[
  {
    "objectID": "TODO.html",
    "href": "TODO.html",
    "title": "todo list",
    "section": "",
    "text": "[] Validation of parameters for combinations that won’t work\n\n\n\n\nFSDP offload and gradient_checkpointing - https://github.com/pytorch/pytorch/issues/82203\nadamw_bnb_8bit doesn’t play well with FSDP offload"
  },
  {
    "objectID": "TODO.html#things-that-are-known-not-to-work",
    "href": "TODO.html#things-that-are-known-not-to-work",
    "title": "todo list",
    "section": "",
    "text": "FSDP offload and gradient_checkpointing - https://github.com/pytorch/pytorch/issues/82203\nadamw_bnb_8bit doesn’t play well with FSDP offload"
  },
  {
    "objectID": "examples/mistral/Mistral-7b-example/code.html",
    "href": "examples/mistral/Mistral-7b-example/code.html",
    "title": "Axolotl",
    "section": "",
    "text": "# Check if GPU is available I used 4x NVIDIA GeForce RTX 3090 (rented 2.1.2-cuda12.1-cudnn8-devel)\nimport torch\nprint('GPU available?', torch.cuda.is_available())\nprint('BF16 is supported?', torch.cuda.is_bf16_supported())\n\nGPU available? True\nBF16 is supported? True\n\n\n\n!pip install huggingface_hub\n\nCollecting huggingface_hub\n  Downloading huggingface_hub-0.20.1-py3-none-any.whl.metadata (12 kB)\nRequirement already satisfied: filelock in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (3.9.0)\nRequirement already satisfied: fsspec>=2023.5.0 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (2023.10.0)\nRequirement already satisfied: requests in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (2.31.0)\nRequirement already satisfied: tqdm>=4.42.1 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (4.65.0)\nRequirement already satisfied: pyyaml>=5.1 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (6.0.1)\nRequirement already satisfied: typing-extensions>=3.7.4.3 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (4.7.1)\nRequirement already satisfied: packaging>=20.9 in /opt/conda/lib/python3.10/site-packages (from huggingface_hub) (23.1)\nRequirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface_hub) (2.0.4)\nRequirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface_hub) (3.4)\nRequirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface_hub) (1.26.18)\nRequirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface_hub) (2023.7.22)\nDownloading huggingface_hub-0.20.1-py3-none-any.whl (330 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 330.1/330.1 kB 8.8 MB/s eta 0:00:00:00:01\nInstalling collected packages: huggingface_hub\nSuccessfully installed huggingface_hub-0.20.1\nWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\n\n\n\nfrom huggingface_hub import notebook_login\n\n\n#Login to huggingface so you can push the model to hub later\nimport sys\nstdout = sys.stdout\nnotebook_login()\n\n\n\n\n\n#I noticed that when you use notebook_login() nothing gets printed after so we use sys \nsys.stdout = stdout\n\n\n#axolotl\n!git clone -b main --depth 1 https://github.com/OpenAccess-AI-Collective/axolotl\n\nCloning into 'axolotl'...\nremote: Enumerating objects: 235, done.\nremote: Counting objects: 100% (235/235), done.\nremote: Compressing objects: 100% (207/207), done.\nremote: Total 235 (delta 48), reused 123 (delta 13), pack-reused 0\nReceiving objects: 100% (235/235), 1.46 MiB | 11.65 MiB/s, done.\nResolving deltas: 100% (48/48), done.\n\n\n\ncd axolotl\n\n/axolotl\n\n\n\n#instaling what is needed inside axolotl file\n!pip install packaging\n!pip install -e '.[flash-attn,deepspeed]'\n!pip install -U git+https://github.com/huggingface/peft.git\n\nRequirement already satisfied: packaging in /opt/conda/lib/python3.10/site-packages (23.1)\nWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\nObtaining file:///axolotl\n  Preparing metadata (setup.py) ... done\nCollecting auto-gptq==0.5.1\n  Downloading auto_gptq-0.5.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (20 kB)\nRequirement already satisfied: packaging in /opt/conda/lib/python3.10/site-packages (23.1)\nCollecting peft==0.6.0\n  Downloading peft-0.6.0-py3-none-any.whl.metadata (23 kB)\nCollecting transformers==4.36.2\n  Downloading transformers-4.36.2-py3-none-any.whl.metadata (126 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 126.8/126.8 kB 9.1 MB/s eta 0:00:00\nCollecting tokenizers==0.15.0\n  Downloading tokenizers-0.15.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (6.7 kB)\nCollecting bitsandbytes>=0.41.1\n  Downloading bitsandbytes-0.41.3.post2-py3-none-any.whl.metadata (9.8 kB)\nCollecting accelerate==0.24.1\n  Downloading accelerate-0.24.1-py3-none-any.whl.metadata (18 kB)\nCollecting addict\n  Downloading addict-2.4.0-py3-none-any.whl (3.8 kB)\nCollecting fire\n  Downloading fire-0.5.0.tar.gz (88 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 88.3/88.3 kB 28.4 MB/s eta 0:00:00\n  Preparing metadata (setup.py) ... done\nRequirement already satisfied: PyYAML>=6.0 in /opt/conda/lib/python3.10/site-packages (6.0.1)\nCollecting datasets>=2.15.0\n  Downloading datasets-2.16.0-py3-none-any.whl.metadata (20 kB)\nCollecting sentencepiece\n  Downloading sentencepiece-0.1.99-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.3 MB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.3/1.3 MB 47.9 MB/s eta 0:00:00\nCollecting wandb\n  Downloading wandb-0.16.1-py3-none-any.whl.metadata (9.8 kB)\nCollecting einops\n  Downloading einops-0.7.0-py3-none-any.whl.metadata (13 kB)\nCollecting optimum==1.13.2\n  Downloading optimum-1.13.2.tar.gz (300 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 301.0/301.0 kB 72.3 MB/s eta 0:00:00\n  Installing build dependencies ... done\n  Getting requirements to build wheel ... done\n  Preparing metadata (pyproject.toml) ... done\nCollecting hf_transfer\n  Downloading hf_transfer-0.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (1.5 kB)\nCollecting colorama\n  Downloading colorama-0.4.6-py2.py3-none-any.whl (25 kB)\nCollecting numba\n  Downloading numba-0.58.1-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl.metadata (2.7 kB)\nRequirement already satisfied: numpy>=1.24.4 in /opt/conda/lib/python3.10/site-packages (1.26.0)\nCollecting bert-score==0.3.13\n  Downloading bert_score-0.3.13-py3-none-any.whl (61 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 61.1/61.1 kB 20.1 MB/s eta 0:00:00\nCollecting evaluate==0.4.0\n  Downloading evaluate-0.4.0-py3-none-any.whl (81 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 81.4/81.4 kB 26.0 MB/s eta 0:00:00\nCollecting rouge-score==0.1.2\n  Downloading rouge_score-0.1.2.tar.gz (17 kB)\n  Preparing metadata (setup.py) ... done\nCollecting scipy\n  Downloading scipy-1.11.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (60 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 60.4/60.4 kB 17.3 MB/s eta 0:00:00\nCollecting scikit-learn==1.2.2\n  Downloading scikit_learn-1.2.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (9.6 MB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 9.6/9.6 MB 83.2 MB/s eta 0:00:0000:0100:01m\nCollecting pynvml\n  Downloading pynvml-11.5.0-py3-none-any.whl (53 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 53.1/53.1 kB 13.6 MB/s eta 0:00:00\nCollecting art\n  Downloading art-6.1-py3-none-any.whl.metadata (69 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 69.9/69.9 kB 21.5 MB/s eta 0:00:00\nCollecting fschat==0.2.34\n  Downloading fschat-0.2.34-py3-none-any.whl.metadata (20 kB)\nCollecting gradio==3.50.2\n  Downloading gradio-3.50.2-py3-none-any.whl.metadata (17 kB)\nCollecting tensorboard\n  Downloading tensorboard-2.15.1-py3-none-any.whl.metadata (1.7 kB)\nCollecting s3fs\n  Downloading s3fs-2023.12.2-py3-none-any.whl.metadata (1.6 kB)\nCollecting gcsfs\n  Downloading gcsfs-2023.12.2.post1-py2.py3-none-any.whl.metadata (1.6 kB)\nCollecting xformers==0.0.23\n  Downloading xformers-0.0.23-cp310-cp310-manylinux2014_x86_64.whl.metadata (1.0 kB)\nCollecting deepspeed\n  Downloading deepspeed-0.12.6.tar.gz (1.2 MB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.2/1.2 MB 109.7 MB/s eta 0:00:00\n  Preparing metadata (setup.py) ... done\nCollecting flash-attn==2.3.3\n  Downloading flash_attn-2.3.3.tar.gz (2.3 MB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2.3/2.3 MB 111.5 MB/s eta 0:00:00\n  Preparing metadata (setup.py) ... done\nRequirement already satisfied: psutil in /opt/conda/lib/python3.10/site-packages (from accelerate==0.24.1) (5.9.0)\nRequirement already satisfied: torch>=1.10.0 in /opt/conda/lib/python3.10/site-packages (from accelerate==0.24.1) (2.1.1)\nRequirement already satisfied: huggingface-hub in /opt/conda/lib/python3.10/site-packages (from accelerate==0.24.1) (0.20.1)\nCollecting rouge (from auto-gptq==0.5.1)\n  Downloading rouge-1.0.1-py3-none-any.whl (13 kB)\nCollecting gekko (from auto-gptq==0.5.1)\n  Downloading gekko-1.0.6-py3-none-any.whl (12.2 MB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 12.2/12.2 MB 77.9 MB/s eta 0:00:0000:010:01\nCollecting safetensors (from auto-gptq==0.5.1)\n  Downloading safetensors-0.4.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (3.8 kB)\nRequirement already satisfied: tqdm in /opt/conda/lib/python3.10/site-packages (from auto-gptq==0.5.1) (4.65.0)\nCollecting pandas>=1.0.1 (from bert-score==0.3.13)\n  Downloading pandas-2.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (18 kB)\nRequirement already satisfied: requests in /opt/conda/lib/python3.10/site-packages (from bert-score==0.3.13) (2.31.0)\nCollecting matplotlib (from bert-score==0.3.13)\n  Downloading matplotlib-3.8.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (5.8 kB)\nCollecting dill (from evaluate==0.4.0)\n  Downloading dill-0.3.7-py3-none-any.whl.metadata (9.9 kB)\nCollecting xxhash (from evaluate==0.4.0)\n  Downloading xxhash-3.4.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (12 kB)\nCollecting multiprocess (from evaluate==0.4.0)\n  Downloading multiprocess-0.70.15-py310-none-any.whl.metadata (7.2 kB)\nRequirement already satisfied: fsspec>=2021.05.0 in /opt/conda/lib/python3.10/site-packages (from fsspec[http]>=2021.05.0->evaluate==0.4.0) (2023.10.0)\nCollecting responses<0.19 (from evaluate==0.4.0)\n  Downloading responses-0.18.0-py3-none-any.whl (38 kB)\nCollecting ninja (from flash-attn==2.3.3)\n  Downloading ninja-1.11.1.1-py2.py3-none-manylinux1_x86_64.manylinux_2_5_x86_64.whl.metadata (5.3 kB)\nCollecting aiohttp (from fschat==0.2.34)\n  Downloading aiohttp-3.9.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (7.4 kB)\nCollecting fastapi (from fschat==0.2.34)\n  Downloading fastapi-0.108.0-py3-none-any.whl.metadata (24 kB)\nCollecting httpx (from fschat==0.2.34)\n  Downloading httpx-0.26.0-py3-none-any.whl.metadata (7.6 kB)\nCollecting markdown2[all] (from fschat==0.2.34)\n  Downloading markdown2-2.4.12-py2.py3-none-any.whl.metadata (2.0 kB)\nCollecting nh3 (from fschat==0.2.34)\n  Downloading nh3-0.2.15-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (1.7 kB)\nRequirement already satisfied: prompt-toolkit>=3.0.0 in /opt/conda/lib/python3.10/site-packages (from fschat==0.2.34) (3.0.36)\nCollecting pydantic<2,>=1 (from fschat==0.2.34)\n  Downloading pydantic-1.10.13-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (149 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 149.6/149.6 kB 42.3 MB/s eta 0:00:00\nCollecting rich>=10.0.0 (from fschat==0.2.34)\n  Downloading rich-13.7.0-py3-none-any.whl.metadata (18 kB)\nCollecting shortuuid (from fschat==0.2.34)\n  Downloading shortuuid-1.0.11-py3-none-any.whl (10 kB)\nCollecting tiktoken (from fschat==0.2.34)\n  Downloading tiktoken-0.5.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (6.6 kB)\nCollecting uvicorn (from fschat==0.2.34)\n  Downloading uvicorn-0.25.0-py3-none-any.whl.metadata (6.4 kB)\nCollecting aiofiles<24.0,>=22.0 (from gradio==3.50.2)\n  Downloading aiofiles-23.2.1-py3-none-any.whl.metadata (9.7 kB)\nCollecting altair<6.0,>=4.2.0 (from gradio==3.50.2)\n  Downloading altair-5.2.0-py3-none-any.whl.metadata (8.7 kB)\nCollecting ffmpy (from gradio==3.50.2)\n  Downloading ffmpy-0.3.1.tar.gz (5.5 kB)\n  Preparing metadata (setup.py) ... done\nCollecting gradio-client==0.6.1 (from gradio==3.50.2)\n  Downloading gradio_client-0.6.1-py3-none-any.whl.metadata (7.1 kB)\nCollecting importlib-resources<7.0,>=1.3 (from gradio==3.50.2)\n  Downloading importlib_resources-6.1.1-py3-none-any.whl.metadata (4.1 kB)\nRequirement already satisfied: jinja2<4.0 in /opt/conda/lib/python3.10/site-packages (from gradio==3.50.2) (3.1.2)\nRequirement already satisfied: markupsafe~=2.0 in /opt/conda/lib/python3.10/site-packages (from gradio==3.50.2) (2.1.1)\nCollecting orjson~=3.0 (from gradio==3.50.2)\n  Downloading orjson-3.9.10-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (49 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 49.3/49.3 kB 14.0 MB/s eta 0:00:00\nRequirement already satisfied: pillow<11.0,>=8.0 in /opt/conda/lib/python3.10/site-packages (from gradio==3.50.2) (10.0.1)\nCollecting pydub (from gradio==3.50.2)\n  Downloading pydub-0.25.1-py2.py3-none-any.whl (32 kB)\nCollecting python-multipart (from gradio==3.50.2)\n  Downloading python_multipart-0.0.6-py3-none-any.whl (45 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 45.7/45.7 kB 13.5 MB/s eta 0:00:00\nCollecting semantic-version~=2.0 (from gradio==3.50.2)\n  Downloading semantic_version-2.10.0-py2.py3-none-any.whl (15 kB)\nRequirement already satisfied: typing-extensions~=4.0 in /opt/conda/lib/python3.10/site-packages (from gradio==3.50.2) (4.7.1)\nCollecting websockets<12.0,>=10.0 (from gradio==3.50.2)\n  Downloading websockets-11.0.3-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (129 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 129.9/129.9 kB 30.8 MB/s eta 0:00:00\nCollecting coloredlogs (from optimum==1.13.2)\n  Downloading coloredlogs-15.0.1-py2.py3-none-any.whl (46 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 46.0/46.0 kB 11.6 MB/s eta 0:00:00\nRequirement already satisfied: sympy in /opt/conda/lib/python3.10/site-packages (from optimum==1.13.2) (1.11.1)\nCollecting absl-py (from rouge-score==0.1.2)\n  Downloading absl_py-2.0.0-py3-none-any.whl.metadata (2.3 kB)\nCollecting nltk (from rouge-score==0.1.2)\n  Downloading nltk-3.8.1-py3-none-any.whl (1.5 MB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.5/1.5 MB 90.4 MB/s eta 0:00:00\nRequirement already satisfied: six>=1.14.0 in /opt/conda/lib/python3.10/site-packages (from rouge-score==0.1.2) (1.16.0)\nCollecting joblib>=1.1.1 (from scikit-learn==1.2.2)\n  Downloading joblib-1.3.2-py3-none-any.whl.metadata (5.4 kB)\nCollecting threadpoolctl>=2.0.0 (from scikit-learn==1.2.2)\n  Downloading threadpoolctl-3.2.0-py3-none-any.whl.metadata (10.0 kB)\nRequirement already satisfied: filelock in /opt/conda/lib/python3.10/site-packages (from transformers==4.36.2) (3.9.0)\nCollecting regex!=2019.12.17 (from transformers==4.36.2)\n  Downloading regex-2023.12.25-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (40 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 40.9/40.9 kB 12.0 MB/s eta 0:00:00\nRequirement already satisfied: networkx in /opt/conda/lib/python3.10/site-packages (from torch>=1.10.0->accelerate==0.24.1) (3.1)\nCollecting pyarrow>=8.0.0 (from datasets>=2.15.0)\n  Downloading pyarrow-14.0.2-cp310-cp310-manylinux_2_28_x86_64.whl.metadata (3.0 kB)\nCollecting pyarrow-hotfix (from datasets>=2.15.0)\n  Downloading pyarrow_hotfix-0.6-py3-none-any.whl.metadata (3.6 kB)\nCollecting hjson (from deepspeed)\n  Downloading hjson-3.1.0-py3-none-any.whl (54 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 54.0/54.0 kB 19.4 MB/s eta 0:00:00\nCollecting py-cpuinfo (from deepspeed)\n  Downloading py_cpuinfo-9.0.0-py3-none-any.whl (22 kB)\nCollecting termcolor (from fire)\n  Downloading termcolor-2.4.0-py3-none-any.whl.metadata (6.1 kB)\nRequirement already satisfied: decorator>4.1.2 in /opt/conda/lib/python3.10/site-packages (from gcsfs) (5.1.1)\nINFO: pip is looking at multiple versions of gcsfs to determine which version is compatible with other requirements. This could take a while.\nCollecting gcsfs\n  Downloading gcsfs-2023.12.1-py2.py3-none-any.whl.metadata (1.6 kB)\n  Downloading gcsfs-2023.12.0-py2.py3-none-any.whl.metadata (1.6 kB)\n  Downloading gcsfs-2023.10.0-py2.py3-none-any.whl.metadata (1.6 kB)\nCollecting google-auth>=1.2 (from gcsfs)\n  Downloading google_auth-2.25.2-py2.py3-none-any.whl.metadata (4.7 kB)\nCollecting google-auth-oauthlib (from gcsfs)\n  Downloading google_auth_oauthlib-1.2.0-py2.py3-none-any.whl.metadata (2.7 kB)\nCollecting google-cloud-storage (from gcsfs)\n  Downloading google_cloud_storage-2.14.0-py2.py3-none-any.whl.metadata (6.1 kB)\nCollecting llvmlite<0.42,>=0.41.0dev0 (from numba)\n  Downloading llvmlite-0.41.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (4.8 kB)\nCollecting aiobotocore<3.0.0,>=2.5.4 (from s3fs)\n  Downloading aiobotocore-2.9.0-py3-none-any.whl.metadata (20 kB)\nINFO: pip is looking at multiple versions of s3fs to determine which version is compatible with other requirements. This could take a while.\nCollecting s3fs\n  Downloading s3fs-2023.12.1-py3-none-any.whl.metadata (1.6 kB)\n  Downloading s3fs-2023.10.0-py3-none-any.whl.metadata (1.6 kB)\nCollecting aiobotocore~=2.7.0 (from s3fs)\n  Downloading aiobotocore-2.7.0-py3-none-any.whl.metadata (20 kB)\nCollecting grpcio>=1.48.2 (from tensorboard)\n  Downloading grpcio-1.60.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (4.0 kB)\nCollecting markdown>=2.6.8 (from tensorboard)\n  Downloading Markdown-3.5.1-py3-none-any.whl.metadata (7.1 kB)\nCollecting protobuf<4.24,>=3.19.6 (from tensorboard)\n  Downloading protobuf-4.23.4-cp37-abi3-manylinux2014_x86_64.whl.metadata (540 bytes)\nRequirement already satisfied: setuptools>=41.0.0 in /opt/conda/lib/python3.10/site-packages (from tensorboard) (68.0.0)\nCollecting tensorboard-data-server<0.8.0,>=0.7.0 (from tensorboard)\n  Downloading tensorboard_data_server-0.7.2-py3-none-manylinux_2_31_x86_64.whl.metadata (1.1 kB)\nCollecting werkzeug>=1.0.1 (from tensorboard)\n  Downloading werkzeug-3.0.1-py3-none-any.whl.metadata (4.1 kB)\nRequirement already satisfied: Click!=8.0.0,>=7.1 in /opt/conda/lib/python3.10/site-packages (from wandb) (8.1.7)\nCollecting GitPython!=3.1.29,>=1.0.0 (from wandb)\n  Downloading GitPython-3.1.40-py3-none-any.whl.metadata (12 kB)\nCollecting sentry-sdk>=1.0.0 (from wandb)\n  Downloading sentry_sdk-1.39.1-py2.py3-none-any.whl.metadata (9.7 kB)\nCollecting docker-pycreds>=0.4.0 (from wandb)\n  Downloading docker_pycreds-0.4.0-py2.py3-none-any.whl (9.0 kB)\nCollecting setproctitle (from wandb)\n  Downloading setproctitle-1.3.3-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (9.9 kB)\nCollecting appdirs>=1.4.3 (from wandb)\n  Downloading appdirs-1.4.4-py2.py3-none-any.whl (9.6 kB)\nCollecting botocore<1.31.65,>=1.31.16 (from aiobotocore~=2.7.0->s3fs)\n  Downloading botocore-1.31.64-py3-none-any.whl.metadata (6.1 kB)\nCollecting wrapt<2.0.0,>=1.10.10 (from aiobotocore~=2.7.0->s3fs)\n  Downloading wrapt-1.16.0-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (6.6 kB)\nCollecting aioitertools<1.0.0,>=0.5.1 (from aiobotocore~=2.7.0->s3fs)\n  Downloading aioitertools-0.11.0-py3-none-any.whl (23 kB)\nRequirement already satisfied: attrs>=17.3.0 in /opt/conda/lib/python3.10/site-packages (from aiohttp->fschat==0.2.34) (23.1.0)\nCollecting multidict<7.0,>=4.5 (from aiohttp->fschat==0.2.34)\n  Downloading multidict-6.0.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (114 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 114.5/114.5 kB 37.1 MB/s eta 0:00:00\nCollecting yarl<2.0,>=1.0 (from aiohttp->fschat==0.2.34)\n  Downloading yarl-1.9.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (31 kB)\nCollecting frozenlist>=1.1.1 (from aiohttp->fschat==0.2.34)\n  Downloading frozenlist-1.4.1-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (12 kB)\nCollecting aiosignal>=1.1.2 (from aiohttp->fschat==0.2.34)\n  Downloading aiosignal-1.3.1-py3-none-any.whl (7.6 kB)\nCollecting async-timeout<5.0,>=4.0 (from aiohttp->fschat==0.2.34)\n  Downloading async_timeout-4.0.3-py3-none-any.whl.metadata (4.2 kB)\nRequirement already satisfied: jsonschema>=3.0 in /opt/conda/lib/python3.10/site-packages (from altair<6.0,>=4.2.0->gradio==3.50.2) (4.20.0)\nRequirement already satisfied: toolz in /opt/conda/lib/python3.10/site-packages (from altair<6.0,>=4.2.0->gradio==3.50.2) (0.12.0)\nCollecting gitdb<5,>=4.0.1 (from GitPython!=3.1.29,>=1.0.0->wandb)\n  Downloading gitdb-4.0.11-py3-none-any.whl.metadata (1.2 kB)\nCollecting cachetools<6.0,>=2.0.0 (from google-auth>=1.2->gcsfs)\n  Downloading cachetools-5.3.2-py3-none-any.whl.metadata (5.2 kB)\nCollecting pyasn1-modules>=0.2.1 (from google-auth>=1.2->gcsfs)\n  Downloading pyasn1_modules-0.3.0-py2.py3-none-any.whl (181 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 181.3/181.3 kB 59.8 MB/s eta 0:00:00\nCollecting rsa<5,>=3.1.4 (from google-auth>=1.2->gcsfs)\n  Downloading rsa-4.9-py3-none-any.whl (34 kB)\nCollecting requests-oauthlib>=0.7.0 (from google-auth-oauthlib->gcsfs)\n  Downloading requests_oauthlib-1.3.1-py2.py3-none-any.whl (23 kB)\nCollecting contourpy>=1.0.1 (from matplotlib->bert-score==0.3.13)\n  Downloading contourpy-1.2.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (5.8 kB)\nCollecting cycler>=0.10 (from matplotlib->bert-score==0.3.13)\n  Downloading cycler-0.12.1-py3-none-any.whl.metadata (3.8 kB)\nCollecting fonttools>=4.22.0 (from matplotlib->bert-score==0.3.13)\n  Downloading fonttools-4.47.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (157 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 157.2/157.2 kB 41.3 MB/s eta 0:00:00\nCollecting kiwisolver>=1.3.1 (from matplotlib->bert-score==0.3.13)\n  Downloading kiwisolver-1.4.5-cp310-cp310-manylinux_2_12_x86_64.manylinux2010_x86_64.whl.metadata (6.4 kB)\nCollecting pyparsing>=2.3.1 (from matplotlib->bert-score==0.3.13)\n  Downloading pyparsing-3.1.1-py3-none-any.whl.metadata (5.1 kB)\nRequirement already satisfied: python-dateutil>=2.7 in /opt/conda/lib/python3.10/site-packages (from matplotlib->bert-score==0.3.13) (2.8.2)\nRequirement already satisfied: pytz>=2020.1 in /opt/conda/lib/python3.10/site-packages (from pandas>=1.0.1->bert-score==0.3.13) (2023.3.post1)\nCollecting tzdata>=2022.1 (from pandas>=1.0.1->bert-score==0.3.13)\n  Downloading tzdata-2023.3-py2.py3-none-any.whl (341 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 341.8/341.8 kB 72.5 MB/s eta 0:00:00\nRequirement already satisfied: wcwidth in /opt/conda/lib/python3.10/site-packages (from prompt-toolkit>=3.0.0->fschat==0.2.34) (0.2.5)\nRequirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests->bert-score==0.3.13) (2.0.4)\nRequirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests->bert-score==0.3.13) (3.4)\nRequirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests->bert-score==0.3.13) (1.26.18)\nRequirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests->bert-score==0.3.13) (2023.7.22)\nCollecting markdown-it-py>=2.2.0 (from rich>=10.0.0->fschat==0.2.34)\n  Downloading markdown_it_py-3.0.0-py3-none-any.whl.metadata (6.9 kB)\nRequirement already satisfied: pygments<3.0.0,>=2.13.0 in /opt/conda/lib/python3.10/site-packages (from rich>=10.0.0->fschat==0.2.34) (2.15.1)\nCollecting h11>=0.8 (from uvicorn->fschat==0.2.34)\n  Downloading h11-0.14.0-py3-none-any.whl (58 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 58.3/58.3 kB 21.6 MB/s eta 0:00:00\nCollecting humanfriendly>=9.1 (from coloredlogs->optimum==1.13.2)\n  Downloading humanfriendly-10.0-py2.py3-none-any.whl (86 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 86.8/86.8 kB 27.2 MB/s eta 0:00:00\nCollecting starlette<0.33.0,>=0.29.0 (from fastapi->fschat==0.2.34)\n  Downloading starlette-0.32.0.post1-py3-none-any.whl.metadata (5.8 kB)\nCollecting typing-extensions~=4.0 (from gradio==3.50.2)\n  Downloading typing_extensions-4.9.0-py3-none-any.whl.metadata (3.0 kB)\nCollecting google-api-core!