add output for train loss in assertian err

.github/workflows/base.yml

@@ -46,18 +46,6 @@ jobs:
             python_version: "3.11"
             pytorch: 2.6.0
             torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-          - cuda: "126"
-            cuda_version: 12.6.3
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.7.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
-          - cuda: "128"
-            cuda_version: 12.6.3
-            cudnn_version: ""
-            python_version: "3.11"
-            pytorch: 2.7.0
-            torch_cuda_arch_list: "7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX"
           - cuda: "128"
             cuda_version: 12.8.1
             cudnn_version: ""

.github/workflows/main.yml

@@ -31,11 +31,6 @@ jobs:
             pytorch: 2.6.0
             axolotl_extras: vllm
             is_latest: true
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.0
-            axolotl_extras: vllm
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -98,11 +93,6 @@ jobs:
             pytorch: 2.6.0
             axolotl_extras:
             is_latest: true
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.0
-            axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:
       - name: Checkout
@@ -148,7 +138,7 @@ jobs:
           - cuda: 124
             cuda_version: 12.4.1
             python_version: "3.11"
-            pytorch: 2.6.0
+            pytorch: 2.4.1
             axolotl_extras:
     runs-on: axolotl-gpu-runner
     steps:

.github/workflows/multi-gpu-e2e.yml

@@ -45,13 +45,6 @@ jobs:
             axolotl_extras: vllm
             num_gpus: 2
             nightly_build: "true"
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.0
-            axolotl_extras:
-            num_gpus: 2
-            nightly_build: "true"
     runs-on: [self-hosted, modal]
     timeout-minutes: 120
     steps:
@@ -74,7 +67,6 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
           echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.multigpu

.github/workflows/tests-nightly.yml

@@ -147,7 +147,6 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
           echo "NIGHTLY_BUILD=${{ matrix.nightly_build }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.e2e_tests

.github/workflows/tests.yml

@@ -49,7 +49,7 @@ jobs:
       max-parallel: 2
       matrix:
         python_version: ["3.11"]
-        pytorch_version: ["2.4.1", "2.5.1", "2.6.0", "2.7.0"]
+        pytorch_version: ["2.4.1", "2.5.1", "2.6.0"]
     timeout-minutes: 20
 
     steps:
@@ -109,7 +109,6 @@ jobs:
       - name: Upload coverage to Codecov
         uses: codecov/codecov-action@v5
         with:
-          token: ${{ secrets.CODECOV_TOKEN }}
           files: ./coverage.xml
           flags: unittests,pytorch-${{ matrix.pytorch_version }}
           fail_ci_if_error: false
@@ -242,7 +241,6 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.e2e_tests
@@ -270,12 +268,6 @@ jobs:
             pytorch: 2.5.1
             num_gpus: 1
             axolotl_extras: vllm
-          - cuda: 126
-            cuda_version: 12.6.3
-            python_version: "3.11"
-            pytorch: 2.7.0
-            num_gpus: 1
-            axolotl_extras:
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -296,7 +288,6 @@ jobs:
           echo "CUDA=${{ matrix.cuda }}" >> $GITHUB_ENV
           echo "MODAL_IMAGE_BUILDER_VERSION=2024.10" >> $GITHUB_ENV
           echo "N_GPUS=${{ matrix.num_gpus }}" >> $GITHUB_ENV
-          echo "CODECOV_TOKEN=${{ secrets.CODECOV_TOKEN }}" >> $GITHUB_ENV
       - name: Run tests job on Modal
         run: |
           modal run cicd.e2e_tests

cicd/cicd.sh

@@ -9,7 +9,8 @@ pytest -v --durations=10 -n8 \
   --ignore=tests/patched/ \
   --ignore=tests/cli \
   /workspace/axolotl/tests/ \
-  --cov=axolotl
+  --cov=axolotl \
+  --cov-report=xml:coverage.xml
 