=2.0.*,!=2.1.*,!=2.2.*,!=2.3.0,<3.0.0dev,>=1.31.5 (from google-cloud-storage->gcsfs)\n  Downloading google_api_core-2.15.0-py3-none-any.whl.metadata (2.7 kB)\nCollecting google-cloud-core<3.0dev,>=2.3.0 (from google-cloud-storage->gcsfs)\n  Downloading google_cloud_core-2.4.1-py2.py3-none-any.whl.metadata (2.7 kB)\nCollecting google-resumable-media>=2.6.0 (from google-cloud-storage->gcsfs)\n  Downloading google_resumable_media-2.7.0-py2.py3-none-any.whl.metadata (2.2 kB)\nCollecting google-crc32c<2.0dev,>=1.0 (from google-cloud-storage->gcsfs)\n  Downloading google_crc32c-1.5.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (32 kB)\nRequirement already satisfied: anyio in /opt/conda/lib/python3.10/site-packages (from httpx->fschat==0.2.34) (4.2.0)\nCollecting httpcore==1.* (from httpx->fschat==0.2.34)\n  Downloading httpcore-1.0.2-py3-none-any.whl.metadata (20 kB)\nRequirement already satisfied: sniffio in /opt/conda/lib/python3.10/site-packages (from httpx->fschat==0.2.34) (1.3.0)\nCollecting wavedrom (from markdown2[all]->fschat==0.2.34)\n  Downloading wavedrom-2.0.3.post3.tar.gz (137 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 137.7/137.7 kB 47.1 MB/s eta 0:00:00\n  Preparing metadata (setup.py) ... done\nRequirement already satisfied: mpmath>=0.19 in /opt/conda/lib/python3.10/site-packages (from sympy->optimum==1.13.2) (1.3.0)\nCollecting jmespath<2.0.0,>=0.7.1 (from botocore<1.31.65,>=1.31.16->aiobotocore~=2.7.0->s3fs)\n  Downloading jmespath-1.0.1-py3-none-any.whl (20 kB)\nCollecting smmap<6,>=3.0.1 (from gitdb<5,>=4.0.1->GitPython!=3.1.29,>=1.0.0->wandb)\n  Downloading smmap-5.0.1-py3-none-any.whl.metadata (4.3 kB)\nCollecting googleapis-common-protos<2.0.dev0,>=1.56.2 (from google-api-core!=2.0.*,!=2.1.*,!=2.2.*,!=2.3.0,<3.0.0dev,>=1.31.5->google-cloud-storage->gcsfs)\n  Downloading googleapis_common_protos-1.62.0-py2.py3-none-any.whl.metadata (1.5 kB)\nRequirement already satisfied: jsonschema-specifications>=2023.03.6 in /opt/conda/lib/python3.10/site-packages (from jsonschema>=3.0->altair<6.0,>=4.2.0->gradio==3.50.2) (2023.12.1)\nRequirement already satisfied: referencing>=0.28.4 in /opt/conda/lib/python3.10/site-packages (from jsonschema>=3.0->altair<6.0,>=4.2.0->gradio==3.50.2) (0.32.0)\nRequirement already satisfied: rpds-py>=0.7.1 in /opt/conda/lib/python3.10/site-packages (from jsonschema>=3.0->altair<6.0,>=4.2.0->gradio==3.50.2) (0.15.2)\nCollecting mdurl~=0.1 (from markdown-it-py>=2.2.0->rich>=10.0.0->fschat==0.2.34)\n  Downloading mdurl-0.1.2-py3-none-any.whl (10.0 kB)\nCollecting pyasn1<0.6.0,>=0.4.6 (from pyasn1-modules>=0.2.1->google-auth>=1.2->gcsfs)\n  Downloading pyasn1-0.5.1-py2.py3-none-any.whl.metadata (8.6 kB)\nCollecting oauthlib>=3.0.0 (from requests-oauthlib>=0.7.0->google-auth-oauthlib->gcsfs)\n  Downloading oauthlib-3.2.2-py3-none-any.whl (151 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 151.7/151.7 kB 50.3 MB/s eta 0:00:00\nRequirement already satisfied: exceptiongroup>=1.0.2 in /opt/conda/lib/python3.10/site-packages (from anyio->httpx->fschat==0.2.34) (1.0.4)\nCollecting svgwrite (from wavedrom->markdown2[all]->fschat==0.2.34)\n  Downloading svgwrite-1.4.3-py3-none-any.whl (67 kB)\n     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 67.1/67.1 kB 21.8 MB/s eta 0:00:00\nDownloading accelerate-0.24.1-py3-none-any.whl (261 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 261.4/261.4 kB 53.4 MB/s eta 0:00:00\nDownloading auto_gptq-0.5.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (4.8 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 4.8/4.8 MB 89.9 MB/s eta 0:00:00ta 0:00:01\nDownloading fschat-0.2.34-py3-none-any.whl (220 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 220.1/220.1 kB 63.9 MB/s eta 0:00:00\nDownloading gradio-3.50.2-py3-none-any.whl (20.3 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 20.3/20.3 MB 82.8 MB/s eta 0:00:00:00:0100:01\nDownloading peft-0.6.0-py3-none-any.whl (134 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 134.9/134.9 kB 40.0 MB/s eta 0:00:00\nDownloading tokenizers-0.15.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.8 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3.8/3.8 MB 87.1 MB/s eta 0:00:00ta 0:00:01\nDownloading transformers-4.36.2-py3-none-any.whl (8.2 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 8.2/8.2 MB 90.1 MB/s eta 0:00:00ta 0:00:01\nDownloading xformers-0.0.23-cp310-cp310-manylinux2014_x86_64.whl (213.0 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 213.0/213.0 MB 36.6 MB/s eta 0:00:0000:0100:01\nDownloading gradio_client-0.6.1-py3-none-any.whl (299 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 299.2/299.2 kB 64.9 MB/s eta 0:00:00\nDownloading bitsandbytes-0.41.3.post2-py3-none-any.whl (92.6 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 92.6/92.6 MB 56.6 MB/s eta 0:00:00:00:0100:01\nDownloading datasets-2.16.0-py3-none-any.whl (507 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 507.1/507.1 kB 87.3 MB/s eta 0:00:00\nDownloading scipy-1.11.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (36.4 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 36.4/36.4 MB 77.4 MB/s eta 0:00:00:00:0100:01\nDownloading art-6.1-py3-none-any.whl (599 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 599.8/599.8 kB 96.0 MB/s eta 0:00:00\nDownloading einops-0.7.0-py3-none-any.whl (44 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 44.6/44.6 kB 13.3 MB/s eta 0:00:00\nDownloading gcsfs-2023.10.0-py2.py3-none-any.whl (33 kB)\nDownloading hf_transfer-0.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.9 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3.9/3.9 MB 99.7 MB/s eta 0:00:00:00:01\nDownloading numba-0.58.1-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl (3.6 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3.6/3.6 MB 100.1 MB/s eta 0:00:0000:01\nDownloading s3fs-2023.10.0-py3-none-any.whl (28 kB)\nDownloading tensorboard-2.15.1-py3-none-any.whl (5.5 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 5.5/5.5 MB 96.9 MB/s eta 0:00:00ta 0:00:01\nDownloading wandb-0.16.1-py3-none-any.whl (2.1 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2.1/2.1 MB 99.4 MB/s eta 0:00:00\nDownloading absl_py-2.0.0-py3-none-any.whl (130 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 130.2/130.2 kB 36.6 MB/s eta 0:00:00\nDownloading aiobotocore-2.7.0-py3-none-any.whl (73 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 73.5/73.5 kB 25.1 MB/s eta 0:00:00\nDownloading aiofiles-23.2.1-py3-none-any.whl (15 kB)\nDownloading aiohttp-3.9.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.2 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.2/1.2 MB 99.6 MB/s eta 0:00:00\nDownloading altair-5.2.0-py3-none-any.whl (996 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 996.9/996.9 kB 110.4 MB/s eta 0:00:00\nDownloading dill-0.3.7-py3-none-any.whl (115 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 115.3/115.3 kB 34.9 MB/s eta 0:00:00\nDownloading GitPython-3.1.40-py3-none-any.whl (190 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 190.6/190.6 kB 47.0 MB/s eta 0:00:00\nDownloading google_auth-2.25.2-py2.py3-none-any.whl (184 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 184.2/184.2 kB 44.9 MB/s eta 0:00:00\nDownloading google_auth_oauthlib-1.2.0-py2.py3-none-any.whl (24 kB)\nDownloading grpcio-1.60.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (5.4 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 5.4/5.4 MB 102.4 MB/s eta 0:00:00a 0:00:01\nDownloading importlib_resources-6.1.1-py3-none-any.whl (33 kB)\nDownloading joblib-1.3.2-py3-none-any.whl (302 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 302.2/302.2 kB 64.0 MB/s eta 0:00:00\nDownloading llvmlite-0.41.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (43.6 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 43.6/43.6 MB 74.3 MB/s eta 0:00:00:00:0100:01\nDownloading Markdown-3.5.1-py3-none-any.whl (102 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 102.2/102.2 kB 34.4 MB/s eta 0:00:00\nDownloading matplotlib-3.8.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (11.6 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 11.6/11.6 MB 99.9 MB/s eta 0:00:00:00:010:01\nDownloading orjson-3.9.10-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (138 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 138.7/138.7 kB 38.8 MB/s eta 0:00:00\nDownloading pandas-2.1.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (12.3 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 12.3/12.3 MB 96.4 MB/s eta 0:00:00:00:010:01\nDownloading protobuf-4.23.4-cp37-abi3-manylinux2014_x86_64.whl (304 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 304.5/304.5 kB 68.5 MB/s eta 0:00:00\nDownloading pyarrow-14.0.2-cp310-cp310-manylinux_2_28_x86_64.whl (38.0 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 38.0/38.0 MB 78.5 MB/s eta 0:00:00:00:0100:01\nDownloading pydantic-1.10.13-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.1 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3.1/3.1 MB 95.6 MB/s eta 0:00:00\nDownloading regex-2023.12.25-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (773 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 774.0/774.0 kB 116.8 MB/s eta 0:00:00\nDownloading rich-13.7.0-py3-none-any.whl (240 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 240.6/240.6 kB 59.9 MB/s eta 0:00:00\nDownloading safetensors-0.4.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.3 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.3/1.3 MB 102.1 MB/s eta 0:00:00\nDownloading sentry_sdk-1.39.1-py2.py3-none-any.whl (254 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 254.1/254.1 kB 71.9 MB/s eta 0:00:00\nDownloading tensorboard_data_server-0.7.2-py3-none-manylinux_2_31_x86_64.whl (6.6 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 6.6/6.6 MB 104.2 MB/s eta 0:00:00a 0:00:01\nDownloading threadpoolctl-3.2.0-py3-none-any.whl (15 kB)\nDownloading uvicorn-0.25.0-py3-none-any.whl (60 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 60.3/60.3 kB 19.1 MB/s eta 0:00:00\nDownloading werkzeug-3.0.1-py3-none-any.whl (226 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 226.7/226.7 kB 67.5 MB/s eta 0:00:00\nDownloading fastapi-0.108.0-py3-none-any.whl (92 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 92.0/92.0 kB 33.8 MB/s eta 0:00:00\nDownloading typing_extensions-4.9.0-py3-none-any.whl (32 kB)\nDownloading google_cloud_storage-2.14.0-py2.py3-none-any.whl (121 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 121.6/121.6 kB 36.9 MB/s eta 0:00:00\nDownloading httpx-0.26.0-py3-none-any.whl (75 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 75.9/75.9 kB 24.7 MB/s eta 0:00:00\nDownloading httpcore-1.0.2-py3-none-any.whl (76 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 76.9/76.9 kB 28.4 MB/s eta 0:00:00\nDownloading multiprocess-0.70.15-py310-none-any.whl (134 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 134.8/134.8 kB 48.7 MB/s eta 0:00:00\nDownloading nh3-0.2.15-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.7 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.7/1.7 MB 108.1 MB/s eta 0:00:00\nDownloading ninja-1.11.1.1-py2.py3-none-manylinux1_x86_64.manylinux_2_5_x86_64.whl (307 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 307.2/307.2 kB 66.5 MB/s eta 0:00:00\nDownloading pyarrow_hotfix-0.6-py3-none-any.whl (7.9 kB)\nDownloading setproctitle-1.3.3-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (30 kB)\nDownloading termcolor-2.4.0-py3-none-any.whl (7.7 kB)\nDownloading tiktoken-0.5.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (2.0 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2.0/2.0 MB 101.9 MB/s eta 0:00:00\nDownloading xxhash-3.4.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (194 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 194.1/194.1 kB 44.4 MB/s eta 0:00:00\nDownloading async_timeout-4.0.3-py3-none-any.whl (5.7 kB)\nDownloading botocore-1.31.64-py3-none-any.whl (11.3 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 11.3/11.3 MB 98.8 MB/s eta 0:00:00:00:010:01\nDownloading cachetools-5.3.2-py3-none-any.whl (9.3 kB)\nDownloading contourpy-1.2.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (310 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 310.7/310.7 kB 69.7 MB/s eta 0:00:00\nDownloading cycler-0.12.1-py3-none-any.whl (8.3 kB)\nDownloading fonttools-4.47.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (4.6 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 4.6/4.6 MB 102.7 MB/s eta 0:00:00a 0:00:01\nDownloading frozenlist-1.4.1-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (239 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 239.5/239.5 kB 71.4 MB/s eta 0:00:00\nDownloading gitdb-4.0.11-py3-none-any.whl (62 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 62.7/62.7 kB 23.3 MB/s eta 0:00:00\nDownloading google_api_core-2.15.0-py3-none-any.whl (121 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 122.0/122.0 kB 32.4 MB/s eta 0:00:00\nDownloading google_cloud_core-2.4.1-py2.py3-none-any.whl (29 kB)\nDownloading google_resumable_media-2.7.0-py2.py3-none-any.whl (80 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 80.6/80.6 kB 22.5 MB/s eta 0:00:00\nDownloading kiwisolver-1.4.5-cp310-cp310-manylinux_2_12_x86_64.manylinux2010_x86_64.whl (1.6 MB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.6/1.6 MB 102.0 MB/s eta 0:00:00\nDownloading markdown_it_py-3.0.0-py3-none-any.whl (87 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 87.5/87.5 kB 25.0 MB/s eta 0:00:00\nDownloading pyparsing-3.1.1-py3-none-any.whl (103 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 103.1/103.1 kB 32.7 MB/s eta 0:00:00\nDownloading starlette-0.32.0.post1-py3-none-any.whl (70 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 70.0/70.0 kB 19.1 MB/s eta 0:00:00\nDownloading wrapt-1.16.0-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (80 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 80.3/80.3 kB 30.3 MB/s eta 0:00:00\nDownloading yarl-1.9.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (301 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 301.6/301.6 kB 80.7 MB/s eta 0:00:00\nDownloading markdown2-2.4.12-py2.py3-none-any.whl (41 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 41.2/41.2 kB 12.5 MB/s eta 0:00:00\nDownloading googleapis_common_protos-1.62.0-py2.py3-none-any.whl (228 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 228.7/228.7 kB 57.7 MB/s eta 0:00:00\nDownloading pyasn1-0.5.1-py2.py3-none-any.whl (84 kB)\n   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 84.9/84.9 kB 30.5 MB/s eta 0:00:00\nDownloading smmap-5.0.1-py3-none-any.whl (24 kB)\nBuilding wheels for collected packages: flash-attn, optimum, rouge-score, deepspeed, fire, ffmpy, wavedrom\n  Building wheel for flash-attn (setup.py) ... done\n  Created wheel for flash-attn: filename=flash_attn-2.3.3-cp310-cp310-linux_x86_64.whl size=57042553 sha256=b1df92cb5bd7657d38b789dd48e907aa3e0bd2715c817eb85f3c4320bb11fb3f\n  Stored in directory: /root/.cache/pip/wheels/e5/e6/fa/941802ec61d1afd320d27160ab1db98e6dba65381f84b76d4a\n  Building wheel for optimum (pyproject.toml) ... done\n  Created wheel for optimum: filename=optimum-1.13.2-py3-none-any.whl size=395599 sha256=ff3a73120e1b6eeeda28f76e3fc8cd4cd826e5d66c869b7848ba150e7af79c62\n  Stored in directory: /root/.cache/pip/wheels/6e/b7/2c/79405d98f0943373d8546daeae25a3d377f7659ca0cbe48699\n  Building wheel for rouge-score (setup.py) ... done\n  Created wheel for rouge-score: filename=rouge_score-0.1.2-py3-none-any.whl size=24932 sha256=8118ecbbcd3529085e794c803f0ddb182fc6c6d3e8a494103b49a94abf1bec37\n  Stored in directory: /root/.cache/pip/wheels/5f/dd/89/461065a73be61a532ff8599a28e9beef17985c9e9c31e541b4\n  Building wheel for deepspeed (setup.py) ... done\n  Created wheel for deepspeed: filename=deepspeed-0.12.6-py3-none-any.whl size=1306729 sha256=35c46b6f0275b0d3063522e0af4f3cbd9ec1c310114d8917d87cbe2bf43346e2\n  Stored in directory: /root/.cache/pip/wheels/a3/dc/a2/f585faaed4dec84108916dcc8e8a7c129a216df8202ca32984\n  Building wheel for fire (setup.py) ... done\n  Created wheel for fire: filename=fire-0.5.0-py2.py3-none-any.whl size=116934 sha256=e76d5185f237f34ec69bb8aa657497bef07408978e4f7efdaef48663bb8cd4ef\n  Stored in directory: /root/.cache/pip/wheels/90/d4/f7/9404e5db0116bd4d43e5666eaa3e70ab53723e1e3ea40c9a95\n  Building wheel for ffmpy (setup.py) ... done\n  Created wheel for ffmpy: filename=ffmpy-0.3.1-py3-none-any.whl size=5579 sha256=da3b54dc0ac1a825a1a233315970ac80b8b4c53ebd9cb2a2cfdeab118f453a64\n  Stored in directory: /root/.cache/pip/wheels/01/a6/d1/1c0828c304a4283b2c1639a09ad86f83d7c487ef34c6b4a1bf\n  Building wheel for wavedrom (setup.py) ... done\n  Created wheel for wavedrom: filename=wavedrom-2.0.3.post3-py2.py3-none-any.whl size=30052 sha256=7f0cbd15d63ee9c120190bac122ab51bbbfc91ee374bc3c046fadb320816c17e\n  Stored in directory: /root/.cache/pip/wheels/9c/52/8c/38b454b42f712f325e26f633287484c7dc1ad469e1580c5954\nSuccessfully built flash-attn optimum rouge-score deepspeed fire ffmpy wavedrom\nInstalling collected packages: sentencepiece, pydub, py-cpuinfo, ninja, nh3, hjson, ffmpy, bitsandbytes, appdirs, addict, xxhash, wrapt, werkzeug, websockets, tzdata, typing-extensions, threadpoolctl, termcolor, tensorboard-data-server, svgwrite, smmap, shortuuid, setproctitle, sentry-sdk, semantic-version, scipy, safetensors, rouge, regex, python-multipart, pyparsing, pynvml, pyasn1, pyarrow-hotfix, pyarrow, protobuf, orjson, oauthlib, multidict, mdurl, markdown2, markdown, llvmlite, kiwisolver, joblib, jmespath, importlib-resources, humanfriendly, hf_transfer, h11, grpcio, google-crc32c, gekko, frozenlist, fonttools, einops, docker-pycreds, dill, cycler, contourpy, colorama, cachetools, async-timeout, art, aioitertools, aiofiles, absl-py, yarl, wavedrom, uvicorn, tiktoken, scikit-learn, rsa, responses, requests-oauthlib, pydantic, pyasn1-modules, pandas, numba, nltk, multiprocess, matplotlib, markdown-it-py, httpcore, googleapis-common-protos, google-resumable-media, gitdb, fire, coloredlogs, botocore, aiosignal, xformers, tokenizers, starlette, rouge-score, rich, httpx, google-auth, GitPython, flash-attn, deepspeed, aiohttp, accelerate, wandb, transformers, gradio-client, google-auth-oauthlib, google-api-core, fastapi, altair, aiobotocore, tensorboard, s3fs, peft, gradio, google-cloud-core, fschat, datasets, bert-score, optimum, google-cloud-storage, evaluate, auto-gptq, gcsfs, axolotl\n  Attempting uninstall: typing-extensions\n    Found existing installation: typing_extensions 4.7.1\n    Uninstalling typing_extensions-4.7.1:\n      Successfully uninstalled typing_extensions-4.7.1\n  Running setup.py develop for axolotl\nSuccessfully installed GitPython-3.1.40 absl-py-2.0.0 accelerate-0.24.1 addict-2.4.0 aiobotocore-2.7.0 aiofiles-23.2.1 aiohttp-3.9.1 aioitertools-0.11.0 aiosignal-1.3.1 altair-5.2.0 appdirs-1.4.4 art-6.1 async-timeout-4.0.3 auto-gptq-0.5.1 axolotl-0.3.0 bert-score-0.3.13 bitsandbytes-0.41.3.post2 botocore-1.31.64 cachetools-5.3.2 colorama-0.4.6 coloredlogs-15.0.1 contourpy-1.2.0 cycler-0.12.1 datasets-2.16.0 deepspeed-0.12.6 dill-0.3.7 docker-pycreds-0.4.0 einops-0.7.0 evaluate-0.4.0 fastapi-0.108.0 ffmpy-0.3.1 fire-0.5.0 flash-attn-2.3.3 fonttools-4.47.0 frozenlist-1.4.1 fschat-0.2.34 gcsfs-2023.10.0 gekko-1.0.6 gitdb-4.0.11 google-api-core-2.15.0 google-auth-2.25.2 google-auth-oauthlib-1.2.0 google-cloud-core-2.4.1 google-cloud-storage-2.14.0 google-crc32c-1.5.0 google-resumable-media-2.7.0 googleapis-common-protos-1.62.0 gradio-3.50.2 gradio-client-0.6.1 grpcio-1.60.0 h11-0.14.0 hf_transfer-0.1.4 hjson-3.1.0 httpcore-1.0.2 httpx-0.26.0 humanfriendly-10.0 importlib-resources-6.1.1 jmespath-1.0.1 joblib-1.3.2 kiwisolver-1.4.5 llvmlite-0.41.1 markdown-3.5.1 markdown-it-py-3.0.0 markdown2-2.4.12 matplotlib-3.8.2 mdurl-0.1.2 multidict-6.0.4 multiprocess-0.70.15 nh3-0.2.15 ninja-1.11.1.1 nltk-3.8.1 numba-0.58.1 oauthlib-3.2.2 optimum-1.13.2 orjson-3.9.10 pandas-2.1.4 peft-0.6.0 protobuf-4.23.4 py-cpuinfo-9.0.0 pyarrow-14.0.2 pyarrow-hotfix-0.6 pyasn1-0.5.1 pyasn1-modules-0.3.0 pydantic-1.10.13 pydub-0.25.1 pynvml-11.5.0 pyparsing-3.1.1 python-multipart-0.0.6 regex-2023.12.25 requests-oauthlib-1.3.1 responses-0.18.0 rich-13.7.0 rouge-1.0.1 rouge-score-0.1.2 rsa-4.9 s3fs-2023.10.0 safetensors-0.4.1 scikit-learn-1.2.2 scipy-1.11.4 semantic-version-2.10.0 sentencepiece-0.1.99 sentry-sdk-1.39.1 setproctitle-1.3.3 shortuuid-1.0.11 smmap-5.0.1 starlette-0.32.0.post1 svgwrite-1.4.3 tensorboard-2.15.1 tensorboard-data-server-0.7.2 termcolor-2.4.0 threadpoolctl-3.2.0 tiktoken-0.5.2 tokenizers-0.15.0 transformers-4.36.2 typing-extensions-4.8.0 tzdata-2023.3 uvicorn-0.25.0 wandb-0.16.1 wavedrom-2.0.3.post3 websockets-11.0.3 werkzeug-3.0.1 wrapt-1.16.0 xformers-0.0.23 xxhash-3.4.1 yarl-1.9.4\nWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\nCollecting git+https://github.com/huggingface/peft.git\n  Cloning https://github.com/huggingface/peft.git to /tmp/pip-req-build-hka8xgk2\n  Running command git clone --filter=blob:none --quiet https://github.com/huggingface/peft.git /tmp/pip-req-build-hka8xgk2\n  Resolved https://github.com/huggingface/peft.git to commit cf04d0353f0343cbf66627228c4495f51669af34\n  Installing build dependencies ... done\n  Getting requirements to build wheel ... done\n  Preparing metadata (pyproject.toml) ... done\nRequirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (1.26.0)\nRequirement already satisfied: packaging>=20.0 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (23.1)\nRequirement already satisfied: psutil in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (5.9.0)\nRequirement already satisfied: pyyaml in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (6.0.1)\nRequirement already satisfied: torch>=1.13.0 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (2.1.1)\nRequirement already satisfied: transformers in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (4.36.2)\nRequirement already satisfied: tqdm in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (4.65.0)\nRequirement already satisfied: accelerate>=0.21.0 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (0.24.1)\nRequirement already satisfied: safetensors in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (0.4.1)\nRequirement already satisfied: huggingface-hub>=0.17.0 in /opt/conda/lib/python3.10/site-packages (from peft==0.7.2.dev0) (0.20.1)\nRequirement already satisfied: filelock in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft==0.7.2.dev0) (3.9.0)\nRequirement already satisfied: fsspec>=2023.5.0 in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft==0.7.2.dev0) (2023.10.0)\nRequirement already satisfied: requests in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft==0.7.2.dev0) (2.31.0)\nRequirement already satisfied: typing-extensions>=3.7.4.3 in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft==0.7.2.dev0) (4.8.0)\nRequirement already satisfied: sympy in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft==0.7.2.dev0) (1.11.1)\nRequirement already satisfied: networkx in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft==0.7.2.dev0) (3.1)\nRequirement already satisfied: jinja2 in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft==0.7.2.dev0) (3.1.2)\nRequirement already satisfied: regex!=2019.12.17 in /opt/conda/lib/python3.10/site-packages (from transformers->peft==0.7.2.dev0) (2023.12.25)\nRequirement already satisfied: tokenizers<0.19,>=0.14 in /opt/conda/lib/python3.10/site-packages (from transformers->peft==0.7.2.dev0) (0.15.0)\nRequirement already satisfied: MarkupSafe>=2.0 in /opt/conda/lib/python3.10/site-packages (from jinja2->torch>=1.13.0->peft==0.7.2.dev0) (2.1.1)\nRequirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft==0.7.2.dev0) (2.0.4)\nRequirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft==0.7.2.dev0) (3.4)\nRequirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft==0.7.2.dev0) (1.26.18)\nRequirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft==0.7.2.dev0) (2023.7.22)\nRequirement already satisfied: mpmath>=0.19 in /opt/conda/lib/python3.10/site-packages (from sympy->torch>=1.13.0->peft==0.7.2.dev0) (1.3.0)\nBuilding wheels for collected packages: peft\n  Building wheel for peft (pyproject.toml) ... done\n  Created wheel for peft: filename=peft-0.7.2.dev0-py3-none-any.whl size=169456 sha256=4c70d23e759fa6abb3827fb2f3a8683be3b24d78777d0f403bbc2c0548e5dd4b\n  Stored in directory: /tmp/pip-ephem-wheel-cache-my5ncou6/wheels/d7/c7/de/1368fac8590e1b103ddc2ec2a28ad51d83aded1a3830e8a087\nSuccessfully built peft\nInstalling collected packages: peft\n  Attempting uninstall: peft\n    Found existing installation: peft 0.6.0\n    Uninstalling peft-0.6.0:\n      Successfully uninstalled peft-0.6.0\nERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.\naxolotl 0.3.0 requires peft==0.6.0, but you have peft 0.7.2.dev0 which is incompatible.\nSuccessfully installed peft-0.7.2.dev0\nWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\n\n\n\n\"\"\"\nTraining using the config.yml file and using deepspeed:zero3_bf16 the most aggressive optimization out of zero1,zero2,zero3 stages which partitions \nnot only optimizer states but also gradients and parameters across GPUs. The bf16 indicate mixed precision training using bfloat16.\nFor more information read axolotl's readme\n\"\"\"\n!accelerate launch -m axolotl.cli.train /folder/config.yml --deepspeed deepspeed_configs/zero3_bf16.json\n\nThe following values were not passed to `accelerate launch` and had defaults used instead:\n    `--num_processes` was set to a value of `3`\n        More than one GPU was found, enabling multi-GPU training.\n        If this was unintended please pass in `--num_processes=1`.\n    `--num_machines` was set to a value of `1`\n    `--mixed_precision` was set to a value of `'no'`\n    `--dynamo_backend` was set to a value of `'no'`\nTo avoid this warning pass in values for each of the problematic parameters or run `accelerate config`.\n/opt/conda/lib/python3.10/site-packages/transformers/deepspeed.py:23: FutureWarning: transformers.deepspeed module is deprecated and will be removed in a future version. Please import deepspeed modules directly from transformers.integrations\n  warnings.warn(\n[2023-12-28 15:44:09,979] [INFO] [datasets.<module>:58] [PID:2814] PyTorch version 2.1.1 available.\n/opt/conda/lib/python3.10/site-packages/transformers/deepspeed.py:23: FutureWarning: transformers.deepspeed module is deprecated and will be removed in a future version. Please import deepspeed modules directly from transformers.integrations\n  warnings.warn(\n/opt/conda/lib/python3.10/site-packages/transformers/deepspeed.py:23: FutureWarning: transformers.deepspeed module is deprecated and will be removed in a future version. Please import deepspeed modules directly from transformers.integrations\n  warnings.warn(\n[2023-12-28 15:44:10,011] [INFO] [datasets.<module>:58] [PID:2812] PyTorch version 2.1.1 available.\n[2023-12-28 15:44:10,013] [INFO] [datasets.<module>:58] [PID:2813] PyTorch version 2.1.1 available.\n[2023-12-28 15:44:10,805] [INFO] [axolotl.normalize_config:150] [PID:2814] [RANK:2] GPU memory usage baseline: 0.000GB (+0.317GB misc)\n[2023-12-28 15:44:10,830] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n[2023-12-28 15:44:10,842] [INFO] [axolotl.normalize_config:150] [PID:2813] [RANK:1] GPU memory usage baseline: 0.000GB (+0.317GB misc)\n[2023-12-28 15:44:10,865] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n[2023-12-28 15:44:10,869] [INFO] [axolotl.normalize_config:150] [PID:2812] [RANK:0] GPU memory usage baseline: 0.000GB (+0.351GB misc)\n[2023-12-28 15:44:10,887] [INFO] [real_accelerator.py:161:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n[2023-12-28 15:44:10,961] [INFO] [comm.py:637:init_distributed] cdb=None\n[2023-12-28 15:44:10,994] [INFO] [comm.py:637:init_distributed] cdb=None\n[2023-12-28 15:44:11,015] [INFO] [comm.py:637:init_distributed] cdb=None\n[2023-12-28 15:44:11,015] [INFO] [comm.py:668:init_distributed] Initializing TorchBackend in DeepSpeed with backend nccl\n                                 dP            dP   dP \n                                 88            88   88 \n      .d8888b. dP.  .dP .d8888b. 88 .d8888b. d8888P 88 \n      88'  `88  `8bd8'  88'  `88 88 88'  `88   88   88 \n      88.  .88  .d88b.  88.  .88 88 88.  .88   88   88 \n      `88888P8 dP'  `dP `88888P' dP `88888P'   dP   dP \n                                                       \n                                                       \n\n[2023-12-28 15:44:11,412] [DEBUG] [axolotl.load_tokenizer:184] [PID:2812] [RANK:0] EOS: 2 / </s>\n[2023-12-28 15:44:11,412] [DEBUG] [axolotl.load_tokenizer:185] [PID:2812] [RANK:0] BOS: 1 / <s>\n[2023-12-28 15:44:11,412] [DEBUG] [axolotl.load_tokenizer:186] [PID:2812] [RANK:0] PAD: 2 / </s>\n[2023-12-28 15:44:11,412] [DEBUG] [axolotl.load_tokenizer:187] [PID:2812] [RANK:0] UNK: 0 / <unk>\n[2023-12-28 15:44:11,413] [INFO] [axolotl.load_tokenized_prepared_datasets:143] [PID:2812] [RANK:0] Loading prepared dataset from disk at tilemachos/GF_new.json/1adc45d2edc1e98ce657814412c6593c...\n[2023-12-28 15:44:11,415] [INFO] [axolotl.load_tokenized_prepared_datasets:145] [PID:2812] [RANK:0] Prepared dataset loaded from disk...\n[2023-12-28 15:44:11,432] [DEBUG] [axolotl.load_tokenizer:184] [PID:2814] [RANK:2] EOS: 2 / </s>\n[2023-12-28 15:44:11,432] [DEBUG] [axolotl.load_tokenizer:185] [PID:2814] [RANK:2] BOS: 1 / <s>\n[2023-12-28 15:44:11,432] [DEBUG] [axolotl.load_tokenizer:186] [PID:2814] [RANK:2] PAD: 2 / </s>\n[2023-12-28 15:44:11,432] [DEBUG] [axolotl.load_tokenizer:187] [PID:2814] [RANK:2] UNK: 0 / <unk>\n[2023-12-28 15:44:11,530] [DEBUG] [axolotl.load_tokenizer:184] [PID:2813] [RANK:1] EOS: 2 / </s>\n[2023-12-28 15:44:11,531] [DEBUG] [axolotl.load_tokenizer:185] [PID:2813] [RANK:1] BOS: 1 / <s>\n[2023-12-28 15:44:11,531] [DEBUG] [axolotl.load_tokenizer:186] [PID:2813] [RANK:1] PAD: 2 / </s>\n[2023-12-28 15:44:11,531] [DEBUG] [axolotl.load_tokenizer:187] [PID:2813] [RANK:1] UNK: 0 / <unk>\n[2023-12-28 15:44:12,158] [INFO] [axolotl.load_tokenized_prepared_datasets:143] [PID:2813] [RANK:1] Loading prepared dataset from disk at tilemachos/GF_new.json/1adc45d2edc1e98ce657814412c6593c...\n[2023-12-28 15:44:12,158] [INFO] [axolotl.load_tokenized_prepared_datasets:143] [PID:2814] [RANK:2] Loading prepared dataset from disk at tilemachos/GF_new.json/1adc45d2edc1e98ce657814412c6593c...\n[2023-12-28 15:44:12,160] [INFO] [axolotl.load_tokenized_prepared_datasets:145] [PID:2813] [RANK:1] Prepared dataset loaded from disk...\n[2023-12-28 15:44:12,161] [INFO] [axolotl.load_tokenized_prepared_datasets:145] [PID:2814] [RANK:2] Prepared dataset loaded from disk...\n[2023-12-28 15:44:12,236] [DEBUG] [axolotl.log:60] [PID:2812] [RANK:0] total_num_tokens: 28120\n[2023-12-28 15:44:12,238] [DEBUG] [axolotl.log:60] [PID:2812] [RANK:0] `total_supervised_tokens: 7990`\n[2023-12-28 15:44:12,238] [DEBUG] [axolotl.log:60] [PID:2812] [RANK:0] total_num_steps: 6\n[2023-12-28 15:44:12,242] [DEBUG] [axolotl.train.log:60] [PID:2812] [RANK:0] loading tokenizer... mistralai/Mistral-7B-v0.1\n[2023-12-28 15:44:12,518] [DEBUG] [axolotl.load_tokenizer:184] [PID:2812] [RANK:0] EOS: 2 / </s>\n[2023-12-28 15:44:12,518] [DEBUG] [axolotl.load_tokenizer:185] [PID:2812] [RANK:0] BOS: 1 / <s>\n[2023-12-28 15:44:12,518] [DEBUG] [axolotl.load_tokenizer:186] [PID:2812] [RANK:0] PAD: 2 / </s>\n[2023-12-28 15:44:12,518] [DEBUG] [axolotl.load_tokenizer:187] [PID:2812] [RANK:0] UNK: 0 / <unk>\n[2023-12-28 15:44:12,518] [DEBUG] [axolotl.train.log:60] [PID:2812] [RANK:0] loading model and peft_config...\n[2023-12-28 15:44:12,589] [DEBUG] [axolotl.load_tokenizer:184] [PID:2814] [RANK:2] EOS: 2 / </s>\n[2023-12-28 15:44:12,589] [DEBUG] [axolotl.load_tokenizer:185] [PID:2814] [RANK:2] BOS: 1 / <s>\n[2023-12-28 15:44:12,589] [DEBUG] [axolotl.load_tokenizer:186] [PID:2814] [RANK:2] PAD: 2 / </s>\n[2023-12-28 15:44:12,589] [DEBUG] [axolotl.load_tokenizer:187] [PID:2814] [RANK:2] UNK: 0 / <unk>\n[2023-12-28 15:44:12,599] [DEBUG] [axolotl.load_tokenizer:184] [PID:2813] [RANK:1] EOS: 2 / </s>\n[2023-12-28 15:44:12,599] [DEBUG] [axolotl.load_tokenizer:185] [PID:2813] [RANK:1] BOS: 1 / <s>\n[2023-12-28 15:44:12,599] [DEBUG] [axolotl.load_tokenizer:186] [PID:2813] [RANK:1] PAD: 2 / </s>\n[2023-12-28 15:44:12,599] [DEBUG] [axolotl.load_tokenizer:187] [PID:2813] [RANK:1] UNK: 0 / <unk>\n[2023-12-28 15:44:13,049] [INFO] [partition_parameters.py:348:__exit__] finished initializing model - num_params = 291, num_elems = 7.24B\nLoading checkpoint shards: 100%|██████████████████| 2/2 [00:11<00:00,  5.81s/it]\nLoading checkpoint shards: 100%|██████████████████| 2/2 [00:11<00:00,  5.98s/it]\n[2023-12-28 15:44:25,395] [INFO] [axolotl.load_model:503] [PID:2813] [RANK:1] GPU memory usage after model load: 7.576GB (+0.524GB cache, +0.708GB misc)\n[2023-12-28 15:44:25,399] [INFO] [axolotl.load_model:526] [PID:2813] [RANK:1] converting PEFT model w/ prepare_model_for_kbit_training\n[2023-12-28 15:44:25,403] [INFO] [axolotl.load_model:538] [PID:2813] [RANK:1] converting modules to torch.bfloat16 for flash attention\ntrainable params: 3,407,872 || all params: 7,245,139,968 || trainable%: 0.04703666202518836\n[2023-12-28 15:44:25,480] [INFO] [axolotl.load_model:568] [PID:2813] [RANK:1] GPU memory usage after adapters: 7.589GB (+1.501GB cache, +0.708GB misc)\n[2023-12-28 15:44:25,572] [INFO] [axolotl.load_model:503] [PID:2814] [RANK:2] GPU memory usage after model load: 7.576GB (+0.410GB cache, +0.708GB misc)\n[2023-12-28 15:44:25,576] [INFO] [axolotl.load_model:526] [PID:2814] [RANK:2] converting PEFT model w/ prepare_model_for_kbit_training\n[2023-12-28 15:44:25,580] [INFO] [axolotl.load_model:538] [PID:2814] [RANK:2] converting modules to torch.bfloat16 for flash attention\ntrainable params: 3,407,872 || all params: 7,245,139,968 || trainable%: 0.04703666202518836\n[2023-12-28 15:44:25,660] [INFO] [axolotl.load_model:568] [PID:2814] [RANK:2] GPU memory usage after adapters: 7.589GB (+1.388GB cache, +0.708GB misc)\nLoading checkpoint shards: 100%|██████████████████| 2/2 [00:12<00:00,  6.30s/it]\n[2023-12-28 15:44:26,170] [INFO] [axolotl.load_model:503] [PID:2812] [RANK:0] GPU memory usage after model load: 7.576GB (+0.776GB cache, +0.741GB misc)\n[2023-12-28 15:44:26,177] [INFO] [axolotl.load_model:526] [PID:2812] [RANK:0] converting PEFT model w/ prepare_model_for_kbit_training\n[2023-12-28 15:44:26,181] [INFO] [axolotl.load_model:538] [PID:2812] [RANK:0] converting modules to torch.bfloat16 for flash attention\ntrainable params: 3,407,872 || all params: 7,245,139,968 || trainable%: 0.04703666202518836\n[2023-12-28 15:44:26,259] [INFO] [axolotl.load_model:568] [PID:2812] [RANK:0] GPU memory usage after adapters: 7.589GB (+1.753GB cache, +0.741GB misc)\n[2023-12-28 15:44:26,293] [INFO] [axolotl.train.log:60] [PID:2812] [RANK:0] Pre-saving adapter config to ./out\n[2023-12-28 15:44:26,296] [INFO] [axolotl.train.log:60] [PID:2812] [RANK:0] Starting trainer...\nUsing /root/.cache/torch_extensions/py310_cu121 as PyTorch extensions root...\nUsing /root/.cache/torch_extensions/py310_cu121 as PyTorch extensions root...\nUsing /root/.cache/torch_extensions/py310_cu121 as PyTorch extensions root...\nDetected CUDA files, patching ldflags\nEmitting ninja build file /root/.cache/torch_extensions/py310_cu121/fused_adam/build.ninja...\nBuilding extension module fused_adam...\nAllowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)\nninja: no work to do.\nLoading extension module fused_adam...\nTime to load fused_adam op: 0.05891108512878418 seconds\nLoading extension module fused_adam...\nTime to load fused_adam op: 0.10173463821411133 seconds\nLoading extension module fused_adam...\nTime to load fused_adam op: 0.10152459144592285 seconds\n/opt/conda/lib/python3.10/site-packages/deepspeed/ops/adam/fused_adam.py:96: UserWarning: The torch.cuda.*DtypeTensor constructors are no longer recommended. It's best to use methods such as torch.tensor(data, dtype=*, device='cuda') to create tensors. (Triggered internally at /opt/conda/conda-bld/pytorch_1699449201336/work/torch/csrc/tensor/python_tensor.cpp:83.)\n  self._dummy_overflow_buf = get_accelerator().IntTensor([0])\n/opt/conda/lib/python3.10/site-packages/deepspeed/ops/adam/fused_adam.py:96: UserWarning: The torch.cuda.*DtypeTensor constructors are no longer recommended. It's best to use methods such as torch.tensor(data, dtype=*, device='cuda') to create tensors. (Triggered internally at /opt/conda/conda-bld/pytorch_1699449201336/work/torch/csrc/tensor/python_tensor.cpp:83.)\n  self._dummy_overflow_buf = get_accelerator().IntTensor([0])\n/opt/conda/lib/python3.10/site-packages/deepspeed/ops/adam/fused_adam.py:96: UserWarning: The torch.cuda.*DtypeTensor constructors are no longer recommended. It's best to use methods such as torch.tensor(data, dtype=*, device='cuda') to create tensors. (Triggered internally at /opt/conda/conda-bld/pytorch_1699449201336/work/torch/csrc/tensor/python_tensor.cpp:83.)\n  self._dummy_overflow_buf = get_accelerator().IntTensor([0])\nParameter Offload: Total persistent parameters: 3674112 in 193 params\n  0%|                                                    | 0/17 [00:00<?, ?it/s]/opt/conda/lib/python3.10/site-packages/torch/utils/checkpoint.py:429: UserWarning: torch.utils.checkpoint: please pass in use_reentrant=True or use_reentrant=False explicitly. The default value of use_reentrant will be updated to be False in the future. To maintain current behavior, pass use_reentrant=True. It is recommended that you use use_reentrant=False. Refer to docs for more details on the differences between the two variants.\n  warnings.warn(\n/opt/conda/lib/python3.10/site-packages/torch/utils/checkpoint.py:429: UserWarning: torch.utils.checkpoint: please pass in use_reentrant=True or use_reentrant=False explicitly. The default value of use_reentrant will be updated to be False in the future. To maintain current behavior, pass use_reentrant=True. It is recommended that you use use_reentrant=False. Refer to docs for more details on the differences between the two variants.\n  warnings.warn(\n/opt/conda/lib/python3.10/site-packages/torch/utils/checkpoint.py:429: UserWarning: torch.utils.checkpoint: please pass in use_reentrant=True or use_reentrant=False explicitly. The default value of use_reentrant will be updated to be False in the future. To maintain current behavior, pass use_reentrant=True. It is recommended that you use use_reentrant=False. Refer to docs for more details on the differences between the two variants.\n  warnings.warn(\n/opt/conda/lib/python3.10/site-packages/bitsandbytes/autograd/_functions.py:322: UserWarning: MatMul8bitLt: inputs will be cast from torch.bfloat16 to float16 during quantization\n  warnings.warn(f\"MatMul8bitLt: inputs will be cast from {A.dtype} to float16 during quantization\")\n/opt/conda/lib/python3.10/site-packages/bitsandbytes/autograd/_functions.py:322: UserWarning: MatMul8bitLt: inputs will be cast from torch.bfloat16 to float16 during quantization\n  warnings.warn(f\"MatMul8bitLt: inputs will be cast from {A.dtype} to float16 during quantization\")\n/opt/conda/lib/python3.10/site-packages/bitsandbytes/autograd/_functions.py:322: UserWarning: MatMul8bitLt: inputs will be cast from torch.bfloat16 to float16 during quantization\n  warnings.warn(f\"MatMul8bitLt: inputs will be cast from {A.dtype} to float16 during quantization\")\n{'loss': 2.0448, 'learning_rate': 2e-05, 'epoch': 0.06}                         \n  6%|██▌                                         | 1/17 [00:28<07:32, 28.30s/it]\n  0%|                                                     | 0/3 [00:00<?, ?it/s]\n 67%|██████████████████████████████               | 2/3 [00:03<00:01,  1.85s/it]\n                                                                                \n{'eval_loss': 1.9694719314575195, 'eval_runtime': 11.391, 'eval_samples_per_second': 1.492, 'eval_steps_per_second': 0.263, 'epoch': 0.06}\n  6%|██▌                                         | 1/17 [00:39<07:32, 28.30s/it]\n100%|█████████████████████████████████████████████| 3/3 [00:07<00:00,  2.65s/it]\n                                                                                [2023-12-28 15:45:35,358] [INFO] [axolotl.callbacks.on_step_end:122] [PID:2812] [RANK:0] GPU memory usage while training: 12.210GB (+4.259GB cache, +0.776GB misc)\n 12%|█████▏                                      | 2/17 [01:04<08:18, 33.20s/it][2023-12-28 15:45:35,358] [INFO] [axolotl.callbacks.on_step_end:122] [PID:2814] [RANK:2] GPU memory usage while training: 12.269GB (+4.522GB cache, +0.743GB misc)\n[2023-12-28 15:45:35,358] [INFO] [axolotl.callbacks.on_step_end:122] [PID:2813] [RANK:1] GPU memory usage while training: 12.283GB (+4.493GB cache, +0.743GB misc)\n{'loss': 2.0022, 'learning_rate': 4e-05, 'epoch': 0.12}                         \n{'loss': 2.1054, 'learning_rate': 6e-05, 'epoch': 0.17}                         \n{'loss': 1.9004, 'learning_rate': 8e-05, 'epoch': 0.23}                         \n{'loss': 1.8794, 'learning_rate': 0.0001, 'epoch': 0.29}                        \n 29%|████████████▉                               | 5/17 [02:20<05:23, 26.92s/it]\n  0%|                                                     | 0/3 [00:00<?, ?it/s]\n 67%|██████████████████████████████               | 2/3 [00:03<00:01,  1.88s/it]\n                                                                                \n{'eval_loss': 1.7912336587905884, 'eval_runtime': 11.3106, 'eval_samples_per_second': 1.503, 'eval_steps_per_second': 0.265, 'epoch': 0.29}\n 29%|████████████▉                               | 5/17 [02:32<05:23, 26.92s/it]\n100%|█████████████████████████████████████████████| 3/3 [00:07<00:00,  2.67s/it]\n{'loss': 1.7871, 'learning_rate': 0.00012, 'epoch': 0.35}                       \n{'loss': 1.7758, 'learning_rate': 0.00014, 'epoch': 0.4}                        \n{'loss': 1.4645, 'learning_rate': 0.00016, 'epoch': 0.46}                       \n{'loss': 1.4009, 'learning_rate': 0.00018, 'epoch': 0.52}                       \n{'loss': 1.3927, 'learning_rate': 0.0002, 'epoch': 0.58}                        \n 59%|█████████████████████████▎                 | 10/17 [04:38<03:04, 26.33s/it]\n  0%|                                                     | 0/3 [00:00<?, ?it/s]\n 67%|██████████████████████████████               | 2/3 [00:03<00:01,  1.89s/it]\n                                                                                \n{'eval_loss': 1.1426481008529663, 'eval_runtime': 11.3344, 'eval_samples_per_second': 1.5, 'eval_steps_per_second': 0.265, 'epoch': 0.58}\n 59%|█████████████████████████▎                 | 10/17 [04:49<03:04, 26.33s/it]\n100%|█████████████████████████████████████████████| 3/3 [00:07<00:00,  2.68s/it]\n{'loss': 1.0122, 'learning_rate': 0.0001900968867902419, 'epoch': 0.63}         \n{'loss': 1.0019, 'learning_rate': 0.00016234898018587337, 'epoch': 0.69}        \n{'loss': 0.8976, 'learning_rate': 0.00012225209339563145, 'epoch': 0.75}        \n{'loss': 0.9301, 'learning_rate': 7.774790660436858e-05, 'epoch': 0.81}         \n{'loss': 0.8595, 'learning_rate': 3.7651019814126654e-05, 'epoch': 0.87}        \n 88%|█████████████████████████████████████▉     | 15/17 [06:55<00:52, 26.17s/it]\n  0%|                                                     | 0/3 [00:00<?, ?it/s]\n 67%|██████████████████████████████               | 2/3 [00:03<00:01,  1.88s/it]\n                                                                                \n{'eval_loss': 0.8175248503684998, 'eval_runtime': 11.2932, 'eval_samples_per_second': 1.505, 'eval_steps_per_second': 0.266, 'epoch': 0.87}\n 88%|█████████████████████████████████████▉     | 15/17 [07:06<00:52, 26.17s/it]\n100%|█████████████████████████████████████████████| 3/3 [00:07<00:00,  2.67s/it]\n{'loss': 0.7931, 'learning_rate': 9.903113209758096e-06, 'epoch': 0.92}         \n{'loss': 0.6909, 'learning_rate': 0.0, 'epoch': 0.98}                           \n100%|███████████████████████████████████████████| 17/17 [07:56<00:00, 28.03s/it]/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1879: UserWarning: Positional args are being deprecated, use kwargs instead. Refer to https://pytorch.org/docs/master/generated/torch.nn.Module.html#torch.nn.Module.state_dict for details.\n  warnings.warn(\n/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1879: UserWarning: Positional args are being deprecated, use kwargs instead. Refer to https://pytorch.org/docs/master/generated/torch.nn.Module.html#torch.nn.Module.state_dict for details.\n  warnings.warn(\n/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1879: UserWarning: Positional args are being deprecated, use kwargs instead. Refer to https://pytorch.org/docs/master/generated/torch.nn.Module.html#torch.nn.Module.state_dict for details.\n  warnings.warn(\n{'train_runtime': 489.0649, 'train_samples_per_second': 0.63, 'train_steps_per_second': 0.035, 'train_loss': 1.408153467318591, 'epoch': 0.98}\n100%|███████████████████████████████████████████| 17/17 [08:09<00:00, 28.77s/it]\n[2023-12-28 15:52:39,488] [INFO] [axolotl.train.log:60] [PID:2812] [RANK:0] Training Completed!!! Saving pre-trained model to ./out"
  },
  {
    "objectID": "docs/multi-node.html",
    "href": "docs/multi-node.html",
    "title": "Multi Node",
    "section": "",
    "text": "You will need to create a configuration for accelerate, either by using accelerate config and follow the instructions or you can use one of the preset below:\n~/.cache/huggingface/accelerate/default_config.yaml\nConfigure your model to use FSDP with for example:",
    "crumbs": [
      "How-To Guides",
      "Multi Node"
    ]
  },
  {
    "objectID": "docs/multi-node.html#machine-configuration",
    "href": "docs/multi-node.html#machine-configuration",
    "title": "Multi Node",
    "section": "Machine configuration",
    "text": "Machine configuration\nOn each machine you need a copy of Axolotl, we suggest using the same commit to ensure compatibility.\nYou will also need to have the same configuration file for your model on each machine.\nOn the main machine only, make sure the port you set as main_process_port is open in TCP and reachable by other machines.\nAll you have to do now is launch using accelerate as you would usually do on each machine and voila, the processes will start once you have launched accelerate on every machine.",
    "crumbs": [
      "How-To Guides",
      "Multi Node"
    ]
  },
  {
    "objectID": "docs/config.html",
    "href": "docs/config.html",
    "title": "Config options",
    "section": "",