 # Run lora kernels tests with coverage append
 pytest -v --durations=10 \
@@ -50,6 +51,11 @@ pytest -v --durations=10 \
   /workspace/axolotl/tests/e2e/ \
   --cov=axolotl \
   --cov-append \
-  --cov-report=xml:e2e-coverage.xml
+  --cov-report=xml:coverage.xml
 
-codecov upload-process -t $CODECOV_TOKEN -f e2e-coverage.xml -F e2e,pytorch-${PYTORCH_VERSION}
+# Upload coverage to Codecov
+if [ -f e2e-coverage.xml ]; then
+  codecov -f e2e-coverage.xml -F e2e,pytorch-${PYTORCH_VERSION}
+else
+  echo "Coverage file not found. Coverage report may have failed."
+fi

cicd/e2e_tests.py

@@ -28,7 +28,6 @@ df_args = {
     "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
     "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
     "NIGHTLY_BUILD": os.environ.get("NIGHTLY_BUILD", ""),
-    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
     "HF_HOME": "/workspace/data/huggingface-cache/hub",
 }
 

cicd/multigpu.py

@@ -29,7 +29,6 @@ df_args = {
     "CUDA": os.environ.get("CUDA", "121"),
     "GITHUB_REF": os.environ.get("GITHUB_REF", "refs/heads/main"),
     "GITHUB_SHA": os.environ.get("GITHUB_SHA", ""),
-    "CODECOV_TOKEN": os.environ.get("CODECOV_TOKEN", ""),
     "HF_HOME": "/workspace/data/huggingface-cache/hub",
 }
 

cicd/multigpu.sh

@@ -1,23 +1,25 @@
 #!/bin/bash
 set -e
 
+# only run one test at a time so as not to OOM the GPU
+pytest -v  --durations=10 -n2 /workspace/axolotl/tests/e2e/multigpu/ --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/
+pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/solo/
+
 # Only run two tests at a time to avoid OOM on GPU (with coverage collection)
 pytest -v -n2 \
-  --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/ \
+  --ignore=/workspace/axolotl/tests/e2e/multigpu/solo/
-  --ignore=/workspace/axolotl/tests/e2e/multigpu/patched/ \
   /workspace/axolotl/tests/e2e/multigpu/ \
-  --cov=axolotl
-
-# Run solo tests with coverage append
-pytest -v --durations=10 -n1 \
-  /workspace/axolotl/tests/e2e/multigpu/solo/ \
   --cov=axolotl \
-  --cov-append
+  --cov-report=xml:multigpu-coverage.xml
 
-pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/patched/ \
+pytest -v  --durations=10 -n1 /workspace/axolotl/tests/e2e/multigpu/solo/ \
   --cov=axolotl \
   --cov-append \
   --cov-report=xml:multigpu-coverage.xml
 
 # Upload coverage to Codecov
-codecov upload-process -t $CODECOV_TOKEN -f multigpu-coverage.xml -F multigpu,docker-tests,pytorch-${PYTORCH_VERSION}
+if [ -f multigpu-coverage.xml ]; then
+  codecov -f multigpu-coverage.xml -F multigpu,docker-tests,pytorch-${PYTORCH_VERSION}
+else
+  echo "Coverage file not found. Coverage report may have failed."
+fi

codecov.yml

@@ -49,6 +49,3 @@ comment:
   require_changes: no
   require_base: no
   require_head: yes
-
-github_checks:
-  annotations: false

docker/Dockerfile-base

@@ -37,7 +37,3 @@ RUN git lfs install --skip-repo && \
     pip3 install awscli && \
     # The base image ships with `pydantic==1.8.2` which is not working
     pip3 install -U --no-cache-dir pydantic==1.10.10
-
-RUN if [ "$PYTORCH_VERSION" = "2.7.0" ] ; then \
-        pip3 install flash-attn==2.7.4.post1; \
-    fi

docs/cli.qmd

@@ -199,17 +199,6 @@ output_dir: # Directory to save evaluation results
 
 See [LM Eval Harness](https://github.com/EleutherAI/lm-evaluation-harness) for more details.
 