    "text": "# This is the huggingface model that contains *.pt, *.safetensors, or *.bin files\n# This can also be a relative path to a model on disk\nbase_model: ./llama-7b-hf\n# You can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)\nbase_model_ignore_patterns:\n# If the base_model repo on hf hub doesn't include configuration .json files,\n# You can set that here, or leave this empty to default to base_model\nbase_model_config: ./llama-7b-hf\n# You can specify to choose a specific model revision from huggingface hub\nrevision_of_model:\n# Optional tokenizer configuration path in case you want to use a different tokenizer\n# than the one defined in the base model\ntokenizer_config:\n# If you want to specify the type of model to load, AutoModelForCausalLM is a good choice too\nmodel_type: AutoModelForCausalLM\n# Corresponding tokenizer for the model AutoTokenizer is a good choice\ntokenizer_type: AutoTokenizer\n# Trust remote code for untrusted source\ntrust_remote_code:\n# use_fast option for tokenizer loading from_pretrained, default to True\ntokenizer_use_fast:\n# Whether to use the legacy tokenizer setting, defaults to True\ntokenizer_legacy:\n# Resize the model embeddings when new tokens are added to multiples of 32\n# This is reported to improve training speed on some models\nresize_token_embeddings_to_32x:\n\n# (Internal use only)\n# Used to identify which the model is based on\nis_falcon_derived_model:\nis_llama_derived_model:\nis_qwen_derived_model:\n# Please note that if you set this to true, `padding_side` will be set to \"left\" by default\nis_mistral_derived_model:\n\n# optional overrides to the base model configuration\noverrides_of_model_config:\n  # RoPE Scaling https://github.com/huggingface/transformers/pull/24653\n  rope_scaling:\n    type: # linear | dynamic\n    factor: # float\n\n# optional overrides to the bnb 4bit quantization configuration\n# https://huggingface.co/docs/transformers/main/main_classes/quantization#transformers.BitsAndBytesConfig\nbnb_config_kwargs:\n  # These are default values\n  llm_int8_has_fp16_weight: false\n  bnb_4bit_quant_type: nf4\n  bnb_4bit_use_double_quant: true\n\n\n# Whether you are training a 4-bit GPTQ quantized model\ngptq: true\n\n# This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer\nload_in_8bit: true\n# Use bitsandbytes 4 bit\nload_in_4bit:\n\n# Use CUDA bf16\nbf16: true # bool or 'full' for `bf16_full_eval`. require >=ampere\n# Use CUDA fp16\nfp16: true\n# Use CUDA tf32\ntf32: true # require >=ampere\n\n# No AMP (automatic mixed precision)\nbfloat16: true # require >=ampere\nfloat16: true\n\n# Limit the memory for all available GPUs to this amount (if an integer, expressed in gigabytes); default: unset\ngpu_memory_limit: 20GiB\n# Do the LoRA/PEFT loading on CPU -- this is required if the base model is so large it takes up most or all of the available GPU VRAM, e.g. during a model and LoRA merge\nlora_on_cpu: true\n\n# A list of one or more datasets to finetune the model with\ndatasets:\n  # HuggingFace dataset repo | s3://,gs:// path | \"json\" for local dataset, make sure to fill data_files\n  - path: vicgalle/alpaca-gpt4\n  # The type of prompt to use for training. [alpaca, sharegpt, gpteacher, oasst, reflection]\n    type: alpaca # format | format:<prompt_style> (chat/instruct) | <prompt_strategies>.load_<load_fn>\n    ds_type: # Optional[str] (json|arrow|parquet|text|csv) defines the datatype when path is a file\n    data_files: # Optional[str] path to source data files\n    shards: # Optional[int] number of shards to split data into\n    name: # Optional[str] name of dataset configuration to load\n    train_on_split: train # Optional[str] name of dataset split to load from\n\n    # Optional[str] fastchat conversation type, only used with type: sharegpt\n    conversation: # Options (see Conversation 'name'): https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py\n    field_human: # Optional[str]. Human key to use for conversation.\n    field_model: # Optional[str]. Assistant key to use for conversation.\n    # Add additional keys from your dataset as input or output roles\n    roles:\n      input: # Optional[List[str]]. These will be masked based on train_on_input\n      output: # Optional[List[str]].\n\n  # Custom user instruction prompt\n  - path: repo\n    type:\n      # The below are defaults. only set what's needed if you use a different column name.\n      system_prompt: \"\"\n      system_format: \"{system}\"\n      field_system: system\n      field_instruction: instruction\n      field_input: input\n      field_output: output\n\n      # Customizable to be single line or multi-line\n      # Use {instruction}/{input} as key to be replaced\n      # 'format' can include {input}\n      format: |-\n        User: {instruction} {input}\n        Assistant:\n      # 'no_input_format' cannot include {input}\n      no_input_format: \"{instruction} \"\n\n      # For `completion` datsets only, uses the provided field instead of `text` column\n      field:\n\n# If false, the datasets will not be shuffled and will keep their original order in `datasets`.\n# The same applies to the `test_datasets` option and the `pretraining_dataset` option. Default is true.\nshuffle_merged_datasets: true\n\n# A list of one or more datasets to eval the model with.\n# You can use either test_datasets, or val_set_size, but not both.\ntest_datasets:\n  - path: /workspace/data/eval.jsonl\n    ds_type: json\n    # You need to specify a split. For \"json\" datasets the default split is called \"train\".\n    split: train\n    type: completion\n    data_files:\n      - /workspace/data/eval.jsonl\n\n# use RL training: 'dpo', 'ipo', 'kto_pair'\nrl:\n\n# Saves the desired chat template to the tokenizer_config.json for easier inferencing\n# Currently supports chatml and inst (mistral/mixtral)\nchat_template: chatml\n# Changes the default system message\ndefault_system_message: You are a helpful assistant. Please give a long and detailed answer. # Currently only supports chatml.\n# Axolotl attempts to save the dataset as an arrow after packing the data together so\n# subsequent training attempts load faster, relative path\ndataset_prepared_path: data/last_run_prepared\n# Push prepared dataset to hub\npush_dataset_to_hub: # repo path\n# The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`\n# if not set.\ndataset_processes: # defaults to os.cpu_count() if not set\n# Keep dataset in memory while preprocessing\n# Only needed if cached dataset is taking too much storage\ndataset_keep_in_memory:\n# push checkpoints to hub\nhub_model_id: # private repo path to push finetuned model\n# how to push checkpoints to hub\n# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy\nhub_strategy:\n# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets\n# Required to be true when used in combination with `push_dataset_to_hub`\nhf_use_auth_token: # boolean\n# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval.\nval_set_size: 0.04\n# Num shards for whole dataset\ndataset_shard_num:\n# Index of shard to use for whole dataset\ndataset_shard_idx:\n\n# The maximum length of an input to train with, this should typically be less than 2048\n# as most models have a token/context limit of 2048\nsequence_len: 2048\n# Pad inputs so each step uses constant sized buffers\n# This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently\npad_to_sequence_len:\n# Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'\nsample_packing:\n# Set to 'false' if getting errors during eval with sample_packing on.\neval_sample_packing:\n# You can set these packing optimizations AFTER starting a training at least once.\n# The trainer will provide recommended values for these values.\nsample_packing_eff_est:\ntotal_num_tokens:\n\n# Passed through to transformers when loading the model when launched without accelerate\n# Use `sequential` when training w/ model parallelism to limit memory\ndevice_map:\n# Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model.\nmax_memory:\n\n# If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model\nadapter: lora\n# If you already have a lora model trained that you want to load, put that here.\n# This means after training, if you want to test the model, you should set this to the value of `output_dir`.\n# Note that if you merge an adapter to the base model, a new subdirectory `merged` will be created under the `output_dir`.\nlora_model_dir:\n\n# LoRA hyperparameters\n# For more details about the following options, see:\n# https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2\nlora_r: 8\nlora_alpha: 16\nlora_dropout: 0.05\nlora_target_modules:\n  - q_proj\n  - v_proj\n#  - k_proj\n#  - o_proj\n#  - gate_proj\n#  - down_proj\n#  - up_proj\nlora_target_linear: # If true, will target all linear modules\npeft_layers_to_transform: # The layer indices to transform, otherwise, apply to all layers\n\n# If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.\n# For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.\n# `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.\n# https://github.com/huggingface/peft/issues/334#issuecomment-1561727994\nlora_modules_to_save:\n#  - embed_tokens\n#  - lm_head\n\nlora_fan_in_fan_out: false\n\npeft:\n  # Configuration options for loftq initialization for LoRA\n  # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization\n  loftq_config:\n    loftq_bits:  # typically 4 bits\n\n# ReLoRA configuration\n# Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed\nrelora_steps: # Number of steps per ReLoRA restart\nrelora_warmup_steps: # Number of per-restart warmup steps\nrelora_anneal_steps: # Number of anneal steps for each relora cycle\nrelora_prune_ratio: # threshold for optimizer magnitude when pruning\nrelora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings\n\n# wandb configuration if you're using it\n# Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.\nwandb_mode: # \"offline\" to save run metadata locally and not sync to the server, \"disabled\" to turn off wandb\nwandb_project: # Your wandb project name\nwandb_entity: # A wandb Team name if using a Team\nwandb_watch:\nwandb_name: # Set the name of your wandb run\nwandb_run_id: # Set the ID of your wandb run\nwandb_log_model: # \"checkpoint\" to log model to wandb Artifacts every `save_steps` or \"end\" to log only at the end of training\n\n# mlflow configuration if you're using it\nmlflow_tracking_uri: # URI to mlflow\nmlflow_experiment_name: # Your experiment name\nhf_mlflow_log_artifacts:  # set to true to copy each saved checkpoint on each save to mlflow artifact registry\n\n# Where to save the full-finetuned model to\noutput_dir: ./completed-model\n\n# Whether to use torch.compile and which backend to use\ntorch_compile:  # bool\ntorch_compile_backend:  # Optional[str]\n\n# Training hyperparameters\n\n# If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.\ngradient_accumulation_steps: 1\n# The number of samples to include in each batch. This is the number of samples sent to each GPU.\nmicro_batch_size: 2\neval_batch_size:\nnum_epochs: 4\nwarmup_steps: 100  # cannot use with warmup_ratio\nwarmup_ratio: 0.05  # cannot use with warmup_steps\nlearning_rate: 0.00003\nlr_quadratic_warmup:\nlogging_steps:\neval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps\nevals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps\nsave_strategy: # Set to `no` to skip checkpoint saves\nsave_steps: # Leave empty to save at each epoch\nsaves_per_epoch: # number of times per epoch to save a checkpoint, mutually exclusive with save_steps\nsave_total_limit: # Checkpoints saved at a time\n# Maximum number of iterations to train for. It precedes num_epochs which means that\n# if both are set, num_epochs will not be guaranteed.\n# e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps\nmax_steps:\n\neval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0\neval_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128\neval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is [\"sacrebleu\", \"comet\", \"ter\", chrf]\n\nloss_watchdog_threshold: # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training)\nloss_watchdog_patience: # Number of high-loss steps in a row before the trainer aborts (default: 3)\n\n# Save model as safetensors (require safetensors package)\nsave_safetensors:\n\n# Whether to mask out or include the human's prompt from the training labels\ntrain_on_inputs: false\n# Group similarly sized data to minimize padding.\n# May be slower to start, as it must download and sort the entire dataset.\n# Note that training loss may have an oscillating pattern with this enabled.\ngroup_by_length: false\n\n# Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing\ngradient_checkpointing: false\n# additional kwargs to pass to the trainer for gradient checkpointing\n# gradient_checkpointing_kwargs:\n#   use_reentrant: true\n\n# Stop training after this many evaluation losses have increased in a row\n# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback\nearly_stopping_patience: 3\n\n# Specify a scheduler and kwargs to use with the optimizer\nlr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine\nlr_scheduler_kwargs:\ncosine_min_lr_ratio: # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr\ncosine_constant_lr_ratio: # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step (https://arxiv.org/pdf/2308.04014.pdf)\n\n# For one_cycle optim\nlr_div_factor: # Learning rate div factor\n\n# Specify optimizer\n# Valid values are driven by the Transformers OptimizerNames class, see:\n# https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134\n#\n# Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of\n# torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used\n# in the examples/ for your model and fine-tuning use case.\n#\n# Valid values for 'optimizer' include:\n# - adamw_hf\n# - adamw_torch\n# - adamw_torch_fused\n# - adamw_torch_xla\n# - adamw_apex_fused\n# - adafactor\n# - adamw_anyprecision\n# - sgd\n# - adagrad\n# - adamw_bnb_8bit\n# - lion_8bit\n# - lion_32bit\n# - paged_adamw_32bit\n# - paged_adamw_8bit\n# - paged_lion_32bit\n# - paged_lion_8bit\n# - galore_adamw\n# - galore_adamw_8bit\n# - galore_adafactor\n# - galore_adamw_layerwise\n# - galore_adamw_8bit_layerwise\n# - galore_adafactor_layerwise\noptimizer:\n# Dictionary of arguments to pass to the optimizer\noptim_args:\n# For Galore Optimizers the following optim_args are available\n# rank:  # type: int\n# update_proj_gap  # type: int\n# scale  # type: float\n# proj_type:  # type: str, default = std\n\n# The target modules to optimize, i.e. the module names that you would like to train, right now this is used only for GaLore algorithm\noptim_target_modules:\n# - self_attn  # for llama\n# - mlp\n\n# Specify weight decay\nweight_decay:\n# adamw hyperparams\nadam_beta1:\nadam_beta2:\nadam_epsilon:\n# Gradient clipping max norm\nmax_grad_norm:\n\n# Augmentation techniques\n# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings\n# currently only supported on Llama and Mistral\nneftune_noise_alpha:\n\n# Whether to bettertransformers\nflash_optimum:\n# Whether to use xformers attention patch https://github.com/facebookresearch/xformers:\nxformers_attention:\n# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:\nflash_attention:\nflash_attn_cross_entropy:  # Whether to use flash-attention cross entropy implementation - advanced use only\nflash_attn_rms_norm:  # Whether to use flash-attention rms norm implementation - advanced use only\nflash_attn_fuse_qkv: # Whether to fuse QKV into a single operation\nflash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation\n# Whether to use scaled-dot-product attention\n# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html\nsdp_attention:\n# Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf\ns2_attention:\n# Resume from a specific checkpoint dir\nresume_from_checkpoint:\n# If resume_from_checkpoint isn't set and you simply want it to start where it left off.\n# Be careful with this being turned on between different models.\nauto_resume_from_checkpoints: false\n\n# Don't mess with this, it's here for accelerate and torchrun\nlocal_rank:\n\n# Add or change special tokens.\n# If you add tokens here, you don't need to add them to the `tokens` list.\nspecial_tokens:\n  # bos_token: \"<s>\"\n  # eos_token: \"</s>\"\n  # unk_token: \"<unk>\"\n\n# Add extra tokens.\ntokens:\n\n# FSDP\nfsdp:\nfsdp_config:\n\n# Deepspeed config path. e.g., deepspeed_configs/zero3.json\ndeepspeed:\n\n# Advanced DDP Arguments\nddp_timeout:\nddp_bucket_cap_mb:\nddp_broadcast_buffers:\n\n# Path to torch distx for optim 'adamw_anyprecision'\ntorchdistx_path:\n\n# Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize\npretraining_dataset:\n\n# Debug mode\ndebug:\n\n# Seed\nseed:\n\n# Allow overwrite yml config using from cli\nstrict:",
    "crumbs": [
      "Reference",
      "Config options"
    ]
  },
  {
    "objectID": "docs/multipack.html",
    "href": "docs/multipack.html",
    "title": "Multipack (Sample Packing)",
    "section": "",
    "text": "Because Flash Attention simply drops the attention mask, we do not need to construct a 4d attention mask. We only need to concatenate the sequences into a single batch and let flash attention know where each new sequence begins.\n4k context, bsz =4, each character represents 256 tokens X represents a padding token\n   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5\n[[ A A A A A A A A A A A ]\n   B B B B B B ]\n   C C C C C C C ]\n   D D D D ]]\n\n[[ E E E E E E E E ]\n [ F F F F ]\n [ G G G ]\n [ H H H H ]]\n\n[[ I I I ]\n [ J J J ]\n [ K K K K K]\n [ L L L ]]\nafter padding to longest input in each step\n   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5\n[[ A A A A A A A A A A A ]\n   B B B B B B X X X X X X ]\n   C C C C C C C X X X X ]\n   D D D D X X X X X X X ]]\n\n[[ E E E E E E E E ]\n [ F F F F X X X X ]\n [ G G G X X X X X ]\n [ H H H H X X X X ]]\n\n[[ I I I X X ]\n [ J J J X X ]\n [ K K K K K ]\n [ L L L X X ]]\nw packing ( note it’s the same effective number of tokens per step, but a true bsz of 1)\n   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5\n[[ A A A A A A A A A A A B B B B B\n   B C C C C C C C D D D D E E E E\n   E E E E F F F F F G G G H H H H\n   I I I J J J J K K K K K L L L X ]]\ncu_seqlens: [[ 0, 11, 17, 24, 28, 36, 41 44, 48, 51, 55, 60, 64]]",
    "crumbs": [
      "How-To Guides",
      "Multipack (Sample Packing)"
    ]
  },
  {
    "objectID": "docs/multipack.html#visualization-of-multipack-with-flash-attention",
    "href": "docs/multipack.html#visualization-of-multipack-with-flash-attention",
    "title": "Multipack (Sample Packing)",
    "section": "",
    "text": "Because Flash Attention simply drops the attention mask, we do not need to construct a 4d attention mask. We only need to concatenate the sequences into a single batch and let flash attention know where each new sequence begins.\n4k context, bsz =4, each character represents 256 tokens X represents a padding token\n   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5\n[[ A A A A A A A A A A A ]\n   B B B B B B ]\n   C C C C C C C ]\n   D D D D ]]\n\n[[ E E E E E E E E ]\n [ F F F F ]\n [ G G G ]\n [ H H H H ]]\n\n[[ I I I ]\n [ J J J ]\n [ K K K K K]\n [ L L L ]]\nafter padding to longest input in each step\n   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5\n[[ A A A A A A A A A A A ]\n   B B B B B B X X X X X X ]\n   C C C C C C C X X X X ]\n   D D D D X X X X X X X ]]\n\n[[ E E E E E E E E ]\n [ F F F F X X X X ]\n [ G G G X X X X X ]\n [ H H H H X X X X ]]\n\n[[ I I I X X ]\n [ J J J X X ]\n [ K K K K K ]\n [ L L L X X ]]\nw packing ( note it’s the same effective number of tokens per step, but a true bsz of 1)\n   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5\n[[ A A A A A A A A A A A B B B B B\n   B C C C C C C C D D D D E E E E\n   E E E E F F F F F G G G H H H H\n   I I I J J J J K K K K K L L L X ]]\ncu_seqlens: [[ 0, 11, 17, 24, 28, 36, 41 44, 48, 51, 55, 60, 64]]",
    "crumbs": [
      "How-To Guides",
      "Multipack (Sample Packing)"
    ]
  },
  {
    "objectID": "docs/multipack.html#multipack-without-flash-attention",
    "href": "docs/multipack.html#multipack-without-flash-attention",
    "title": "Multipack (Sample Packing)",
    "section": "Multipack without Flash Attention",
    "text": "Multipack without Flash Attention\nMultipack can still be achieved without Flash attention, but with lower packing efficiency as we are not able to join multiple batches into a single batch due to context length limits without flash attention. We can use either Pytorch’s Scaled Dot Product Attention implementation or native Pytorch attention implementation along with 4d attention masks to pack sequences together and avoid cross attention.",
    "crumbs": [
      "How-To Guides",
      "Multipack (Sample Packing)"
    ]
  },
  {
    "objectID": "docs/mac.html",
    "href": "docs/mac.html",
    "title": "Mac M-series",
    "section": "",
    "text": "Currently Axolotl on Mac is partially usable, many of the dependencies of Axolotl including Pytorch do not support MPS or have incomplete support.\nCurrent support:\n\nSupport for all models\nFull training of models\nLoRA training\nSample packing\nFP16 and BF16 (awaiting AMP support for MPS in Pytorch)\nTri-dao’s flash-attn (until it is supported use spd_attention as an alternative)\nxformers\nbitsandbytes (meaning no 4/8 bits loading and bnb optimizers)\nqlora\nDeepSpeed\n\nUntested: - FSDP",
    "crumbs": [
      "How-To Guides",
      "Mac M-series"
    ]
  },
  {
    "objectID": "docs/fsdp_qlora.html",
    "href": "docs/fsdp_qlora.html",
    "title": "FDSP + QLoRA",
    "section": "",
    "text": "Using FSDP with QLoRA is essential for fine-tuning larger (70b+ parameter) LLMs on consumer GPUs. For example, you can use FSDP + QLoRA to train a 70b model on two 24GB GPUs1.\nBelow, we describe how to use this feature in Axolotl."
  },
  {
    "objectID": "docs/fsdp_qlora.html#background",
    "href": "docs/fsdp_qlora.html#background",
    "title": "FDSP + QLoRA",
    "section": "",
    "text": "Using FSDP with QLoRA is essential for fine-tuning larger (70b+ parameter) LLMs on consumer GPUs. For example, you can use FSDP + QLoRA to train a 70b model on two 24GB GPUs1.\nBelow, we describe how to use this feature in Axolotl."
  },
  {
    "objectID": "docs/fsdp_qlora.html#usage",
    "href": "docs/fsdp_qlora.html#usage",
    "title": "FDSP + QLoRA",
    "section": "Usage",
    "text": "Usage\nTo enable QLoRA with FSDP, you need to perform the following steps:\n\n![Tip] See the example config file in addition to reading these instructions.\n\n\nSet adapter: qlora in your axolotl config file.\nEnable FSDP in your axolotl config, as described here.\nUse one of the supported model types: llama, mistral or mixtral."
  },
  {
    "objectID": "docs/fsdp_qlora.html#example-config",
    "href": "docs/fsdp_qlora.html#example-config",
    "title": "FDSP + QLoRA",
    "section": "Example Config",
    "text": "Example Config\nexamples/llama-2/qlora-fsdp.yml contains an example of how to enable QLoRA + FSDP in axolotl."
  },
  {
    "objectID": "docs/fsdp_qlora.html#references",
    "href": "docs/fsdp_qlora.html#references",
    "title": "FDSP + QLoRA",
    "section": "References",
    "text": "References\n\nPR #1378 enabling QLoRA in FSDP in Axolotl.\nBlog Post from the Answer.AI team describing the work that enabled QLoRA in FSDP.\nRelated HuggingFace PRs Enabling FDSP + QLoRA:\n\nAccelerate PR#2544\nTransformers PR#29587\nTRL PR#1416\nPEFT PR#1550"
  },
  {
    "objectID": "docs/fsdp_qlora.html#footnotes",
    "href": "docs/fsdp_qlora.html#footnotes",
    "title": "FDSP + QLoRA",
    "section": "Footnotes",
    "text": "Footnotes\n\n\nThis was enabled by this work from the Answer.AI team.↩︎"
  },
  {
    "objectID": "FAQS.html",
    "href": "FAQS.html",
    "title": "FAQs",
    "section": "",
    "text": "FAQs\n\nCan you train StableLM with this? Yes, but only with a single GPU atm. Multi GPU support is coming soon! Just waiting on this PR\nWill this work with Deepspeed? That’s still a WIP, but setting export ACCELERATE_USE_DEEPSPEED=true should work in some cases\nError invalid argument at line 359 in file /workspace/bitsandbytes/csrc/pythonInterface.c /arrow/cpp/src/arrow/filesystem/s3fs.cc:2598:  arrow::fs::FinalizeS3 was not called even though S3 was initialized. This could lead to a segmentation fault at exit. Try reinstalling bitsandbytes and transformers from source."