-### delinearize-llama4
-
-Delinearizes a Llama 4 linearized model into a regular HuggingFace Llama 4 model. This only works with the non-quantized linearized model.
-
-```bash
-axolotl delinearize-llama4 --model path/to/model_dir --output path/to/output_dir
-```
-
-This would be necessary to use with other frameworks. If you have an adapter, merge it with the non-quantized linearized model before delinearizing.
-
-
 ## Legacy CLI Usage
 
 While the new Click-based CLI is preferred, Axolotl still supports the legacy module-based CLI:

docs/custom_integrations.qmd

@@ -49,8 +49,7 @@ sections = [
     ("Knowledge Distillation (KD)", "kd"),
     ("Liger Kernels", "liger"),
     ("Language Model Evaluation Harness (LM Eval)", "lm_eval"),
-    ("Spectrum", "spectrum"),
+    ("Spectrum", "spectrum")
-    ("LLMCompressor", "llm_compressor")
 ]
 
 for section_name, folder_name in sections:

docs/installation.qmd

@@ -19,12 +19,6 @@ This guide covers all the ways you can install and set up Axolotl for your envir
 
 ## Installation Methods {#sec-installation-methods}
 
-::: {.callout-important}
-Please make sure to have Pytorch installed before installing Axolotl in your local environment.
-
-Follow the instructions at: [https://pytorch.org/get-started/locally/](https://pytorch.org/get-started/locally/)
-:::
-
 ### PyPI Installation (Recommended) {#sec-pypi}
 
 ```{.bash}

examples/glm4/qlora-32b.yaml

@@ -1,62 +0,0 @@
-base_model: THUDM/GLM-4-32B-0414
-# Automatically upload checkpoint and final model to HF
-# hub_model_id: username/custom_model_name
-
-load_in_4bit: true
-
-datasets:
-  - path: teknium/GPT4-LLM-Cleaned
-    type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0
-output_dir: ./outputs/qlora-out
-
-adapter: qlora
-lora_model_dir:
-
-sequence_len: 2048
-sample_packing: true
-eval_sample_packing: true
-pad_to_sequence_len: true
-
-lora_r: 16
-lora_alpha: 32
-lora_dropout: 0.05
-lora_target_modules:
-  - gate_proj
-  - down_proj
-  - up_proj
-  - q_proj
-  - v_proj
-  - k_proj
-  - o_proj
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 2
-micro_batch_size: 2
-num_epochs: 1
-optimizer: adamw_8bit
-lr_scheduler: cosine
-learning_rate: 0.0002
-
-bf16: auto
-tf32: false
-
-gradient_checkpointing: true
-resume_from_checkpoint:
-logging_steps: 1
-flash_attention: true
-
-loss_watchdog_threshold: 5.0
-loss_watchdog_patience: 3
-
-warmup_steps: 10
-evals_per_epoch: 1
-saves_per_epoch: 1
-weight_decay: 0.0
-special_tokens:

examples/llama-3/sparse-finetuning.yaml

@@ -1,77 +0,0 @@
-base_model: neuralmagic/Sparse-Llama-3.1-8B-2of4
-
-plugins:
-  - axolotl.integrations.llm_compressor.LLMCompressorPlugin
-
-load_in_8bit: false
-load_in_4bit: false
-strict: false
-
-datasets:
-  - path: tatsu-lab/alpaca
-    type: alpaca
-dataset_prepared_path: last_run_prepared
-val_set_size: 0.05
-output_dir: ./outputs/out
-
-sequence_len: 4096
-sample_packing: true
-pad_to_sequence_len: true
-eval_sample_packing: false
-
-wandb_project:
-wandb_entity:
-wandb_watch:
-wandb_name:
-wandb_log_model:
-
-gradient_accumulation_steps: 8
-micro_batch_size: 1
-num_epochs: 1
-optimizer: paged_adamw_8bit
-lr_scheduler: cosine
-learning_rate: 2e-5