  },
  {
    "objectID": "docs/faq.html",
    "href": "docs/faq.html",
    "title": "FAQ",
    "section": "",
    "text": "Q: The trainer stopped and hasn’t progressed in several minutes.\n\nA: Usually an issue with the GPUs communicating with each other. See the NCCL doc\n\nQ: Exitcode -9\n\nA: This usually happens when you run out of system RAM.\n\nQ: Exitcode -7 while using deepspeed\n\nA: Try upgrading deepspeed w: pip install -U deepspeed\n\nQ: AttributeError: ‘DummyOptim’ object has no attribute ‘step’\n\nA: You may be using deepspeed with single gpu. Please don’t set deepspeed: in yaml or cli.",
    "crumbs": [
      "FAQ"
    ]
  },
  {
    "objectID": "docs/rlhf.html",
    "href": "docs/rlhf.html",
    "title": "RLHF (Beta)",
    "section": "",
    "text": "Overview\nReinforcement Learning from Human Feedback is a method whereby a language model is optimized from data using human feedback. Various methods include, but not limited to:\n\nProximal Policy Optimization (PPO) (not yet supported in axolotl)\nDirect Preference Optimization (DPO)\nIdentity Preference Optimization (IPO)\n\n\n\nRLHF using Axolotl\n\n[!IMPORTANT] This is a BETA feature and many features are not fully implemented. You are encouraged to open new PRs to improve the integration and functionality.\n\nThe various RL training methods are implemented in trl and wrapped via axolotl. Below are various examples with how you can use various preference datasets to train models that use ChatML\n\nDPO\nrl: dpo\ndatasets:\n  - path: Intel/orca_dpo_pairs\n    split: train\n    type: chatml.intel\n  - path: argilla/ultrafeedback-binarized-preferences\n    split: train\n    type: chatml.argilla\n\n\nIPO\nrl: ipo\n\n\nORPO\nPaper: https://arxiv.org/abs/2403.07691\nrl: orpo\norpo_alpha: 0.1\nremove_unused_columns: false\n\nchat_template: chatml\ndatasets:\n  - path: argilla/ultrafeedback-binarized-preferences-cleaned\n    type: orpo.chat_template\n\n\nUsing local dataset files\ndatasets:\n  - ds_type: json\n    data_files:\n      - orca_rlhf.jsonl\n    split: train\n    type: chatml.intel\n\n\nTrl autounwrap for peft\nTrl supports autounwrapping peft models, so that a ref model does not need to be additionally loaded, leading to less VRAM needed. This is on by default. To turn it off, pass the following config.\n# load ref model when adapter training.\nrl_adapter_ref_model: true",
    "crumbs": [
      "How-To Guides",
      "RLHF (Beta)"
    ]
  },
  {
    "objectID": "docs/nccl.html",
    "href": "docs/nccl.html",
    "title": "NCCL",
    "section": "",
    "text": "NVIDIA NCCL is a library to facilitate and optimize multi-GPU communication operations, such as broadcast, all-gather, reduce, all-reduce, etc. Broadly, NCCL configuration is highly environment-specific and is configured via several environment variables. A common NCCL-related problem occurs when a long-running operation times out causing the training process to abort:\nWatchdog caught collective operation timeout: WorkNCCL(SeqNum=42, OpType=ALLGATHER, Timeout(ms)=1800000) ran for 1806948 milliseconds before timing out.\nOften, this timeout will happen after 30 minutes (the default setting) and is accompanied by below-average power consumption with near 100% GPU utilization before the error is raised. Nvidia recommends disabling PCI access control services (ACS) as a possible solution if this is available to you.\nForcing cross-GPU communication via NVLink may help without increasing timeouts. To verify that your configuration is leveraging NVLink run the following command:\nnvidia-smi nvlink --status\nTo force NCCL to use NVLink, simply set this in the environment:\nexport NCCL_P2P_LEVEL=NVL\nIf NVLink is not available in your environment there are other options for NCCL_P2P_LEVEL in the table below:\n\n\n\n\n\n\n\nNCCL_P2P_LEVEL\nDescription\n\n\n\n\nPIX\nP2P data transfers through no more than a single PCIe bridge. Faster data transfer rates vs to paths involving multiple bridges, but slower compared to direct GPU-to-GPU communication.\n\n\nPXB\nP2P data transfers through multiple PCIe bridges but not going through the PCIe Host Bridge; this path involves a complex routing process, potentially incurring a moderate level of latency.\n\n\nPHB\nP2P data transfers occur over the PCIe and through a PCIe Host Bridge, typically involving the CPU, which can facilitate direct memory access but might introduce additional latency compared to more direct paths (ex PIX, NVL)\n\n\n\nTo validate that acceptable data transfer speeds exist for your training job, running NCCL Tests can help pinpoint bottlenecks, for example:\n./build/all_reduce_perf -b 8 -e 128M -f 2 -g 3\nIt can be useful when debugging NCCL communication timeouts to activate additional logging in both PyTorch and NCCL:\nexport NCCL_DEBUG=INFO\nexport NCCL_DEBUG_SUBSYS=ALL\nexport TORCH_DISTRIBUTED_DEBUG=INFO\nexport TORCHELASTIC_ERROR_FILE=/PATH/TO/torcherror.log\nFinally, if you believe your training job needs more time you can increase the timeout past 30 minutes by setting the ddp_timeout value in the Axolotl configuration. See PyTorch init_process_group for documentation on this value.",
    "crumbs": [
      "How-To Guides",
      "NCCL"
    ]
  },
  {
    "objectID": "docs/debugging.html",
    "href": "docs/debugging.html",
    "title": "Debugging",
    "section": "",
    "text": "This document provides some tips and tricks for debugging Axolotl. It also provides an example configuration for debugging with VSCode. A good debugging setup is essential to understanding how Axolotl code works behind the scenes.",
    "crumbs": [
      "How-To Guides",
      "Debugging"
    ]
  },
  {
    "objectID": "docs/debugging.html#table-of-contents",
    "href": "docs/debugging.html#table-of-contents",
    "title": "Debugging",
    "section": "Table of Contents",
    "text": "Table of Contents\n\nGeneral Tips\nDebugging with VSCode\n\nBackground\nConfiguration\nCustomizing your debugger\nVideo Tutorial\n\nDebugging With Docker\n\nSetup\nAttach To Container\nVideo - Attaching To Docker On Remote Host",
    "crumbs": [
      "How-To Guides",
      "Debugging"
    ]
  },
  {
    "objectID": "docs/debugging.html#general-tips",
    "href": "docs/debugging.html#general-tips",
    "title": "Debugging",
    "section": "General Tips",
    "text": "General Tips\nWhile debugging it’s helpful to simplify your test scenario as much as possible. Here are some tips for doing so:\n\n[!Important] All of these tips are incorporated into the example configuration for debugging with VSCode below.\n\n\nMake sure you are using the latest version of axolotl: This project changes often and bugs get fixed fast. Check your git branch and make sure you have pulled the latest changes from main.\nEliminate concurrency: Restrict the number of processes to 1 for both training and data preprocessing:\n\nSet CUDA_VISIBLE_DEVICES to a single GPU, ex: export CUDA_VISIBLE_DEVICES=0.\nSet dataset_processes: 1 in your axolotl config or run the training command with --dataset_processes=1.\n\nUse a small dataset: Construct or use a small dataset from HF Hub. When using a small dataset, you will often have to make sure sample_packing: False and eval_sample_packing: False to avoid errors. If you are in a pinch and don’t have time to construct a small dataset but want to use from the HF Hub, you can shard the data (this will still tokenize the entire dataset, but will only use a fraction of the data for training. For example, to shard the dataset into 20 pieces, add the following to your axolotl config): yaml     dataset:         ...         shards: 20\nUse a small model: A good example of a small model is TinyLlama/TinyLlama-1.1B-Chat-v1.0.\nMinimize iteration time: Make sure the training loop finishes as fast as possible, with these settings.\n\nmicro_batch_size: 1\nmax_steps: 1\nval_set_size: 0\n\nClear Caches: Axolotl caches certain steps and so does the underlying HuggingFace trainer. You may want to clear some of these caches when debugging.\n\nData preprocessing: When debugging data preprocessing, which includes prompt template formation, you may want to delete the directory set in dataset_prepared_path: in your axolotl config. If you didn’t set this value, the default is last_run_prepared.\nHF Hub: If you are debugging data preprocessing, you should clear the relevant HF cache HuggingFace cache, by deleting the appropriate ~/.cache/huggingface/datasets/... folder(s).\nThe recommended approach is to redirect all outputs and caches to a temporary folder and delete selected subfolders before each run. This is demonstrated in the example configuration below.",
    "crumbs": [
      "How-To Guides",
      "Debugging"
    ]
  },
  {
    "objectID": "docs/debugging.html#debugging-with-vscode",
    "href": "docs/debugging.html#debugging-with-vscode",
    "title": "Debugging",
    "section": "Debugging with VSCode",
    "text": "Debugging with VSCode\n\nBackground\nThe below example shows how to configure VSCode to debug data preprocessing of the sharegpt format. This is the format used when you have the following in your axolotl config:\ndatasets:\n  - path: <path to your sharegpt formatted dataset> # example on HF Hub: philschmid/guanaco-sharegpt-style\n    type: sharegpt\n\n[!Important] If you are already familiar with advanced VSCode debugging, you can skip the below explanation and look at the files .vscode/launch.json and .vscode/tasks.json for an example configuration.\n\n\n[!Tip] If you prefer to watch a video, rather than read, you can skip to the video tutorial below (but doing both is recommended).\n\n\n\nSetup\nMake sure you have an editable install of Axolotl, which ensures that changes you make to the code are reflected at runtime. Run the following commands from the root of this project:\npip3 install packaging\npip3 install -e '.[flash-attn,deepspeed]'\n\nRemote Hosts\nIf you developing on a remote host, you can easily use VSCode to debug remotely. To do so, you will need to follow this remote - SSH guide. You can also see the video below on Docker and Remote SSH debugging.\n\n\n\nConfiguration\nThe easiest way to get started is to modify the .vscode/launch.json file in this project. This is just an example configuration, so you may need to modify or copy it to suit your needs.\nFor example, to mimic the command cd devtools && CUDA_VISIBLE_DEVICES=0 accelerate launch -m axolotl.cli.train dev_sharegpt.yml, you would use the below configuration1. Note that we add additional flags that override the axolotl config and incorporate the tips above (see the comments). We also set the working directory to devtools and set the env variable HF_HOME to a temporary folder that is later partially deleted. This is because we want to delete the HF dataset cache before each run in order to ensure that the data preprocessing code is run from scratch.\n// .vscode/launch.json\n{\n    \"version\": \"0.2.0\",\n    \"configurations\": [\n        {\n            \"name\": \"Debug axolotl prompt - sharegpt\",\n            \"type\": \"python\",\n            \"module\": \"accelerate.commands.launch\",\n            \"request\": \"launch\",\n            \"args\": [\n                \"-m\", \"axolotl.cli.train\", \"dev_sharegpt.yml\",\n                // The flags below simplify debugging by overriding the axolotl config\n                // with the debugging tips above.  Modify as needed.\n                \"--dataset_processes=1\",      // limits data preprocessing to one process\n                \"--max_steps=1\",              // limits training to just one step\n                \"--batch_size=1\",             // minimizes batch size\n                \"--micro_batch_size=1\",       // minimizes batch size\n                \"--val_set_size=0\",           // disables validation\n                \"--sample_packing=False\",     // disables sample packing which is necessary for small datasets\n                \"--eval_sample_packing=False\",// disables sample packing on eval set\n                \"--dataset_prepared_path=temp_debug/axolotl_outputs/data\", // send data outputs to a temp folder\n                \"--output_dir=temp_debug/axolotl_outputs/model\" // send model outputs to a temp folder\n                ],\n            \"console\": \"integratedTerminal\",      // show output in the integrated terminal\n            \"cwd\": \"${workspaceFolder}/devtools\", // set working directory to devtools from the root of the project\n            \"justMyCode\": true,                   // step through only axolotl code\n            \"env\": {\"CUDA_VISIBLE_DEVICES\": \"0\",  // Since we aren't doing distributed training, we need to limit to one GPU\n                    \"HF_HOME\": \"${workspaceFolder}/devtools/temp_debug/.hf-cache\"}, // send HF cache to a temp folder\n            \"preLaunchTask\": \"cleanup-for-dataprep\", // delete temp folders (see below)\n        }\n    ]\n}\nAdditional notes about this configuration:\n\nThe argument justMyCode is set to true such that you step through only the axolotl code. If you want to step into dependencies, set this to false.\nThe preLaunchTask: cleanup-for-dataprep is defined in .vscode/tasks.json and is used to delete the following folders before debugging, which is essential to ensure that the data pre-processing code is run from scratch:\n\n./devtools/temp_debug/axolotl_outputs\n./devtools/temp_debug/.hf-cache/datasets\n\n\n\n[!Tip] You may not want to delete these folders. For example, if you are debugging model training instead of data pre-processing, you may NOT want to delete the cache or output folders. You may also need to add additional tasks to the tasks.json file depending on your use case.\n\nBelow is the ./vscode/tasks.json file that defines the cleanup-for-dataprep task. This task is run before each debugging session when you use the above configuration. Note how there are two tasks that delete the two folders mentioned above. The third task cleanup-for-dataprep is a composite task that combines the two tasks. A composite task is necessary because VSCode does not allow you to specify multiple tasks in the preLaunchTask argument of the launch.json file.\n// .vscode/tasks.json\n// this file is used by launch.json\n{\n    \"version\": \"2.0.0\",\n    \"tasks\": [\n      // this task changes into the devtools directory and deletes the temp_debug/axolotl_outputs folder\n      {\n        \"label\": \"delete-outputs\",\n        \"type\": \"shell\",\n        \"command\": \"rm -rf temp_debug/axolotl_outputs\",\n        \"options\":{ \"cwd\": \"${workspaceFolder}/devtools\"},\n        \"problemMatcher\": []\n      },\n      // this task changes into the devtools directory and deletes the `temp_debug/.hf-cache/datasets` folder\n      {\n        \"label\": \"delete-temp-hf-dataset-cache\",\n        \"type\": \"shell\",\n        \"command\": \"rm -rf temp_debug/.hf-cache/datasets\",\n        \"options\":{ \"cwd\": \"${workspaceFolder}/devtools\"},\n        \"problemMatcher\": []\n      },\n        // this task combines the two tasks above\n      {\n       \"label\": \"cleanup-for-dataprep\",\n       \"dependsOn\": [\"delete-outputs\", \"delete-temp-hf-dataset-cache\"],\n      }\n    ]\n}\n\n\nCustomizing your debugger\nYour debugging use case may differ from the example above. The easiest thing to do is to put your own axolotl config in the devtools folder and modify the launch.json file to use your config. You may also want to modify the preLaunchTask to delete different folders or not delete anything at all.\n\n\nVideo Tutorial\nThe following video tutorial walks through the above configuration and demonstrates how to debug with VSCode, (click the image below to watch):\n\n\n\nHamel Husain’s tutorial: Debugging Axolotl w/VSCode",
    "crumbs": [
      "How-To Guides",
      "Debugging"
    ]
  },
  {
    "objectID": "docs/debugging.html#debugging-with-docker",
    "href": "docs/debugging.html#debugging-with-docker",
    "title": "Debugging",
    "section": "Debugging With Docker",
    "text": "Debugging With Docker\nUsing official Axolotl Docker images is a great way to debug your code, and is a very popular way to use Axolotl. Attaching VSCode to Docker takes a few more steps.\n\nSetup\nOn the host that is running axolotl (ex: if you are using a remote host), clone the axolotl repo and change your current directory to the root:\ngit clone https://github.com/OpenAccess-AI-Collective/axolotl\ncd axolotl\n\n[!Tip] If you already have axolotl cloned on your host, make sure you have the latest changes and change into the root of the project.\n\nNext, run the desired docker image and mount the current directory. Below is a docker command you can run to do this:2\ndocker run --privileged --gpus '\"all\"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src=\"${PWD}\",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface winglian/axolotl:main-py3.10-cu118-2.0.1\n\n[!Tip] To understand which containers are available, see the Docker section of the README and the DockerHub repo. For details of how the Docker containers are built, see axolotl’s Docker CI builds.\n\nYou will now be in the container. Next, perform an editable install of Axolotl:\npip3 install packaging\npip3 install -e '.[flash-attn,deepspeed]'\n\n\nAttach To Container\nNext, if you are using a remote host, Remote into this host with VSCode. If you are using a local host, you can skip this step.\nNext, select Dev Containers: Attach to Running Container... using the command palette (CMD + SHIFT + P) in VSCode. You will be prompted to select a container to attach to. Select the container you just created. You will now be in the container with a working directory that is at the root of the project. Any changes you make to the code will be reflected both in the container and on the host.\nNow you are ready to debug as described above (see Debugging with VSCode).\n\n\nVideo - Attaching To Docker On Remote Host\nHere is a short video that demonstrates how to attach to a Docker container on a remote host:\n\n\n\nHamel Husain’s tutorial: Debugging Axolotl Part 2: Attaching to Docker on a Remote Host",
    "crumbs": [
      "How-To Guides",
      "Debugging"
    ]
  },
  {
    "objectID": "docs/debugging.html#footnotes",
    "href": "docs/debugging.html#footnotes",
    "title": "Debugging",
    "section": "Footnotes",
    "text": "Footnotes\n\n\nThe config actually mimics the command CUDA_VISIBLE_DEVICES=0 python -m accelerate.commands.launch -m axolotl.cli.train devtools/sharegpt.yml, but this is the same thing.↩︎\nMany of the below flags are recommended best practices by Nvidia when using nvidia-container-toolkit. You can read more about these flags here.↩︎",
    "crumbs": [
      "How-To Guides",
      "Debugging"
    ]
  },
  {
    "objectID": "docs/input_output.html",
    "href": "docs/input_output.html",
    "title": "Template-free prompt construction",
    "section": "",
    "text": "Background\n\nMasking Inputs\nYou may not want prompt templates\nThe input_output format\n\nUsage\n\n1. Prepare Data\n2. Use type: input_output\n3. Check the prompts",
    "crumbs": [
      "How-To Guides",
      "Template-free prompt construction"
    ]
  },
  {
    "objectID": "docs/input_output.html#background",
    "href": "docs/input_output.html#background",
    "title": "Template-free prompt construction",
    "section": "Background",
    "text": "Background\n\n\nMasking Inputs\nOne of the most popular features of axolotl is setting the following configuration value:\ntrain_on_inputs: false\nIf you declare a dataset formats such as alpaca or chatml, axolotl knows what is an input (i.e. human) vs. an output (i.e. the assistant) and masks the input labels so that your model can focus on predicting the outputs only.\n\n\n\nYou may not want prompt templates\nHowever, there are many situations where you don’t want to use one of these formats or templates (I usually don’t!). This is because they can:\n\nAdd unnecessary boilerplate to your prompts.\nCreate artifacts like special delimiters <|im_start|> that can quickly become footguns if you don’t include them correctly at inference time.\nEnforce a chat interface when you do not want one. Sometimes you just want to fine-tune a model to a very specific task and do NOT want multi-turn conversations, roles, etc.\nLimit you to only certain roles that the template allows.\n\n\n\n\nThe input_output format\nYou can construct your prompts without a template by using the input_output format, by setting type: input_output in your configuration file like this:\nconfig.yml\ntrain_on_inputs: false # Mask segments of your data\ndatasets:\n  - path: output.jsonl\n    type: input_output  # use template free prompt construction\nUnlike type: completion, which is also template-free, type: input_output allows you to mask segments of your text. More details on how this works are described below.",
    "crumbs": [
      "How-To Guides",
      "Template-free prompt construction"
    ]
  },
  {
    "objectID": "docs/input_output.html#usage",
    "href": "docs/input_output.html#usage",
    "title": "Template-free prompt construction",
    "section": "Usage",
    "text": "Usage\nThis is how you can use the input_output format:\n\n\n1. Prepare Data\nTo use the input_output format, collect your data in the following format into a jsonl file (below is the first row from the file output.jsonl` pretty printed):\n$ head -n1 output.jsonl | python -m json.tool\n\n{.cell-output .cell-output-stdout}\n    {\n        \"segments\": [\n            {\n                \"label\": true,\n                \"text\": \"<s>Hello\\n\"\n            },\n            {\n                \"label\": true,\n                \"text\": \"hi there!. \"\n            },\n            {\n                \"label\": false,\n                \"text\": \"goodbye \"\n            },\n            {\n                \"label\": true,\n                \"text\": \"farewell</s>\"\n            }\n        ]\n    }\nSet label:false when you want to mask a segment of text so that the model isn’t trained on it. Some things to keep in mind:\n\n[!IMPORTANT] 1. EOS, BOS, spaces, newlines etc. are entirely up to you. Axolotl concatenates all the segments as-is. The tokenizer doesn’t add anything additional. Notice how I added spaces, newlines, <s> (BOS), and </s> (EOS) myself. 2. Make sure you check the materialized output to validate that the prompt is getting assembled how you like.\n\n\n\n\n2. Use type: input_output\nLet’s materialize data with our output.jsonl file by setting type: input_output in our axolotl config:\n# training_config.yaml\nbase_model: mistralai/Mistral-7B-v0.1\ndata_seed: 49\nseed: 49\n\ndatasets:\n  - path: output.jsonl\n    type: input_output\nval_set_size: 0.1\n\nsequence_len: 896\nsample_packing: false\n\nmicro_batch_size: 2\ngradient_accumulation_steps: 3\neval_batch_size: 2\nnum_epochs: 1\nlearning_rate: 0.0002\n\ntrain_on_inputs: false\nspecial_tokens:\n  bos_token: \"<s>\"\n  eos_token: \"</s>\"\n  unk_token: \"<unk>\"\nYou can use the following command to materialize your data. The --debug flag will print the tokens, along with the labels so you can verify that the correct items are being ignored:\n$ python -m axolotl.cli.preprocess training_config.yaml --debug\n\n...\n[2024-03-05 23:36:46,969] [INFO] [axolotl.check_example_labels:35] [PID:607731] [RANK:0] <s>(1, 1) Hello(22557, 22557)\n(13, 13) hi(12014, 12014) there(736, 736) !(28808, 28808) .(28723, 28723) (28705, 28705) good(-100, 1179) bye(-100, 17664) (-100, 28705) fare(19111, 19111) well(5458, 5458) </s>(2, 2)\nThe format is decoded_token(label, token_id), for example, <s>(1, 1) means that the token is <s>, the label is 1 and the token_id is 1. When the label is -100 then that token is ignored for training.\n\n\n\n3. Check the prompts\nHere is another way to check the materialized output:\nfrom transformers import AutoTokenizer\nfrom datasets import load_from_disk\nimport yaml\n\ndirectory = !ls last_run_prepared/\nwith open('training_config.yaml', 'r') as f:\n    cfg = yaml.safe_load(f)\nmodel_id = cfg['base_model']\ntok = AutoTokenizer.from_pretrained(model_id)\nds = load_from_disk(f'last_run_prepared/{directory[0]}/')\n>>> row = ds[0]\n>>> print(tok.decode(row['input_ids']))\n<s> Hello\n    hi there!.  goodbye  farewell</s>\nWe can check that the right tokens are ingored by comparing the labels to each token:\nimport pandas as pd\npd.DataFrame([{'token': tok.decode(i), 'label': l, 'id':i} for i,l in\n              zip(row['input_ids'], row['labels'])])\n\n\n\ntoken\nlabel\nid\n\n\n\n\n0\n<s>\n1\n\n\n1\nHello\n22557\n\n\n2\n\\n\n13\n\n\n3\nhi\n12014\n\n\n4\nthere\n736\n\n\n5\n!\n28808\n\n\n6\n.\n28723\n\n\n7\n\n28705\n\n\n8\ngood\n-100\n\n\n9\nbye\n-100\n\n\n10\n\n-100\n\n\n11\nfare\n19111\n\n\n12\nwell\n5458\n\n\n13\n</s>\n2\n\n\n\nIf we look at the input data, the above table seems correct! (The jsonl version is repeated below for reference):\n$ head -n1 output.jsonl | python -m json.tool\n\n{.cell-output .cell-output-stdout}\n    {\n        \"segments\": [\n            {\n                \"label\": true,\n                \"text\": \"<s>Hello\\n\"\n            },\n            {\n                \"label\": true,\n                \"text\": \"hi there!. \"\n            },\n            {\n                \"label\": false,\n                \"text\": \"goodbye \"\n            },\n            {\n                \"label\": true,\n                \"text\": \"farewell</s>\"\n            }\n        ]\n    }",
    "crumbs": [
      "How-To Guides",
      "Template-free prompt construction"
    ]
  },
  {
    "objectID": "examples/colab-notebooks/colab-axolotl-example.html",
    "href": "examples/colab-notebooks/colab-axolotl-example.html",
    "title": "Example notebook for running Axolotl on google colab",
    "section": "",
    "text": "import torch\n# Check so there is a gpu available, a T4(free tier) is enough to run this notebook\nassert (torch.cuda.is_available()==True)"
  },
  {
    "objectID": "examples/colab-notebooks/colab-axolotl-example.html#install-axolotl-and-dependencies",
    "href": "examples/colab-notebooks/colab-axolotl-example.html#install-axolotl-and-dependencies",
    "title": "Example notebook for running Axolotl on google colab",
    "section": "Install Axolotl and dependencies",
    "text": "Install Axolotl and dependencies\n\n!pip install torch==\"2.1.2\"\n!pip install -e git+https://github.com/OpenAccess-AI-Collective/axolotl#egg=axolotl\n!pip install flash-attn==\"2.5.0\"\n!pip install deepspeed==\"0.13.1\""
  },
  {
    "objectID": "examples/colab-notebooks/colab-axolotl-example.html#create-an-yaml-config-file",
    "href": "examples/colab-notebooks/colab-axolotl-example.html#create-an-yaml-config-file",
    "title": "Example notebook for running Axolotl on google colab",
    "section": "Create an yaml config file",
    "text": "Create an yaml config file\n\nimport yaml\n\n# Your YAML string\nyaml_string = \"\"\"\nbase_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T\nmodel_type: LlamaForCausalLM\ntokenizer_type: LlamaTokenizer\nis_llama_derived_model: true\n\nload_in_8bit: false\nload_in_4bit: true\nstrict: false\n\ndatasets:\n  - path: mhenrichsen/alpaca_2k_test\n    type: alpaca\ndataset_prepared_path:\nval_set_size: 0.05\noutput_dir: ./qlora-out\n\nadapter: qlora\nlora_model_dir:\n\nsequence_len: 1096\nsample_packing: true\npad_to_sequence_len: true\n\nlora_r: 32\nlora_alpha: 16\nlora_dropout: 0.05\nlora_target_modules:\nlora_target_linear: true\nlora_fan_in_fan_out:\n\nwandb_project:\nwandb_entity:\nwandb_watch:\nwandb_name:\nwandb_log_model:\n\nmlflow_experiment_name: colab-example\n\ngradient_accumulation_steps: 1\nmicro_batch_size: 1\nnum_epochs: 4\nmax_steps: 20\noptimizer: paged_adamw_32bit\nlr_scheduler: cosine\nlearning_rate: 0.0002\n\ntrain_on_inputs: false\ngroup_by_length: false\nbf16: false\nfp16: true\ntf32: false\n\ngradient_checkpointing: true\nearly_stopping_patience:\nresume_from_checkpoint:\nlocal_rank:\nlogging_steps: 1\nxformers_attention:\nflash_attention: false\n\nwarmup_steps: 10\nevals_per_epoch:\nsaves_per_epoch:\ndebug:\ndeepspeed:\nweight_decay: 0.0\nfsdp:\nfsdp_config:\nspecial_tokens:\n\n\"\"\"\n\n# Convert the YAML string to a Python dictionary\nyaml_dict = yaml.safe_load(yaml_string)\n\n# Specify your file path\nfile_path = 'test_axolotl.yaml'\n\n# Write the YAML file\nwith open(file_path, 'w') as file:\n    yaml.dump(yaml_dict, file)"
  },
  {
    "objectID": "examples/colab-notebooks/colab-axolotl-example.html#launch-the-training",
    "href": "examples/colab-notebooks/colab-axolotl-example.html#launch-the-training",
    "title": "Example notebook for running Axolotl on google colab",
    "section": "Launch the training",
    "text": "Launch the training\n\n# Buy using the ! the comand will be executed as a bash command\n!accelerate launch -m axolotl.cli.train /content/test_axolotl.yaml"
  },
  {
    "objectID": "examples/colab-notebooks/colab-axolotl-example.html#play-with-inference",
    "href": "examples/colab-notebooks/colab-axolotl-example.html#play-with-inference",
    "title": "Example notebook for running Axolotl on google colab",
    "section": "Play with inference",
    "text": "Play with inference\n\n# Buy using the ! the comand will be executed as a bash command\n!accelerate launch -m axolotl.cli.inference /content/test_axolotl.yaml \\\n    --qlora_model_dir=\"./qlora-out\" --gradio"
  },
  {
    "objectID": "index.html",
    "href": "index.html",
    "title": "Axolotl",
    "section": "",
    "text": "Axolotl is a tool designed to streamline the fine-tuning of various AI models, offering support for multiple configurations and architectures.\nFeatures: - Train various Huggingface models such as llama, pythia, falcon, mpt - Supports fullfinetune, lora, qlora, relora, and gptq - Customize configurations using a simple yaml file or CLI overwrite - Load different dataset formats, use custom formats, or bring your own tokenized datasets - Integrated with xformer, flash attention, rope scaling, and multipacking - Works with single GPU or multiple GPUs via FSDP or Deepspeed - Easily run with Docker locally or on the cloud - Log results and optionally checkpoints to wandb or mlflow - And more!\n  \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nfp16/fp32\nlora\nqlora\ngptq\ngptq w/flash attn\nflash attn\nxformers attn\n\n\n\n\nllama\n✅\n✅\n✅\n✅\n✅\n✅\n✅\n\n\nMistral\n✅\n✅\n✅\n✅\n✅\n✅\n✅\n\n\nMixtral-MoE\n✅\n✅\n✅\n❓\n❓\n❓\n❓\n\n\nPythia\n✅\n✅\n✅\n❌\n❌\n❌\n❓\n\n\ncerebras\n✅\n✅\n✅\n❌\n❌\n❌\n❓\n\n\nbtlm\n✅\n✅\n✅\n❌\n❌\n❌\n❓\n\n\nmpt\n✅\n❌\n❓\n❌\n❌\n❌\n❓\n\n\nfalcon\n✅\n✅\n✅\n❌\n❌\n❌\n❓\n\n\ngpt-j\n✅\n✅\n✅\n❌\n❌\n❓\n❓\n\n\nXGen\n✅\n❓\n✅\n❓\n❓\n❓\n✅\n\n\nphi\n✅\n✅\n✅\n❓\n❓\n❓\n❓\n\n\nRWKV\n✅\n❓\n❓\n❓\n❓\n❓\n❓\n\n\nQwen\n✅\n✅\n✅\n❓\n❓\n❓\n❓\n\n\nGemma\n✅\n✅\n✅\n❓\n❓\n✅\n❓\n\n\n\n✅: supported ❌: not supported ❓: untested\n\n\n\nGet started with Axolotl in just a few steps! This quickstart guide will walk you through setting up and running a basic fine-tuning task.\nRequirements: Python >=3.10 and Pytorch >=2.1.1.\ngit clone https://github.com/OpenAccess-AI-Collective/axolotl\ncd axolotl\n\npip3 install packaging\npip3 install -e '.[flash-attn,deepspeed]'\n\n\n# preprocess datasets - optional but recommended\nCUDA_VISIBLE_DEVICES=\"\" python -m axolotl.cli.preprocess examples/openllama-3b/lora.yml\n\n# finetune lora\naccelerate launch -m axolotl.cli.train examples/openllama-3b/lora.yml\n\n# inference\naccelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \\\n    --lora_model_dir=\"./lora-out\"\n\n# gradio\naccelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \\\n    --lora_model_dir=\"./lora-out\" --gradio\n\n# remote yaml files - the yaml config can be hosted on a public URL\n# Note: the yaml config must directly link to the **raw** yaml\naccelerate launch -m axolotl.cli.train https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/examples/openllama-3b/lora.yml\n\n\n\n\n\n\n\n\ndocker run --gpus '\"all\"' --rm -it winglian/axolotl:main-latest\nOr run on the current files for development:\ndocker compose up -d\n\n[!Tip] If you want to debug axolotl or prefer to use Docker as your development environment, see the debugging guide’s section on Docker.\n\n\n\nDocker advanced\n\nA more powerful Docker command to run would be this:\ndocker run --privileged --gpus '\"all\"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src=\"${PWD}\",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface winglian/axolotl:main-latest\nIt additionally: * Prevents memory issues when running e.g. deepspeed (e.g. you could hit SIGBUS/signal 7 error) through --ipc and --ulimit args. * Persists the downloaded HF data (models etc.) and your modifications to axolotl code through --mount/-v args. * The --name argument simply makes it easier to refer to the container in vscode (Dev Containers: Attach to Running Container...) or in your terminal. * The --privileged flag gives all capabilities to the container. * The --shm-size 10g argument increases the shared memory size. Use this if you see exitcode: -7 errors using deepspeed.\nMore information on nvidia website\n\n\n\n\n\nInstall python >=3.10\nInstall pytorch stable https://pytorch.org/get-started/locally/\nInstall Axolotl along with python dependencies bash         pip3 install packaging         pip3 install -e '.[flash-attn,deepspeed]'\n(Optional) Login to Huggingface to use gated models/datasets. bash         huggingface-cli login Get the token at huggingface.co/settings/tokens\n\n\n\n\nFor cloud GPU providers that support docker images, use winglian/axolotl-cloud:main-latest\n\non Latitude.sh use this direct link\non JarvisLabs.ai use this direct link\non RunPod use this direct link\n\n\n\n\n\n\n\n\nClick to Expand\n\n\nInstall python\n\nsudo apt update\nsudo apt install -y python3.10\n\nsudo update-alternatives --install /usr/bin/python python /usr/bin/python3.10 1\nsudo update-alternatives --config python # pick 3.10 if given option\npython -V # should be 3.10\n\nInstall pip\n\nwget https://bootstrap.pypa.io/get-pip.py\npython get-pip.py\n\nInstall torch\n\npip3 install -U torch --index-url https://download.pytorch.org/whl/cu118\n\nAxolotl\n\ngit clone https://github.com/OpenAccess-AI-Collective/axolotl\ncd axolotl\n\npip3 install packaging\npip3 install -e '.[flash-attn,deepspeed]'\npip3 install protobuf==3.20.3\npip3 install -U --ignore-installed requests Pillow psutil scipy\n\nSet path\n\nexport LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH\n\n\n\n\n\n\nClick to Expand\n\nUse a Deeplearning linux OS with cuda and pytorch installed. Then follow instructions on quickstart.\nMake sure to run the below to uninstall xla.\npip uninstall -y torch_xla[tpu]\n\n\n\n\n\nPlease use WSL or Docker!\n\n\n\nUse the below instead of the install method in QuickStart.\npip3 install -e '.'\nMore info: mac.md\n\n\n\nTo launch on GPU instances (both on-demand and spot instances) on 7+ clouds (GCP, AWS, Azure, OCI, and more), you can use SkyPilot:\npip install \"skypilot-nightly[gcp,aws,azure,oci,lambda,kubernetes,ibm,scp]\"  # choose your clouds\nsky check\nGet the example YAMLs of using Axolotl to finetune mistralai/Mistral-7B-v0.1:\ngit clone https://github.com/skypilot-org/skypilot.git\ncd skypilot/llm/axolotl\nUse one command to launch:\n# On-demand\nHF_TOKEN=xx sky launch axolotl.yaml --env HF_TOKEN\n\n# Managed spot (auto-recovery on preemption)\nHF_TOKEN=xx BUCKET=<unique-name> sky spot launch axolotl-spot.yaml --env HF_TOKEN --env BUCKET\n\n\n\n\nAxolotl supports a variety of dataset formats. Below are some of the formats you can use. Have dataset(s) in one of the following format (JSONL recommended):\n\n\n\ncompletion: raw corpus\n{\"text\": \"...\"}\n\nNote: Axolotl usually loads the entire dataset into memory. This will be challenging for large datasets. Use the following config to enable streaming:\npretraining_dataset: # hf path only\n\n\n\n\n\n\nalpaca: instruction; input(optional)\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\"}\n\n\n\nSee other formats\n\n\njeopardy: question and answer\n{\"question\": \"...\", \"category\": \"...\", \"answer\": \"...\"}\noasst: instruction\n{\"INSTRUCTION\": \"...\", \"RESPONSE\": \"...\"}\ngpteacher: instruction; input(optional)\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}\nreflection: instruction with reflect; input(optional)\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\", \"reflection\": \"...\", \"corrected\": \"...\"}\nexplainchoice: question, choices, (solution OR explanation)\n{\"question\": \"...\", \"choices\": [\"...\"], \"solution\": \"...\", \"explanation\": \"...\"}\nconcisechoice: question, choices, (solution OR explanation)\n{\"question\": \"...\", \"choices\": [\"...\"], \"solution\": \"...\", \"explanation\": \"...\"}\nsummarizetldr: article and summary\n{\"article\": \"...\", \"summary\": \"...\"}\nalpaca_chat: basic instruct for alpaca chat\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}\nalpaca_chat.load_qa: question and answer for alpaca chat\n{\"question\": \"...\", \"answer\": \"...\"}\nalpaca_chat.load_concise: question and answer for alpaca chat, for concise answers\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}\nalpaca_chat.load_camel_ai: question and answer for alpaca chat, for load_camel_ai\n{\"message_1\": \"...\", \"message_2\": \"...\"}\nalpaca_w_system.load_open_orca: support for open orca datasets with included system prompts, instruct\n{\"system_prompt\": \"...\", \"question\": \"...\", \"response\": \"...\"}\ncontext_qa: in context question answering from an article\n{\"article\": \"...\", \"question\": \"...\", \"answer\": \"...\"}\ncontext_qa.load_v2: in context question answering (alternate)\n{\"context\": \"...\", \"question\": \"...\", \"answer\": \"...\"}\ncontext_qa.load_404: in context question answering from an article, with default response for no answer from context\n{\"article\": \"...\", \"unanswerable_question\": \"...\"}\ncreative_acr.load_answer: instruction and revision\n{\"instruction\": \"...\", \"revision\": \"...\"}\ncreative_acr.load_critique: critique\n{\"scores\": \"...\", \"critiques\": \"...\", \"instruction\": \"...\", \"answer\": \"...\"}\ncreative_acr.load_revise: critique and revise\n{\"scores\": \"...\", \"critiques\": \"...\", \"instruction\": \"...\", \"answer\": \"...\", \"revision\": \"...\"}\nmetharme: instruction, adds additional eos tokens\n{\"prompt\": \"...\", \"generation\": \"...\"}\n\n\n\n\n\n\ninput_output: template-free prompt construction\n {\"segments\": [{\"label\": true|false, \"text\": \"...\"}]}\n\nThis is a special format that allows you to construct prompts without using templates. This is for advanced users who want more freedom with prompt construction. See these docs for more details.\n\n\n\n\nsharegpt: conversations where from is human/gpt. (optional: first row with role system to override default system prompt)\n{\"conversations\": [{\"from\": \"...\", \"value\": \"...\"}]}\n\n\n\nSee other formats\n\n\npygmalion: pygmalion\n{\"conversations\": [{\"role\": \"...\", \"value\": \"...\"}]}\nsharegpt.load_role: conversations where role is used instead of from\n{\"conversations\": [{\"role\": \"...\", \"value\": \"...\"}]}\nsharegpt.load_guanaco: conversations where from is prompter/assistant instead of default sharegpt\n{\"conversations\": [{\"from\": \"...\", \"value\": \"...\"}]}\nsharegpt_jokes: creates a chat where bot is asked to tell a joke, then explain why the joke is funny\n{\"conversations\": [{\"title\": \"...\", \"text\": \"...\", \"explanation\": \"...\"}]}\n\n\nNote: type: sharegpt opens a special config conversation: that enables conversions to many Conversation types. See dataset section under all yaml options.\n\n\n\n\nFor a dataset that is preprocessed for instruction purposes:\n{\"input\": \"...\", \"output\": \"...\"}\nYou can use this example in your YAML config:\ndatasets:\n  - path: repo\n    type:\n      system_prompt: \"\"\n      field_system: system\n      field_instruction: input\n      field_output: output\n      format: \"[INST] {instruction} [/INST]\"\n      no_input_format: \"[INST] {instruction} [/INST]\"\nSee full config options under all yaml options.\n\n\n\n\nDo not pass a type:\nColumns in Dataset must be exactly input_ids, attention_mask, labels\n\n- path: ...\n\n\n\n\nSee examples for quick start. It is recommended to duplicate and modify to your needs. The most important options are:\n\nmodel\nbase_model: ./llama-7b-hf # local or huggingface repo\nNote: The code will load the right architecture.\ndataset\ndatasets:\n    # huggingface repo\n  - path: vicgalle/alpaca-gpt4\n    type: alpaca\n\n    # huggingface repo with specific configuration/subset\n  - path: EleutherAI/pile\n    name: enron_emails\n    type: completion # format from earlier\n    field: text # Optional[str] default: text, field to use for completion data\n\n    # huggingface repo with multiple named configurations/subsets\n  - path: bigcode/commitpackft\n    name:\n      - ruby\n      - python\n      - typescript\n    type: ... # unimplemented custom format\n\n    # fastchat conversation\n    # See 'conversation' options: https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py\n  - path: ...\n    type: sharegpt\n    conversation: chatml # default: vicuna_v1.1\n\n    # local\n  - path: data.jsonl # or json\n    ds_type: json # see other options below\n    type: alpaca\n\n    # dataset with splits, but no train split\n  - path: knowrohit07/know_sql\n    type: context_qa.load_v2\n    train_on_split: validation\n\n    # loading from s3 or gcs\n    # s3 creds will be loaded from the system default and gcs only supports public access\n  - path: s3://path_to_ds # Accepts folder with arrow/parquet or file path like above. Supports s3, gcs.\n    ...\n\n    # Loading Data From a Public URL\n    # - The file format is `json` (which includes `jsonl`) by default. For different formats, adjust the `ds_type` option accordingly.\n  - path: https://some.url.com/yourdata.jsonl # The URL should be a direct link to the file you wish to load. URLs must use HTTPS protocol, not HTTP.\n    ds_type: json # this is the default, see other options below.\nloading\nload_in_4bit: true\nload_in_8bit: true\n\nbf16: auto # require >=ampere, auto will detect if your GPU supports this and choose automatically.\nfp16: # leave empty to use fp16 when bf16 is 'auto'. set to false if you want to fallback to fp32\ntf32: true # require >=ampere\n\nbfloat16: true # require >=ampere, use instead of bf16 when you don't want AMP (automatic mixed precision)\nfloat16: true # use instead of fp16 when you don't want AMP\nNote: Repo does not do 4-bit quantization.\nlora\nadapter: lora # 'qlora' or leave blank for full finetune\nlora_r: 8\nlora_alpha: 16\nlora_dropout: 0.05\nlora_target_modules:\n  - q_proj\n  - v_proj\n\n\n\nAll yaml options (click to expand)\n\n# This is the huggingface model that contains *.pt, *.safetensors, or *.bin files\n# This can also be a relative path to a model on disk\nbase_model: ./llama-7b-hf\n# You can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)\nbase_model_ignore_patterns:\n# If the base_model repo on hf hub doesn't include configuration .json files,\n# You can set that here, or leave this empty to default to base_model\nbase_model_config: ./llama-7b-hf\n# You can specify to choose a specific model revision from huggingface hub\nrevision_of_model:\n# Optional tokenizer configuration path in case you want to use a different tokenizer\n# than the one defined in the base model\ntokenizer_config:\n# If you want to specify the type of model to load, AutoModelForCausalLM is a good choice too\nmodel_type: AutoModelForCausalLM\n# Corresponding tokenizer for the model AutoTokenizer is a good choice\ntokenizer_type: AutoTokenizer\n# Trust remote code for untrusted source\ntrust_remote_code:\n# use_fast option for tokenizer loading from_pretrained, default to True\ntokenizer_use_fast:\n# Whether to use the legacy tokenizer setting, defaults to True\ntokenizer_legacy:\n# Resize the model embeddings when new tokens are added to multiples of 32\n# This is reported to improve training speed on some models\nresize_token_embeddings_to_32x:\n\n# (Internal use only)\n# Used to identify which the model is based on\nis_falcon_derived_model:\nis_llama_derived_model:\nis_qwen_derived_model:\n# Please note that if you set this to true, `padding_side` will be set to \"left\" by default\nis_mistral_derived_model:\n\n# optional overrides to the base model configuration\noverrides_of_model_config:\n  # RoPE Scaling https://github.com/huggingface/transformers/pull/24653\n  rope_scaling:\n    type: # linear | dynamic\n    factor: # float\n\n# optional overrides to the bnb 4bit quantization configuration\n# https://huggingface.co/docs/transformers/main/main_classes/quantization#transformers.BitsAndBytesConfig\nbnb_config_kwargs:\n  # These are default values\n  llm_int8_has_fp16_weight: false\n  bnb_4bit_quant_type: nf4\n  bnb_4bit_use_double_quant: true\n\n\n# Whether you are training a 4-bit GPTQ quantized model\ngptq: true\n\n# This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer\nload_in_8bit: true\n# Use bitsandbytes 4 bit\nload_in_4bit:\n\n# Use CUDA bf16\nbf16: true # bool or 'full' for `bf16_full_eval`. require >=ampere\n# Use CUDA fp16\nfp16: true\n# Use CUDA tf32\ntf32: true # require >=ampere\n\n# No AMP (automatic mixed precision)\nbfloat16: true # require >=ampere\nfloat16: true\n\n# Limit the memory for all available GPUs to this amount (if an integer, expressed in gigabytes); default: unset\ngpu_memory_limit: 20GiB\n# Do the LoRA/PEFT loading on CPU -- this is required if the base model is so large it takes up most or all of the available GPU VRAM, e.g. during a model and LoRA merge\nlora_on_cpu: true\n\n# A list of one or more datasets to finetune the model with\ndatasets:\n  # HuggingFace dataset repo | s3://,gs:// path | \"json\" for local dataset, make sure to fill data_files\n  - path: vicgalle/alpaca-gpt4\n  # The type of prompt to use for training. [alpaca, sharegpt, gpteacher, oasst, reflection]\n    type: alpaca # format | format:<prompt_style> (chat/instruct) | <prompt_strategies>.load_<load_fn>\n    ds_type: # Optional[str] (json|arrow|parquet|text|csv) defines the datatype when path is a file\n    data_files: # Optional[str] path to source data files\n    shards: # Optional[int] number of shards to split data into\n    name: # Optional[str] name of dataset configuration to load\n    train_on_split: train # Optional[str] name of dataset split to load from\n\n    # Optional[str] fastchat conversation type, only used with type: sharegpt\n    conversation: # Options (see Conversation 'name'): https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py\n    field_human: # Optional[str]. Human key to use for conversation.\n    field_model: # Optional[str]. Assistant key to use for conversation.\n    # Add additional keys from your dataset as input or output roles\n    roles:\n      input: # Optional[List[str]]. These will be masked based on train_on_input\n      output: # Optional[List[str]].\n\n  # Custom user instruction prompt\n  - path: repo\n    type:\n      # The below are defaults. only set what's needed if you use a different column name.\n      system_prompt: \"\"\n      system_format: \"{system}\"\n      field_system: system\n      field_instruction: instruction\n      field_input: input\n      field_output: output\n\n      # Customizable to be single line or multi-line\n      # Use {instruction}/{input} as key to be replaced\n      # 'format' can include {input}\n      format: |-\n        User: {instruction} {input}\n        Assistant:\n      # 'no_input_format' cannot include {input}\n      no_input_format: \"{instruction} \"\n\n      # For `completion` datsets only, uses the provided field instead of `text` column\n      field:\n\n# If false, the datasets will not be shuffled and will keep their original order in `datasets`.\n# The same applies to the `test_datasets` option and the `pretraining_dataset` option. Default is true.\nshuffle_merged_datasets: true\n\n# A list of one or more datasets to eval the model with.\n# You can use either test_datasets, or val_set_size, but not both.\ntest_datasets:\n  - path: /workspace/data/eval.jsonl\n    ds_type: json\n    # You need to specify a split. For \"json\" datasets the default split is called \"train\".\n    split: train\n    type: completion\n    data_files:\n      - /workspace/data/eval.jsonl\n\n# use RL training: 'dpo', 'ipo', 'kto_pair'\nrl:\n\n# Saves the desired chat template to the tokenizer_config.json for easier inferencing\n# Currently supports chatml and inst (mistral/mixtral)\nchat_template: chatml\n# Changes the default system message\ndefault_system_message: You are a helpful assistant. Please give a long and detailed answer. # Currently only supports chatml.\n# Axolotl attempts to save the dataset as an arrow after packing the data together so\n# subsequent training attempts load faster, relative path\ndataset_prepared_path: data/last_run_prepared\n# Push prepared dataset to hub\npush_dataset_to_hub: # repo path\n# The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`\n# if not set.\ndataset_processes: # defaults to os.cpu_count() if not set\n# Keep dataset in memory while preprocessing\n# Only needed if cached dataset is taking too much storage\ndataset_keep_in_memory:\n# push checkpoints to hub\nhub_model_id: # private repo path to push finetuned model\n# how to push checkpoints to hub\n# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy\nhub_strategy:\n# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets\n# Required to be true when used in combination with `push_dataset_to_hub`\nhf_use_auth_token: # boolean\n# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval.\nval_set_size: 0.04\n# Num shards for whole dataset\ndataset_shard_num:\n# Index of shard to use for whole dataset\ndataset_shard_idx:\n\n# The maximum length of an input to train with, this should typically be less than 2048\n# as most models have a token/context limit of 2048\nsequence_len: 2048\n# Pad inputs so each step uses constant sized buffers\n# This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently\npad_to_sequence_len:\n# Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'\nsample_packing:\n# Set to 'false' if getting errors during eval with sample_packing on.\neval_sample_packing:\n# You can set these packing optimizations AFTER starting a training at least once.\n# The trainer will provide recommended values for these values.\nsample_packing_eff_est:\ntotal_num_tokens:\n\n# Passed through to transformers when loading the model when launched without accelerate\n# Use `sequential` when training w/ model parallelism to limit memory\ndevice_map:\n# Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model.\nmax_memory:\n\n# If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model\nadapter: lora\n# If you already have a lora model trained that you want to load, put that here.\n# This means after training, if you want to test the model, you should set this to the value of `output_dir`.\n# Note that if you merge an adapter to the base model, a new subdirectory `merged` will be created under the `output_dir`.\nlora_model_dir:\n\n# LoRA hyperparameters\n# For more details about the following options, see:\n# https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2\nlora_r: 8\nlora_alpha: 16\nlora_dropout: 0.05\nlora_target_modules:\n  - q_proj\n  - v_proj\n#  - k_proj\n#  - o_proj\n#  - gate_proj\n#  - down_proj\n#  - up_proj\nlora_target_linear: # If true, will target all linear modules\npeft_layers_to_transform: # The layer indices to transform, otherwise, apply to all layers\n\n# If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.\n# For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.\n# `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.\n# https://github.com/huggingface/peft/issues/334#issuecomment-1561727994\nlora_modules_to_save:\n#  - embed_tokens\n#  - lm_head\n\nlora_fan_in_fan_out: false\n\npeft:\n  # Configuration options for loftq initialization for LoRA\n  # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization\n  loftq_config:\n    loftq_bits:  # typically 4 bits\n\n# ReLoRA configuration\n# Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed\nrelora_steps: # Number of steps per ReLoRA restart\nrelora_warmup_steps: # Number of per-restart warmup steps\nrelora_anneal_steps: # Number of anneal steps for each relora cycle\nrelora_prune_ratio: # threshold for optimizer magnitude when pruning\nrelora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings\n\n# wandb configuration if you're using it\n# Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.\nwandb_mode: # \"offline\" to save run metadata locally and not sync to the server, \"disabled\" to turn off wandb\nwandb_project: # Your wandb project name\nwandb_entity: # A wandb Team name if using a Team\nwandb_watch:\nwandb_name: # Set the name of your wandb run\nwandb_run_id: # Set the ID of your wandb run\nwandb_log_model: # \"checkpoint\" to log model to wandb Artifacts every `save_steps` or \"end\" to log only at the end of training\n\n# mlflow configuration if you're using it\nmlflow_tracking_uri: # URI to mlflow\nmlflow_experiment_name: # Your experiment name\nhf_mlflow_log_artifacts:  # set to true to copy each saved checkpoint on each save to mlflow artifact registry\n\n# Where to save the full-finetuned model to\noutput_dir: ./completed-model\n\n# Whether to use torch.compile and which backend to use\ntorch_compile:  # bool\ntorch_compile_backend:  # Optional[str]\n\n# Training hyperparameters\n\n# If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.\ngradient_accumulation_steps: 1\n# The number of samples to include in each batch. This is the number of samples sent to each GPU.\nmicro_batch_size: 2\neval_batch_size:\nnum_epochs: 4\nwarmup_steps: 100  # cannot use with warmup_ratio\nwarmup_ratio: 0.05  # cannot use with warmup_steps\nlearning_rate: 0.00003\nlr_quadratic_warmup:\nlogging_steps:\neval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps\nevals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps\nsave_strategy: # Set to `no` to skip checkpoint saves\nsave_steps: # Leave empty to save at each epoch\nsaves_per_epoch: # number of times per epoch to save a checkpoint, mutually exclusive with save_steps\nsave_total_limit: # Checkpoints saved at a time\n# Maximum number of iterations to train for. It precedes num_epochs which means that\n# if both are set, num_epochs will not be guaranteed.\n# e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps\nmax_steps:\n\neval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0\neval_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128\neval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is [\"sacrebleu\", \"comet\", \"ter\", chrf]\n\nloss_watchdog_threshold: # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training)\nloss_watchdog_patience: # Number of high-loss steps in a row before the trainer aborts (default: 3)\n\n# Save model as safetensors (require safetensors package)\nsave_safetensors:\n\n# Whether to mask out or include the human's prompt from the training labels\ntrain_on_inputs: false\n# Group similarly sized data to minimize padding.\n# May be slower to start, as it must download and sort the entire dataset.\n# Note that training loss may have an oscillating pattern with this enabled.\ngroup_by_length: false\n\n# Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing\ngradient_checkpointing: false\n# additional kwargs to pass to the trainer for gradient checkpointing\n# gradient_checkpointing_kwargs:\n#   use_reentrant: true\n\n# Stop training after this many evaluation losses have increased in a row\n# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback\nearly_stopping_patience: 3\n\n# Specify a scheduler and kwargs to use with the optimizer\nlr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine\nlr_scheduler_kwargs:\ncosine_min_lr_ratio: # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr\ncosine_constant_lr_ratio: # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step (https://arxiv.org/pdf/2308.04014.pdf)\n\n# For one_cycle optim\nlr_div_factor: # Learning rate div factor\n\n# Specify optimizer\n# Valid values are driven by the Transformers OptimizerNames class, see:\n# https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134\n#\n# Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of\n# torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used\n# in the examples/ for your model and fine-tuning use case.\n#\n# Valid values for 'optimizer' include:\n# - adamw_hf\n# - adamw_torch\n# - adamw_torch_fused\n# - adamw_torch_xla\n# - adamw_apex_fused\n# - adafactor\n# - adamw_anyprecision\n# - sgd\n# - adagrad\n# - adamw_bnb_8bit\n# - lion_8bit\n# - lion_32bit\n# - paged_adamw_32bit\n# - paged_adamw_8bit\n# - paged_lion_32bit\n# - paged_lion_8bit\n# - galore_adamw\n# - galore_adamw_8bit\n# - galore_adafactor\n# - galore_adamw_layerwise\n# - galore_adamw_8bit_layerwise\n# - galore_adafactor_layerwise\noptimizer:\n# Dictionary of arguments to pass to the optimizer\noptim_args:\n# For Galore Optimizers the following optim_args are available\n# rank:  # type: int\n# update_proj_gap  # type: int\n# scale  # type: float\n# proj_type:  # type: str, default = std\n\n# The target modules to optimize, i.e. the module names that you would like to train, right now this is used only for GaLore algorithm\noptim_target_modules:\n# - self_attn  # for llama\n# - mlp\n\n# Specify weight decay\nweight_decay:\n# adamw hyperparams\nadam_beta1:\nadam_beta2:\nadam_epsilon:\n# Gradient clipping max norm\nmax_grad_norm:\n\n# Augmentation techniques\n# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings\n# currently only supported on Llama and Mistral\nneftune_noise_alpha:\n\n# Whether to bettertransformers\nflash_optimum:\n# Whether to use xformers attention patch https://github.com/facebookresearch/xformers:\nxformers_attention:\n# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:\nflash_attention:\nflash_attn_cross_entropy:  # Whether to use flash-attention cross entropy implementation - advanced use only\nflash_attn_rms_norm:  # Whether to use flash-attention rms norm implementation - advanced use only\nflash_attn_fuse_qkv: # Whether to fuse QKV into a single operation\nflash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation\n# Whether to use scaled-dot-product attention\n# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html\nsdp_attention:\n# Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf\ns2_attention:\n# Resume from a specific checkpoint dir\nresume_from_checkpoint:\n# If resume_from_checkpoint isn't set and you simply want it to start where it left off.\n# Be careful with this being turned on between different models.\nauto_resume_from_checkpoints: false\n\n# Don't mess with this, it's here for accelerate and torchrun\nlocal_rank:\n\n# Add or change special tokens.\n# If you add tokens here, you don't need to add them to the `tokens` list.\nspecial_tokens:\n  # bos_token: \"<s>\"\n  # eos_token: \"</s>\"\n  # unk_token: \"<unk>\"\n\n# Add extra tokens.\ntokens:\n\n# FSDP\nfsdp:\nfsdp_config:\n\n# Deepspeed config path. e.g., deepspeed_configs/zero3.json\ndeepspeed:\n\n# Advanced DDP Arguments\nddp_timeout:\nddp_bucket_cap_mb:\nddp_broadcast_buffers:\n\n# Path to torch distx for optim 'adamw_anyprecision'\ntorchdistx_path:\n\n# Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize\npretraining_dataset:\n\n# Debug mode\ndebug:\n\n# Seed\nseed:\n\n# Allow overwrite yml config using from cli\nstrict:\n\n\n\nUnderstanding of batch size and gradient accumulation steps\n\n Gradient accumulation means accumulating gradients over several mini-batches and updating the model weights afterward. When the samples in each batch are diverse, this technique doesn’t significantly impact learning.\nThis method allows for effective training with larger effective batch sizes without needing proportionally larger memory. Here’s why:\n\nMemory Consumption with Batch Size: The primary reason increasing the batch size impacts memory is due to the storage requirements for intermediate activations. When you forward propagate a batch through a network, you have to store the activations at each layer for each sample in the batch, because these activations are used during backpropagation to compute gradients. Therefore, larger batches mean more activations, leading to greater GPU memory consumption.\nGradient Accumulation: With gradient accumulation, you’re effectively simulating a larger batch size by accumulating gradients over several smaller batches (or micro-batches). However, at any given time, you’re only forward and backward propagating a micro-batch. This means you only store activations for the micro-batch, not the full accumulated batch. As a result, you can simulate the effect of a larger batch size without the memory cost of storing activations for a large batch.\n\nExample 1: Micro batch size: 3 Gradient accumulation steps: 2 Number of GPUs: 3 Total batch size = 3 * 2 * 3 = 18\n| GPU 1          | GPU 2          | GPU 3          |\n|----------------|----------------|----------------|\n| S1, S2, S3     | S4, S5, S6     | S7, S8, S9     |\n| e1, e2, e3     | e4, e5, e6     | e7, e8, e9     |\n|----------------|----------------|----------------|\n| → (accumulate) | → (accumulate) | → (accumulate) |\n|----------------|----------------|----------------|\n| S10, S11, S12  | S13, S14, S15  | S16, S17, S18  |\n| e10, e11, e12  | e13, e14, e15  | e16, e17, e18  |\n|----------------|----------------|----------------|\n| → (apply)      | → (apply)      | → (apply)      |\n\nAccumulated gradient for the weight w1 after the second iteration (considering all GPUs):\nTotal gradient for w1 = e1 + e2 + e3 + e4 + e5 + e6 + e7 + e8 + e9 + e10 + e11 + e12 + e13 + e14 + e15 + e16 + e17 + e18\n\nWeight update for w1:\nw1_new = w1_old - learning rate x (Total gradient for w1 / 18)\nExample 2: Micro batch size: 2 Gradient accumulation steps: 1 Number of GPUs: 3 Total batch size = 2 * 1 * 3 = 6\n| GPU 1     | GPU 2     | GPU 3     |\n|-----------|-----------|-----------|\n| S1, S2    | S3, S4    | S5, S6    |\n| e1, e2    | e3, e4    | e5, e6    |\n|-----------|-----------|-----------|\n| → (apply) | → (apply) | → (apply) |\n\nAccumulated gradient for the weight w1 (considering all GPUs):\nTotal gradient for w1 = e1 + e2 + e3 + e4 + e5 + e6\n\nWeight update for w1:\nw1_new = w1_old - learning rate × (Total gradient for w1 / 6)\n\n\n\n\nRun\naccelerate launch -m axolotl.cli.train your_config.yml\n\n[!TIP] You can also reference a config file that is hosted on a public URL, for example accelerate launch -m axolotl.cli.train https://yourdomain.com/your_config.yml\n\n\n\nYou can optionally pre-tokenize dataset with the following before finetuning. This is recommended for large datasets.\n\nSet dataset_prepared_path: to a local folder for saving and loading pre-tokenized dataset.\n(Optional): Set push_dataset_to_hub: hf_user/repo to push it to Huggingface.\n(Optional): Use --debug to see preprocessed examples.\n\npython -m axolotl.cli.preprocess your_config.yml\n\n\n\nBelow are the options available in axolotl for training with multiple GPUs. Note that DeepSpeed is the recommended multi-GPU option currently because FSDP may experience loss instability.\n\n\nDeepspeed is an optimization suite for multi-gpu systems allowing you to train much larger models than you might typically be able to fit into your GPU’s VRAM. More information about the various optimization types for deepspeed is available at https://huggingface.co/docs/accelerate/main/en/usage_guides/deepspeed#what-is-integrated\nWe provide several default deepspeed JSON configurations for ZeRO stage 1, 2, and 3.\ndeepspeed: deepspeed_configs/zero1.json\naccelerate launch -m axolotl.cli.train examples/llama-2/config.py --deepspeed deepspeed_configs/zero1.json\n\n\n\n\nllama FSDP\n\nfsdp:\n  - full_shard\n  - auto_wrap\nfsdp_config:\n  fsdp_offload_params: true\n  fsdp_state_dict_type: FULL_STATE_DICT\n  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer\n\n\n\nAxolotl supports training with FSDP and QLoRA, see these docs for more information.\n\n\n\nMake sure your WANDB_API_KEY environment variable is set (recommended) or you login to wandb with wandb login.\n\nwandb options\n\nwandb_mode:\nwandb_project:\nwandb_entity:\nwandb_watch:\nwandb_name:\nwandb_log_model:\n\n\n\nIt is important to have special tokens like delimiters, end-of-sequence, beginning-of-sequence in your tokenizer’s vocabulary. This will help you avoid tokenization issues and help your model train better. You can do this in axolotl like this:\nspecial_tokens:\n  bos_token: \"<s>\"\n  eos_token: \"</s>\"\n  unk_token: \"<unk>\"\ntokens: # these are delimiters\n  - \"<|im_start|>\"\n  - \"<|im_end|>\"\nWhen you include these tokens in your axolotl config, axolotl adds these tokens to the tokenizer’s vocabulary.\n\n\n\n\n\nAxolotl allows you to load your model in an interactive terminal playground for quick experimentation. The config file is the same config file used for training.\nPass the appropriate flag to the inference command, depending upon what kind of model was trained:\n\nPretrained LORA:\npython -m axolotl.cli.inference examples/your_config.yml --lora_model_dir=\"./lora-output-dir\"\nFull weights finetune:\npython -m axolotl.cli.inference examples/your_config.yml --base_model=\"./completed-model\"\nFull weights finetune w/ a prompt from a text file:\ncat /tmp/prompt.txt | python -m axolotl.cli.inference examples/your_config.yml \\\n  --base_model=\"./completed-model\" --prompter=None --load_in_8bit=True\n– With gradio hosting\npython -m axolotl.cli.inference examples/your_config.yml --gradio\n\nPlease use --sample_packing False if you have it on and receive the error similar to below:\n\nRuntimeError: stack expects each tensor to be equal size, but got [1, 32, 1, 128] at entry 0 and [1, 32, 8, 128] at entry 1\n\n\n\n\nThe following command will merge your LORA adapater with your base model. You can optionally pass the argument --lora_model_dir to specify the directory where your LORA adapter was saved, otherwhise, this will be inferred from output_dir in your axolotl config file. The merged model is saved in the sub-directory {lora_model_dir}/merged.\npython3 -m axolotl.cli.merge_lora your_config.yml --lora_model_dir=\"./completed-model\"\nYou may need to use the gpu_memory_limit and/or lora_on_cpu config options to avoid running out of memory. If you still run out of CUDA memory, you can try to merge in system RAM with\nCUDA_VISIBLE_DEVICES=\"\" python3 -m axolotl.cli.merge_lora ...\nalthough this will be very slow, and using the config options above are recommended instead.\n\n\n\n\nSee also the FAQ’s and debugging guide.\n\nIf you encounter a ‘Cuda out of memory’ error, it means your GPU ran out of memory during the training process. Here’s how to resolve it:\n\nPlease reduce any below - micro_batch_size - eval_batch_size - gradient_accumulation_steps - sequence_len\nIf it does not help, try running without deepspeed and without accelerate (replace “accelerate launch” with “python”) in the command.\nUsing adamw_bnb_8bit might also save you some memory.\n\nfailed (exitcode: -9)\n\nUsually means your system has run out of system memory. Similarly, you should consider reducing the same settings as when you run out of VRAM. Additionally, look into upgrading your system RAM which should be simpler than GPU upgrades.\n\nRuntimeError: expected scalar type Float but found Half\n\nTry set fp16: true\n\nNotImplementedError: No operator found for memory_efficient_attention_forward …\n\nTry to turn off xformers.\n\naccelerate config missing\n\nIt’s safe to ignore it.\n\nNCCL Timeouts during training\n\nSee the NCCL guide.\n\n\nFor many formats, Axolotl constructs prompts by concatenating token ids after tokenizing strings. The reason for concatenating token ids rather than operating on strings is to maintain precise accounting for attention masks.\nIf you decode a prompt constructed by axolotl, you might see spaces between tokens (or lack thereof) that you do not expect, especially around delimiters and special tokens. When you are starting out with a new format, you should always do the following:\n\nMaterialize some data using python -m axolotl.cli.preprocess your_config.yml --debug, and then decode the first few rows with your model’s tokenizer.\nDuring inference, right before you pass a tensor of token ids to your model, decode these tokens back into a string.\nMake sure the inference string from #2 looks exactly like the data you fine tuned on from #1, including spaces and new lines. If they aren’t the same, adjust your inference server accordingly.\nAs an additional troubleshooting step, you can look at the token ids between 1 and 2 to make sure they are identical.\n\nHaving misalignment between your prompts during training and inference can cause models to perform very poorly, so it is worth checking this. See this blog post for a concrete example.\n\n\n\n\nSee this debugging guide for tips on debugging Axolotl, along with an example configuration for debugging with VSCode.\n\n\n\nJoin our Discord server where we our community members can help you.\nNeed dedicated support? Please contact us at ✉️wing@openaccessaicollective.org for dedicated support options.\n\n\n\nBuilding something cool with Axolotl? Consider adding a badge to your model card.\n[<img src=\"https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png\" alt=\"Built with Axolotl\" width=\"200\" height=\"32\"/>](https://github.com/OpenAccess-AI-Collective/axolotl)\n\n\n\n\nCheck out some of the projects and models that have been built using Axolotl! Have a model you’d like to add to our Community Showcase? Open a PR with your model.\nOpen Access AI Collective - Minotaur 13b - Manticore 13b - Hippogriff 30b\nPocketDoc Labs - Dan’s PersonalityEngine 13b LoRA\n\n\n\nPlease read the contributing guide\nBugs? Please check the open issues else create a new Issue.\nPRs are greatly welcome!\nPlease run below to setup env\ngit clone https://github.com/OpenAccess-AI-Collective/axolotl\ncd axolotl\n\npip3 install packaging\npip3 install -e '.[flash-attn,deepspeed]'\n\npip3 install -r requirements-dev.txt -r requirements-tests.txt\npre-commit install\n\n# test\npytest tests/\n\n# optional: run against all files\npre-commit run --all-files\nThanks to all of our contributors to date. Help drive open source AI progress forward by contributing to Axolotl.\n  \n\n\n\nOpenAccess AI Collective is run by volunteer contributors such as winglian, NanoCode012, tmm1, mhenrichsen, casper-hansen, hamelsmu and many more who help us accelerate forward by fixing bugs, answering community questions and implementing new features. Axolotl needs donations from sponsors for the compute needed to run our unit & integration tests, troubleshooting community issues, and providing bounties. If you love axolotl, consider sponsoring the project via GitHub Sponsors, Ko-fi or reach out directly to wing@openaccessaicollective.org.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nJarvisLabs.ai",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#axolotl-supports",
    "href": "index.html#axolotl-supports",
    "title": "Axolotl",
    "section": "",
    "text": "fp16/fp32\nlora\nqlora\ngptq\ngptq w/flash attn\nflash attn\nxformers attn\n\n\n\n\nllama\n✅\n✅\n✅\n✅\n✅\n✅\n✅\n\n\nMistral\n✅\n✅\n✅\n✅\n✅\n✅\n✅\n\n\nMixtral-MoE\n✅\n✅\n✅\n❓\n❓\n❓\n❓\n\n\nPythia\n✅\n✅\n✅\n❌\n❌\n❌\n❓\n\n\ncerebras\n✅\n✅\n✅\n❌\n❌\n❌\n❓\n\n\nbtlm\n✅\n✅\n✅\n❌\n❌\n❌\n❓\n\n\nmpt\n✅\n❌\n❓\n❌\n❌\n❌\n❓\n\n\nfalcon\n✅\n✅\n✅\n❌\n❌\n❌\n❓\n\n\ngpt-j\n✅\n✅\n✅\n❌\n❌\n❓\n❓\n\n\nXGen\n✅\n❓\n✅\n❓\n❓\n❓\n✅\n\n\nphi\n✅\n✅\n✅\n❓\n❓\n❓\n❓\n\n\nRWKV\n✅\n❓\n❓\n❓\n❓\n❓\n❓\n\n\nQwen\n✅\n✅\n✅\n❓\n❓\n❓\n❓\n\n\nGemma\n✅\n✅\n✅\n❓\n❓\n✅\n❓\n\n\n\n✅: supported ❌: not supported ❓: untested",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#quickstart",
    "href": "index.html#quickstart",
    "title": "Axolotl",
    "section": "",
    "text": "Get started with Axolotl in just a few steps! This quickstart guide will walk you through setting up and running a basic fine-tuning task.\nRequirements: Python >=3.10 and Pytorch >=2.1.1.\ngit clone https://github.com/OpenAccess-AI-Collective/axolotl\ncd axolotl\n\npip3 install packaging\npip3 install -e '.[flash-attn,deepspeed]'\n\n\n# preprocess datasets - optional but recommended\nCUDA_VISIBLE_DEVICES=\"\" python -m axolotl.cli.preprocess examples/openllama-3b/lora.yml\n\n# finetune lora\naccelerate launch -m axolotl.cli.train examples/openllama-3b/lora.yml\n\n# inference\naccelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \\\n    --lora_model_dir=\"./lora-out\"\n\n# gradio\naccelerate launch -m axolotl.cli.inference examples/openllama-3b/lora.yml \\\n    --lora_model_dir=\"./lora-out\" --gradio\n\n# remote yaml files - the yaml config can be hosted on a public URL\n# Note: the yaml config must directly link to the **raw** yaml\naccelerate launch -m axolotl.cli.train https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/examples/openllama-3b/lora.yml",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#advanced-setup",
    "href": "index.html#advanced-setup",
    "title": "Axolotl",
    "section": "",
    "text": "docker run --gpus '\"all\"' --rm -it winglian/axolotl:main-latest\nOr run on the current files for development:\ndocker compose up -d\n\n[!Tip] If you want to debug axolotl or prefer to use Docker as your development environment, see the debugging guide’s section on Docker.\n\n\n\nDocker advanced\n\nA more powerful Docker command to run would be this:\ndocker run --privileged --gpus '\"all\"' --shm-size 10g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src=\"${PWD}\",target=/workspace/axolotl -v ${HOME}/.cache/huggingface:/root/.cache/huggingface winglian/axolotl:main-latest\nIt additionally: * Prevents memory issues when running e.g. deepspeed (e.g. you could hit SIGBUS/signal 7 error) through --ipc and --ulimit args. * Persists the downloaded HF data (models etc.) and your modifications to axolotl code through --mount/-v args. * The --name argument simply makes it easier to refer to the container in vscode (Dev Containers: Attach to Running Container...) or in your terminal. * The --privileged flag gives all capabilities to the container. * The --shm-size 10g argument increases the shared memory size. Use this if you see exitcode: -7 errors using deepspeed.\nMore information on nvidia website\n\n\n\n\n\nInstall python >=3.10\nInstall pytorch stable https://pytorch.org/get-started/locally/\nInstall Axolotl along with python dependencies bash         pip3 install packaging         pip3 install -e '.[flash-attn,deepspeed]'\n(Optional) Login to Huggingface to use gated models/datasets. bash         huggingface-cli login Get the token at huggingface.co/settings/tokens\n\n\n\n\nFor cloud GPU providers that support docker images, use winglian/axolotl-cloud:main-latest\n\non Latitude.sh use this direct link\non JarvisLabs.ai use this direct link\non RunPod use this direct link\n\n\n\n\n\n\n\n\nClick to Expand\n\n\nInstall python\n\nsudo apt update\nsudo apt install -y python3.10\n\nsudo update-alternatives --install /usr/bin/python python /usr/bin/python3.10 1\nsudo update-alternatives --config python # pick 3.10 if given option\npython -V # should be 3.10\n\nInstall pip\n\nwget https://bootstrap.pypa.io/get-pip.py\npython get-pip.py\n\nInstall torch\n\npip3 install -U torch --index-url https://download.pytorch.org/whl/cu118\n\nAxolotl\n\ngit clone https://github.com/OpenAccess-AI-Collective/axolotl\ncd axolotl\n\npip3 install packaging\npip3 install -e '.[flash-attn,deepspeed]'\npip3 install protobuf==3.20.3\npip3 install -U --ignore-installed requests Pillow psutil scipy\n\nSet path\n\nexport LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH\n\n\n\n\n\n\nClick to Expand\n\nUse a Deeplearning linux OS with cuda and pytorch installed. Then follow instructions on quickstart.\nMake sure to run the below to uninstall xla.\npip uninstall -y torch_xla[tpu]\n\n\n\n\n\nPlease use WSL or Docker!\n\n\n\nUse the below instead of the install method in QuickStart.\npip3 install -e '.'\nMore info: mac.md\n\n\n\nTo launch on GPU instances (both on-demand and spot instances) on 7+ clouds (GCP, AWS, Azure, OCI, and more), you can use SkyPilot:\npip install \"skypilot-nightly[gcp,aws,azure,oci,lambda,kubernetes,ibm,scp]\"  # choose your clouds\nsky check\nGet the example YAMLs of using Axolotl to finetune mistralai/Mistral-7B-v0.1:\ngit clone https://github.com/skypilot-org/skypilot.git\ncd skypilot/llm/axolotl\nUse one command to launch:\n# On-demand\nHF_TOKEN=xx sky launch axolotl.yaml --env HF_TOKEN\n\n# Managed spot (auto-recovery on preemption)\nHF_TOKEN=xx BUCKET=<unique-name> sky spot launch axolotl-spot.yaml --env HF_TOKEN --env BUCKET\n\n\n\n\nAxolotl supports a variety of dataset formats. Below are some of the formats you can use. Have dataset(s) in one of the following format (JSONL recommended):\n\n\n\ncompletion: raw corpus\n{\"text\": \"...\"}\n\nNote: Axolotl usually loads the entire dataset into memory. This will be challenging for large datasets. Use the following config to enable streaming:\npretraining_dataset: # hf path only\n\n\n\n\n\n\nalpaca: instruction; input(optional)\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\"}\n\n\n\nSee other formats\n\n\njeopardy: question and answer\n{\"question\": \"...\", \"category\": \"...\", \"answer\": \"...\"}\noasst: instruction\n{\"INSTRUCTION\": \"...\", \"RESPONSE\": \"...\"}\ngpteacher: instruction; input(optional)\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}\nreflection: instruction with reflect; input(optional)\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\", \"reflection\": \"...\", \"corrected\": \"...\"}\nexplainchoice: question, choices, (solution OR explanation)\n{\"question\": \"...\", \"choices\": [\"...\"], \"solution\": \"...\", \"explanation\": \"...\"}\nconcisechoice: question, choices, (solution OR explanation)\n{\"question\": \"...\", \"choices\": [\"...\"], \"solution\": \"...\", \"explanation\": \"...\"}\nsummarizetldr: article and summary\n{\"article\": \"...\", \"summary\": \"...\"}\nalpaca_chat: basic instruct for alpaca chat\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}\nalpaca_chat.load_qa: question and answer for alpaca chat\n{\"question\": \"...\", \"answer\": \"...\"}\nalpaca_chat.load_concise: question and answer for alpaca chat, for concise answers\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}\nalpaca_chat.load_camel_ai: question and answer for alpaca chat, for load_camel_ai\n{\"message_1\": \"...\", \"message_2\": \"...\"}\nalpaca_w_system.load_open_orca: support for open orca datasets with included system prompts, instruct\n{\"system_prompt\": \"...\", \"question\": \"...\", \"response\": \"...\"}\ncontext_qa: in context question answering from an article\n{\"article\": \"...\", \"question\": \"...\", \"answer\": \"...\"}\ncontext_qa.load_v2: in context question answering (alternate)\n{\"context\": \"...\", \"question\": \"...\", \"answer\": \"...\"}\ncontext_qa.load_404: in context question answering from an article, with default response for no answer from context\n{\"article\": \"...\", \"unanswerable_question\": \"...\"}\ncreative_acr.load_answer: instruction and revision\n{\"instruction\": \"...\", \"revision\": \"...\"}\ncreative_acr.load_critique: critique\n{\"scores\": \"...\", \"critiques\": \"...\", \"instruction\": \"...\", \"answer\": \"...\"}\ncreative_acr.load_revise: critique and revise\n{\"scores\": \"...\", \"critiques\": \"...\", \"instruction\": \"...\", \"answer\": \"...\", \"revision\": \"...\"}\nmetharme: instruction, adds additional eos tokens\n{\"prompt\": \"...\", \"generation\": \"...\"}\n\n\n\n\n\n\ninput_output: template-free prompt construction\n {\"segments\": [{\"label\": true|false, \"text\": \"...\"}]}\n\nThis is a special format that allows you to construct prompts without using templates. This is for advanced users who want more freedom with prompt construction. See these docs for more details.\n\n\n\n\nsharegpt: conversations where from is human/gpt. (optional: first row with role system to override default system prompt)\n{\"conversations\": [{\"from\": \"...\", \"value\": \"...\"}]}\n\n\n\nSee other formats\n\n\npygmalion: pygmalion\n{\"conversations\": [{\"role\": \"...\", \"value\": \"...\"}]}\nsharegpt.load_role: conversations where role is used instead of from\n{\"conversations\": [{\"role\": \"...\", \"value\": \"...\"}]}\nsharegpt.load_guanaco: conversations where from is prompter/assistant instead of default sharegpt\n{\"conversations\": [{\"from\": \"...\", \"value\": \"...\"}]}\nsharegpt_jokes: creates a chat where bot is asked to tell a joke, then explain why the joke is funny\n{\"conversations\": [{\"title\": \"...\", \"text\": \"...\", \"explanation\": \"...\"}]}\n\n\nNote: type: sharegpt opens a special config conversation: that enables conversions to many Conversation types. See dataset section under all yaml options.\n\n\n\n\nFor a dataset that is preprocessed for instruction purposes:\n{\"input\": \"...\", \"output\": \"...\"}\nYou can use this example in your YAML config:\ndatasets:\n  - path: repo\n    type:\n      system_prompt: \"\"\n      field_system: system\n      field_instruction: input\n      field_output: output\n      format: \"[INST] {instruction} [/INST]\"\n      no_input_format: \"[INST] {instruction} [/INST]\"\nSee full config options under all yaml options.\n\n\n\n\nDo not pass a type:\nColumns in Dataset must be exactly input_ids, attention_mask, labels\n\n- path: ...\n\n\n\n\nSee examples for quick start. It is recommended to duplicate and modify to your needs. The most important options are:\n\nmodel\nbase_model: ./llama-7b-hf # local or huggingface repo\nNote: The code will load the right architecture.\ndataset\ndatasets:\n    # huggingface repo\n  - path: vicgalle/alpaca-gpt4\n    type: alpaca\n\n    # huggingface repo with specific configuration/subset\n  - path: EleutherAI/pile\n    name: enron_emails\n    type: completion # format from earlier\n    field: text # Optional[str] default: text, field to use for completion data\n\n    # huggingface repo with multiple named configurations/subsets\n  - path: bigcode/commitpackft\n    name:\n      - ruby\n      - python\n      - typescript\n    type: ... # unimplemented custom format\n\n    # fastchat conversation\n    # See 'conversation' options: https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py\n  - path: ...\n    type: sharegpt\n    conversation: chatml # default: vicuna_v1.1\n\n    # local\n  - path: data.jsonl # or json\n    ds_type: json # see other options below\n    type: alpaca\n\n    # dataset with splits, but no train split\n  - path: knowrohit07/know_sql\n    type: context_qa.load_v2\n    train_on_split: validation\n\n    # loading from s3 or gcs\n    # s3 creds will be loaded from the system default and gcs only supports public access\n  - path: s3://path_to_ds # Accepts folder with arrow/parquet or file path like above. Supports s3, gcs.\n    ...\n\n    # Loading Data From a Public URL\n    # - The file format is `json` (which includes `jsonl`) by default. For different formats, adjust the `ds_type` option accordingly.\n  - path: https://some.url.com/yourdata.jsonl # The URL should be a direct link to the file you wish to load. URLs must use HTTPS protocol, not HTTP.\n    ds_type: json # this is the default, see other options below.\nloading\nload_in_4bit: true\nload_in_8bit: true\n\nbf16: auto # require >=ampere, auto will detect if your GPU supports this and choose automatically.\nfp16: # leave empty to use fp16 when bf16 is 'auto'. set to false if you want to fallback to fp32\ntf32: true # require >=ampere\n\nbfloat16: true # require >=ampere, use instead of bf16 when you don't want AMP (automatic mixed precision)\nfloat16: true # use instead of fp16 when you don't want AMP\nNote: Repo does not do 4-bit quantization.\nlora\nadapter: lora # 'qlora' or leave blank for full finetune\nlora_r: 8\nlora_alpha: 16\nlora_dropout: 0.05\nlora_target_modules:\n  - q_proj\n  - v_proj\n\n\n\nAll yaml options (click to expand)\n\n# This is the huggingface model that contains *.pt, *.safetensors, or *.bin files\n# This can also be a relative path to a model on disk\nbase_model: ./llama-7b-hf\n# You can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)\nbase_model_ignore_patterns:\n# If the base_model repo on hf hub doesn't include configuration .json files,\n# You can set that here, or leave this empty to default to base_model\nbase_model_config: ./llama-7b-hf\n# You can specify to choose a specific model revision from huggingface hub\nrevision_of_model:\n# Optional tokenizer configuration path in case you want to use a different tokenizer\n# than the one defined in the base model\ntokenizer_config:\n# If you want to specify the type of model to load, AutoModelForCausalLM is a good choice too\nmodel_type: AutoModelForCausalLM\n# Corresponding tokenizer for the model AutoTokenizer is a good choice\ntokenizer_type: AutoTokenizer\n# Trust remote code for untrusted source\ntrust_remote_code:\n# use_fast option for tokenizer loading from_pretrained, default to True\ntokenizer_use_fast:\n# Whether to use the legacy tokenizer setting, defaults to True\ntokenizer_legacy:\n# Resize the model embeddings when new tokens are added to multiples of 32\n# This is reported to improve training speed on some models\nresize_token_embeddings_to_32x:\n\n# (Internal use only)\n# Used to identify which the model is based on\nis_falcon_derived_model:\nis_llama_derived_model:\nis_qwen_derived_model:\n# Please note that if you set this to true, `padding_side` will be set to \"left\" by default\nis_mistral_derived_model:\n\n# optional overrides to the base model configuration\noverrides_of_model_config:\n  # RoPE Scaling https://github.com/huggingface/transformers/pull/24653\n  rope_scaling:\n    type: # linear | dynamic\n    factor: # float\n\n# optional overrides to the bnb 4bit quantization configuration\n# https://huggingface.co/docs/transformers/main/main_classes/quantization#transformers.BitsAndBytesConfig\nbnb_config_kwargs:\n  # These are default values\n  llm_int8_has_fp16_weight: false\n  bnb_4bit_quant_type: nf4\n  bnb_4bit_use_double_quant: true\n\n\n# Whether you are training a 4-bit GPTQ quantized model\ngptq: true\n\n# This will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer\nload_in_8bit: true\n# Use bitsandbytes 4 bit\nload_in_4bit:\n\n# Use CUDA bf16\nbf16: true # bool or 'full' for `bf16_full_eval`. require >=ampere\n# Use CUDA fp16\nfp16: true\n# Use CUDA tf32\ntf32: true # require >=ampere\n\n# No AMP (automatic mixed precision)\nbfloat16: true # require >=ampere\nfloat16: true\n\n# Limit the memory for all available GPUs to this amount (if an integer, expressed in gigabytes); default: unset\ngpu_memory_limit: 20GiB\n# Do the LoRA/PEFT loading on CPU -- this is required if the base model is so large it takes up most or all of the available GPU VRAM, e.g. during a model and LoRA merge\nlora_on_cpu: true\n\n# A list of one or more datasets to finetune the model with\ndatasets:\n  # HuggingFace dataset repo | s3://,gs:// path | \"json\" for local dataset, make sure to fill data_files\n  - path: vicgalle/alpaca-gpt4\n  # The type of prompt to use for training. [alpaca, sharegpt, gpteacher, oasst, reflection]\n    type: alpaca # format | format:<prompt_style> (chat/instruct) | <prompt_strategies>.load_<load_fn>\n    ds_type: # Optional[str] (json|arrow|parquet|text|csv) defines the datatype when path is a file\n    data_files: # Optional[str] path to source data files\n    shards: # Optional[int] number of shards to split data into\n    name: # Optional[str] name of dataset configuration to load\n    train_on_split: train # Optional[str] name of dataset split to load from\n\n    # Optional[str] fastchat conversation type, only used with type: sharegpt\n    conversation: # Options (see Conversation 'name'): https://github.com/lm-sys/FastChat/blob/main/fastchat/conversation.py\n    field_human: # Optional[str]. Human key to use for conversation.\n    field_model: # Optional[str]. Assistant key to use for conversation.\n    # Add additional keys from your dataset as input or output roles\n    roles:\n      input: # Optional[List[str]]. These will be masked based on train_on_input\n      output: # Optional[List[str]].\n\n  # Custom user instruction prompt\n  - path: repo\n    type:\n      # The below are defaults. only set what's needed if you use a different column name.\n      system_prompt: \"\"\n      system_format: \"{system}\"\n      field_system: system\n      field_instruction: instruction\n      field_input: input\n      field_output: output\n\n      # Customizable to be single line or multi-line\n      # Use {instruction}/{input} as key to be replaced\n      # 'format' can include {input}\n      format: |-\n        User: {instruction} {input}\n        Assistant:\n      # 'no_input_format' cannot include {input}\n      no_input_format: \"{instruction} \"\n\n      # For `completion` datsets only, uses the provided field instead of `text` column\n      field:\n\n# If false, the datasets will not be shuffled and will keep their original order in `datasets`.\n# The same applies to the `test_datasets` option and the `pretraining_dataset` option. Default is true.\nshuffle_merged_datasets: true\n\n# A list of one or more datasets to eval the model with.\n# You can use either test_datasets, or val_set_size, but not both.\ntest_datasets:\n  - path: /workspace/data/eval.jsonl\n    ds_type: json\n    # You need to specify a split. For \"json\" datasets the default split is called \"train\".\n    split: train\n    type: completion\n    data_files:\n      - /workspace/data/eval.jsonl\n\n# use RL training: 'dpo', 'ipo', 'kto_pair'\nrl:\n\n# Saves the desired chat template to the tokenizer_config.json for easier inferencing\n# Currently supports chatml and inst (mistral/mixtral)\nchat_template: chatml\n# Changes the default system message\ndefault_system_message: You are a helpful assistant. Please give a long and detailed answer. # Currently only supports chatml.\n# Axolotl attempts to save the dataset as an arrow after packing the data together so\n# subsequent training attempts load faster, relative path\ndataset_prepared_path: data/last_run_prepared\n# Push prepared dataset to hub\npush_dataset_to_hub: # repo path\n# The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()`\n# if not set.\ndataset_processes: # defaults to os.cpu_count() if not set\n# Keep dataset in memory while preprocessing\n# Only needed if cached dataset is taking too much storage\ndataset_keep_in_memory:\n# push checkpoints to hub\nhub_model_id: # private repo path to push finetuned model\n# how to push checkpoints to hub\n# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy\nhub_strategy:\n# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets\n# Required to be true when used in combination with `push_dataset_to_hub`\nhf_use_auth_token: # boolean\n# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval.\nval_set_size: 0.04\n# Num shards for whole dataset\ndataset_shard_num:\n# Index of shard to use for whole dataset\ndataset_shard_idx:\n\n# The maximum length of an input to train with, this should typically be less than 2048\n# as most models have a token/context limit of 2048\nsequence_len: 2048\n# Pad inputs so each step uses constant sized buffers\n# This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently\npad_to_sequence_len:\n# Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true'\nsample_packing:\n# Set to 'false' if getting errors during eval with sample_packing on.\neval_sample_packing:\n# You can set these packing optimizations AFTER starting a training at least once.\n# The trainer will provide recommended values for these values.\nsample_packing_eff_est:\ntotal_num_tokens:\n\n# Passed through to transformers when loading the model when launched without accelerate\n# Use `sequential` when training w/ model parallelism to limit memory\ndevice_map:\n# Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model.\nmax_memory:\n\n# If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model\nadapter: lora\n# If you already have a lora model trained that you want to load, put that here.\n# This means after training, if you want to test the model, you should set this to the value of `output_dir`.\n# Note that if you merge an adapter to the base model, a new subdirectory `merged` will be created under the `output_dir`.\nlora_model_dir:\n\n# LoRA hyperparameters\n# For more details about the following options, see:\n# https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2\nlora_r: 8\nlora_alpha: 16\nlora_dropout: 0.05\nlora_target_modules:\n  - q_proj\n  - v_proj\n#  - k_proj\n#  - o_proj\n#  - gate_proj\n#  - down_proj\n#  - up_proj\nlora_target_linear: # If true, will target all linear modules\npeft_layers_to_transform: # The layer indices to transform, otherwise, apply to all layers\n\n# If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens.\n# For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models.\n# `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities.\n# https://github.com/huggingface/peft/issues/334#issuecomment-1561727994\nlora_modules_to_save:\n#  - embed_tokens\n#  - lm_head\n\nlora_fan_in_fan_out: false\n\npeft:\n  # Configuration options for loftq initialization for LoRA\n  # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization\n  loftq_config:\n    loftq_bits:  # typically 4 bits\n\n# ReLoRA configuration\n# Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed\nrelora_steps: # Number of steps per ReLoRA restart\nrelora_warmup_steps: # Number of per-restart warmup steps\nrelora_anneal_steps: # Number of anneal steps for each relora cycle\nrelora_prune_ratio: # threshold for optimizer magnitude when pruning\nrelora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings\n\n# wandb configuration if you're using it\n# Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`.\nwandb_mode: # \"offline\" to save run metadata locally and not sync to the server, \"disabled\" to turn off wandb\nwandb_project: # Your wandb project name\nwandb_entity: # A wandb Team name if using a Team\nwandb_watch:\nwandb_name: # Set the name of your wandb run\nwandb_run_id: # Set the ID of your wandb run\nwandb_log_model: # \"checkpoint\" to log model to wandb Artifacts every `save_steps` or \"end\" to log only at the end of training\n\n# mlflow configuration if you're using it\nmlflow_tracking_uri: # URI to mlflow\nmlflow_experiment_name: # Your experiment name\nhf_mlflow_log_artifacts:  # set to true to copy each saved checkpoint on each save to mlflow artifact registry\n\n# Where to save the full-finetuned model to\noutput_dir: ./completed-model\n\n# Whether to use torch.compile and which backend to use\ntorch_compile:  # bool\ntorch_compile_backend:  # Optional[str]\n\n# Training hyperparameters\n\n# If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps.\ngradient_accumulation_steps: 1\n# The number of samples to include in each batch. This is the number of samples sent to each GPU.\nmicro_batch_size: 2\neval_batch_size:\nnum_epochs: 4\nwarmup_steps: 100  # cannot use with warmup_ratio\nwarmup_ratio: 0.05  # cannot use with warmup_steps\nlearning_rate: 0.00003\nlr_quadratic_warmup:\nlogging_steps:\neval_steps: # Leave empty to eval at each epoch, integers for every N steps. decimal for fraction of total steps\nevals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps\nsave_strategy: # Set to `no` to skip checkpoint saves\nsave_steps: # Leave empty to save at each epoch\nsaves_per_epoch: # number of times per epoch to save a checkpoint, mutually exclusive with save_steps\nsave_total_limit: # Checkpoints saved at a time\n# Maximum number of iterations to train for. It precedes num_epochs which means that\n# if both are set, num_epochs will not be guaranteed.\n# e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps\nmax_steps:\n\neval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0\neval_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128\neval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is [\"sacrebleu\", \"comet\", \"ter\", chrf]\n\nloss_watchdog_threshold: # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training)\nloss_watchdog_patience: # Number of high-loss steps in a row before the trainer aborts (default: 3)\n\n# Save model as safetensors (require safetensors package)\nsave_safetensors:\n\n# Whether to mask out or include the human's prompt from the training labels\ntrain_on_inputs: false\n# Group similarly sized data to minimize padding.\n# May be slower to start, as it must download and sort the entire dataset.\n# Note that training loss may have an oscillating pattern with this enabled.\ngroup_by_length: false\n\n# Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing\ngradient_checkpointing: false\n# additional kwargs to pass to the trainer for gradient checkpointing\n# gradient_checkpointing_kwargs:\n#   use_reentrant: true\n\n# Stop training after this many evaluation losses have increased in a row\n# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback\nearly_stopping_patience: 3\n\n# Specify a scheduler and kwargs to use with the optimizer\nlr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine\nlr_scheduler_kwargs:\ncosine_min_lr_ratio: # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr\ncosine_constant_lr_ratio: # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step (https://arxiv.org/pdf/2308.04014.pdf)\n\n# For one_cycle optim\nlr_div_factor: # Learning rate div factor\n\n# Specify optimizer\n# Valid values are driven by the Transformers OptimizerNames class, see:\n# https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134\n#\n# Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of\n# torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used\n# in the examples/ for your model and fine-tuning use case.\n#\n# Valid values for 'optimizer' include:\n# - adamw_hf\n# - adamw_torch\n# - adamw_torch_fused\n# - adamw_torch_xla\n# - adamw_apex_fused\n# - adafactor\n# - adamw_anyprecision\n# - sgd\n# - adagrad\n# - adamw_bnb_8bit\n# - lion_8bit\n# - lion_32bit\n# - paged_adamw_32bit\n# - paged_adamw_8bit\n# - paged_lion_32bit\n# - paged_lion_8bit\n# - galore_adamw\n# - galore_adamw_8bit\n# - galore_adafactor\n# - galore_adamw_layerwise\n# - galore_adamw_8bit_layerwise\n# - galore_adafactor_layerwise\noptimizer:\n# Dictionary of arguments to pass to the optimizer\noptim_args:\n# For Galore Optimizers the following optim_args are available\n# rank:  # type: int\n# update_proj_gap  # type: int\n# scale  # type: float\n# proj_type:  # type: str, default = std\n\n# The target modules to optimize, i.e. the module names that you would like to train, right now this is used only for GaLore algorithm\noptim_target_modules:\n# - self_attn  # for llama\n# - mlp\n\n# Specify weight decay\nweight_decay:\n# adamw hyperparams\nadam_beta1:\nadam_beta2:\nadam_epsilon:\n# Gradient clipping max norm\nmax_grad_norm:\n\n# Augmentation techniques\n# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings\n# currently only supported on Llama and Mistral\nneftune_noise_alpha:\n\n# Whether to bettertransformers\nflash_optimum:\n# Whether to use xformers attention patch https://github.com/facebookresearch/xformers:\nxformers_attention:\n# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:\nflash_attention:\nflash_attn_cross_entropy:  # Whether to use flash-attention cross entropy implementation - advanced use only\nflash_attn_rms_norm:  # Whether to use flash-attention rms norm implementation - advanced use only\nflash_attn_fuse_qkv: # Whether to fuse QKV into a single operation\nflash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation\n# Whether to use scaled-dot-product attention\n# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html\nsdp_attention:\n# Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf\ns2_attention:\n# Resume from a specific checkpoint dir\nresume_from_checkpoint:\n# If resume_from_checkpoint isn't set and you simply want it to start where it left off.\n# Be careful with this being turned on between different models.\nauto_resume_from_checkpoints: false\n\n# Don't mess with this, it's here for accelerate and torchrun\nlocal_rank:\n\n# Add or change special tokens.\n# If you add tokens here, you don't need to add them to the `tokens` list.\nspecial_tokens:\n  # bos_token: \"<s>\"\n  # eos_token: \"</s>\"\n  # unk_token: \"<unk>\"\n\n# Add extra tokens.\ntokens:\n\n# FSDP\nfsdp:\nfsdp_config:\n\n# Deepspeed config path. e.g., deepspeed_configs/zero3.json\ndeepspeed:\n\n# Advanced DDP Arguments\nddp_timeout:\nddp_bucket_cap_mb:\nddp_broadcast_buffers:\n\n# Path to torch distx for optim 'adamw_anyprecision'\ntorchdistx_path:\n\n# Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize\npretraining_dataset:\n\n# Debug mode\ndebug:\n\n# Seed\nseed:\n\n# Allow overwrite yml config using from cli\nstrict:\n\n\n\nUnderstanding of batch size and gradient accumulation steps\n\n Gradient accumulation means accumulating gradients over several mini-batches and updating the model weights afterward. When the samples in each batch are diverse, this technique doesn’t significantly impact learning.\nThis method allows for effective training with larger effective batch sizes without needing proportionally larger memory. Here’s why:\n\nMemory Consumption with Batch Size: The primary reason increasing the batch size impacts memory is due to the storage requirements for intermediate activations. When you forward propagate a batch through a network, you have to store the activations at each layer for each sample in the batch, because these activations are used during backpropagation to compute gradients. Therefore, larger batches mean more activations, leading to greater GPU memory consumption.\nGradient Accumulation: With gradient accumulation, you’re effectively simulating a larger batch size by accumulating gradients over several smaller batches (or micro-batches). However, at any given time, you’re only forward and backward propagating a micro-batch. This means you only store activations for the micro-batch, not the full accumulated batch. As a result, you can simulate the effect of a larger batch size without the memory cost of storing activations for a large batch.\n\nExample 1: Micro batch size: 3 Gradient accumulation steps: 2 Number of GPUs: 3 Total batch size = 3 * 2 * 3 = 18\n| GPU 1          | GPU 2          | GPU 3          |\n|----------------|----------------|----------------|\n| S1, S2, S3     | S4, S5, S6     | S7, S8, S9     |\n| e1, e2, e3     | e4, e5, e6     | e7, e8, e9     |\n|----------------|----------------|----------------|\n| → (accumulate) | → (accumulate) | → (accumulate) |\n|----------------|----------------|----------------|\n| S10, S11, S12  | S13, S14, S15  | S16, S17, S18  |\n| e10, e11, e12  | e13, e14, e15  | e16, e17, e18  |\n|----------------|----------------|----------------|\n| → (apply)      | → (apply)      | → (apply)      |\n\nAccumulated gradient for the weight w1 after the second iteration (considering all GPUs):\nTotal gradient for w1 = e1 + e2 + e3 + e4 + e5 + e6 + e7 + e8 + e9 + e10 + e11 + e12 + e13 + e14 + e15 + e16 + e17 + e18\n\nWeight update for w1:\nw1_new = w1_old - learning rate x (Total gradient for w1 / 18)\nExample 2: Micro batch size: 2 Gradient accumulation steps: 1 Number of GPUs: 3 Total batch size = 2 * 1 * 3 = 6\n| GPU 1     | GPU 2     | GPU 3     |\n|-----------|-----------|-----------|\n| S1, S2    | S3, S4    | S5, S6    |\n| e1, e2    | e3, e4    | e5, e6    |\n|-----------|-----------|-----------|\n| → (apply) | → (apply) | → (apply) |\n\nAccumulated gradient for the weight w1 (considering all GPUs):\nTotal gradient for w1 = e1 + e2 + e3 + e4 + e5 + e6\n\nWeight update for w1:\nw1_new = w1_old - learning rate × (Total gradient for w1 / 6)\n\n\n\n\nRun\naccelerate launch -m axolotl.cli.train your_config.yml\n\n[!TIP] You can also reference a config file that is hosted on a public URL, for example accelerate launch -m axolotl.cli.train https://yourdomain.com/your_config.yml\n\n\n\nYou can optionally pre-tokenize dataset with the following before finetuning. This is recommended for large datasets.\n\nSet dataset_prepared_path: to a local folder for saving and loading pre-tokenized dataset.\n(Optional): Set push_dataset_to_hub: hf_user/repo to push it to Huggingface.\n(Optional): Use --debug to see preprocessed examples.\n\npython -m axolotl.cli.preprocess your_config.yml\n\n\n\nBelow are the options available in axolotl for training with multiple GPUs. Note that DeepSpeed is the recommended multi-GPU option currently because FSDP may experience loss instability.\n\n\nDeepspeed is an optimization suite for multi-gpu systems allowing you to train much larger models than you might typically be able to fit into your GPU’s VRAM. More information about the various optimization types for deepspeed is available at https://huggingface.co/docs/accelerate/main/en/usage_guides/deepspeed#what-is-integrated\nWe provide several default deepspeed JSON configurations for ZeRO stage 1, 2, and 3.\ndeepspeed: deepspeed_configs/zero1.json\naccelerate launch -m axolotl.cli.train examples/llama-2/config.py --deepspeed deepspeed_configs/zero1.json\n\n\n\n\nllama FSDP\n\nfsdp:\n  - full_shard\n  - auto_wrap\nfsdp_config:\n  fsdp_offload_params: true\n  fsdp_state_dict_type: FULL_STATE_DICT\n  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer\n\n\n\nAxolotl supports training with FSDP and QLoRA, see these docs for more information.\n\n\n\nMake sure your WANDB_API_KEY environment variable is set (recommended) or you login to wandb with wandb login.\n\nwandb options\n\nwandb_mode:\nwandb_project:\nwandb_entity:\nwandb_watch:\nwandb_name:\nwandb_log_model:\n\n\n\nIt is important to have special tokens like delimiters, end-of-sequence, beginning-of-sequence in your tokenizer’s vocabulary. This will help you avoid tokenization issues and help your model train better. You can do this in axolotl like this:\nspecial_tokens:\n  bos_token: \"<s>\"\n  eos_token: \"</s>\"\n  unk_token: \"<unk>\"\ntokens: # these are delimiters\n  - \"<|im_start|>\"\n  - \"<|im_end|>\"\nWhen you include these tokens in your axolotl config, axolotl adds these tokens to the tokenizer’s vocabulary.\n\n\n\n\n\nAxolotl allows you to load your model in an interactive terminal playground for quick experimentation. The config file is the same config file used for training.\nPass the appropriate flag to the inference command, depending upon what kind of model was trained:\n\nPretrained LORA:\npython -m axolotl.cli.inference examples/your_config.yml --lora_model_dir=\"./lora-output-dir\"\nFull weights finetune:\npython -m axolotl.cli.inference examples/your_config.yml --base_model=\"./completed-model\"\nFull weights finetune w/ a prompt from a text file:\ncat /tmp/prompt.txt | python -m axolotl.cli.inference examples/your_config.yml \\\n  --base_model=\"./completed-model\" --prompter=None --load_in_8bit=True\n– With gradio hosting\npython -m axolotl.cli.inference examples/your_config.yml --gradio\n\nPlease use --sample_packing False if you have it on and receive the error similar to below:\n\nRuntimeError: stack expects each tensor to be equal size, but got [1, 32, 1, 128] at entry 0 and [1, 32, 8, 128] at entry 1\n\n\n\n\nThe following command will merge your LORA adapater with your base model. You can optionally pass the argument --lora_model_dir to specify the directory where your LORA adapter was saved, otherwhise, this will be inferred from output_dir in your axolotl config file. The merged model is saved in the sub-directory {lora_model_dir}/merged.\npython3 -m axolotl.cli.merge_lora your_config.yml --lora_model_dir=\"./completed-model\"\nYou may need to use the gpu_memory_limit and/or lora_on_cpu config options to avoid running out of memory. If you still run out of CUDA memory, you can try to merge in system RAM with\nCUDA_VISIBLE_DEVICES=\"\" python3 -m axolotl.cli.merge_lora ...\nalthough this will be very slow, and using the config options above are recommended instead.",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#common-errors",
    "href": "index.html#common-errors",
    "title": "Axolotl",
    "section": "",
    "text": "See also the FAQ’s and debugging guide.\n\nIf you encounter a ‘Cuda out of memory’ error, it means your GPU ran out of memory during the training process. Here’s how to resolve it:\n\nPlease reduce any below - micro_batch_size - eval_batch_size - gradient_accumulation_steps - sequence_len\nIf it does not help, try running without deepspeed and without accelerate (replace “accelerate launch” with “python”) in the command.\nUsing adamw_bnb_8bit might also save you some memory.\n\nfailed (exitcode: -9)\n\nUsually means your system has run out of system memory. Similarly, you should consider reducing the same settings as when you run out of VRAM. Additionally, look into upgrading your system RAM which should be simpler than GPU upgrades.\n\nRuntimeError: expected scalar type Float but found Half\n\nTry set fp16: true\n\nNotImplementedError: No operator found for memory_efficient_attention_forward …\n\nTry to turn off xformers.\n\naccelerate config missing\n\nIt’s safe to ignore it.\n\nNCCL Timeouts during training\n\nSee the NCCL guide.\n\n\nFor many formats, Axolotl constructs prompts by concatenating token ids after tokenizing strings. The reason for concatenating token ids rather than operating on strings is to maintain precise accounting for attention masks.\nIf you decode a prompt constructed by axolotl, you might see spaces between tokens (or lack thereof) that you do not expect, especially around delimiters and special tokens. When you are starting out with a new format, you should always do the following:\n\nMaterialize some data using python -m axolotl.cli.preprocess your_config.yml --debug, and then decode the first few rows with your model’s tokenizer.\nDuring inference, right before you pass a tensor of token ids to your model, decode these tokens back into a string.\nMake sure the inference string from #2 looks exactly like the data you fine tuned on from #1, including spaces and new lines. If they aren’t the same, adjust your inference server accordingly.\nAs an additional troubleshooting step, you can look at the token ids between 1 and 2 to make sure they are identical.\n\nHaving misalignment between your prompts during training and inference can cause models to perform very poorly, so it is worth checking this. See this blog post for a concrete example.",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#debugging-axolotl",
    "href": "index.html#debugging-axolotl",
    "title": "Axolotl",
    "section": "",
    "text": "See this debugging guide for tips on debugging Axolotl, along with an example configuration for debugging with VSCode.",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#need-help",
    "href": "index.html#need-help",
    "title": "Axolotl",
    "section": "",
    "text": "Join our Discord server where we our community members can help you.\nNeed dedicated support? Please contact us at ✉️wing@openaccessaicollective.org for dedicated support options.",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#badge",
    "href": "index.html#badge",
    "title": "Axolotl",
    "section": "",
    "text": "Building something cool with Axolotl? Consider adding a badge to your model card.\n[<img src=\"https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png\" alt=\"Built with Axolotl\" width=\"200\" height=\"32\"/>](https://github.com/OpenAccess-AI-Collective/axolotl)",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#community-showcase",
    "href": "index.html#community-showcase",
    "title": "Axolotl",
    "section": "",
    "text": "Check out some of the projects and models that have been built using Axolotl! Have a model you’d like to add to our Community Showcase? Open a PR with your model.\nOpen Access AI Collective - Minotaur 13b - Manticore 13b - Hippogriff 30b\nPocketDoc Labs - Dan’s PersonalityEngine 13b LoRA",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#contributing",
    "href": "index.html#contributing",
    "title": "Axolotl",
    "section": "",
    "text": "Please read the contributing guide\nBugs? Please check the open issues else create a new Issue.\nPRs are greatly welcome!\nPlease run below to setup env\ngit clone https://github.com/OpenAccess-AI-Collective/axolotl\ncd axolotl\n\npip3 install packaging\npip3 install -e '.[flash-attn,deepspeed]'\n\npip3 install -r requirements-dev.txt -r requirements-tests.txt\npre-commit install\n\n# test\npytest tests/\n\n# optional: run against all files\npre-commit run --all-files\nThanks to all of our contributors to date. Help drive open source AI progress forward by contributing to Axolotl.",
    "crumbs": [
      "Home"
    ]
  },
  {
    "objectID": "index.html#sponsors",
    "href": "index.html#sponsors",
    "title": "Axolotl",
    "section": "",
    "text": "OpenAccess AI Collective is run by volunteer contributors such as winglian, NanoCode012, tmm1, mhenrichsen, casper-hansen, hamelsmu and many more who help us accelerate forward by fixing bugs, answering community questions and implementing new features. Axolotl needs donations from sponsors for the compute needed to run our unit & integration tests, troubleshooting community issues, and providing bounties. If you love axolotl, consider sponsoring the project via GitHub Sponsors, Ko-fi or reach out directly to wing@openaccessaicollective.org.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\nJarvisLabs.ai",
    "crumbs": [
      "Home"
    ]
  }
]