-
-train_on_inputs: false
-group_by_length: false
-bf16: auto
-fp16:
-tf32: false
-
-gradient_checkpointing: true
-gradient_checkpointing_kwargs:
-  use_reentrant: false
-early_stopping_patience:
-resume_from_checkpoint:
-logging_steps: 1
-xformers_attention:
-flash_attention: true
-
-warmup_steps: 100
-evals_per_epoch: 2
-eval_table_size:
-saves_per_epoch: 1
-debug:
-deepspeed:
-weight_decay: 0.0
-fsdp:
-fsdp_config:
-special_tokens:
-  pad_token: <|end_of_text|>
-
-llmcompressor:
-  recipe:
-    finetuning_stage:
-      finetuning_modifiers:
-        ConstantPruningModifier:
-          targets: [
-            're:.*q_proj.weight',
-            're:.*k_proj.weight',
-            're:.*v_proj.weight',
-            're:.*o_proj.weight',
-            're:.*gate_proj.weight',
-            're:.*up_proj.weight',
-            're:.*down_proj.weight',
-          ]
-          start: 0
-  save_compressed: true

examples/llama-4/README.md

@@ -26,11 +26,3 @@ Multi-GPU (4xH100) for Llama 4 Scout uses 62.8GB VRAM/GPU @ 4k contenxt length @
 ### Llama 4 Maverick 17Bx128Experts (400B)
 
 Coming Soon
-
-## Delinearized Llama 4 Models
-
-We provide a script to delinearize Llama 4 linearized models into regular HuggingFace Llama 4 models.
-
-```bash
-axolotl delinearize-llama4 --model path/to/model_dir --output path/to/output_dir
-```

requirements-tests.txt

@@ -1,5 +1,4 @@
 codecov
-codecov-cli
 pytest
 pytest-cov
 pytest-retry

requirements.txt

@@ -6,7 +6,7 @@ triton>=3.0.0
 mamba-ssm==1.2.0.post1
 xformers>=0.0.23.post1
 autoawq==0.2.7.post3
-liger-kernel==0.5.8
+liger-kernel==0.5.6
 # END section
 
 packaging==23.2
@@ -19,7 +19,6 @@ datasets==3.5.0
 deepspeed>=0.15.4
 trl==0.16.1
 hf_xet==1.0.0
-hqq==0.2.5
 
 optimum==1.16.2
 hf_transfer

setup.py

@@ -51,7 +51,7 @@ def parse_requirements(extras_require_map):
             try:
                 torch_version = version("torch")
             except PackageNotFoundError:
-                torch_version = "2.6.0"  # default to torch 2.6
+                torch_version = "2.5.1"
             _install_requires.append(f"torch=={torch_version}")
 
             version_match = re.match(r"^(\d+)\.(\d+)(?:\.(\d+))?", torch_version)
@@ -64,15 +64,9 @@ def parse_requirements(extras_require_map):
             else:
                 raise ValueError("Invalid version format")
 
-            if (major, minor) >= (2, 7):
+            if (major, minor) >= (2, 6):
                 _install_requires.pop(_install_requires.index(xformers_version))
-                # _install_requires.append("xformers==0.0.29.post3")  # xformers seems to be hard pinned to 2.6.0
+                _install_requires.append("xformers==0.0.29.post2")
-                extras_require_map["vllm"] = ["vllm==0.8.3"]
-            elif (major, minor) >= (2, 6):
-                _install_requires.pop(_install_requires.index(xformers_version))
-                _install_requires.append(
-                    "xformers==0.0.29.post2"
-                )  # vllm needs post2 w torch 2.6
                 extras_require_map["vllm"] = ["vllm==0.8.3"]
             elif (major, minor) >= (2, 5):
                 _install_requires.pop(_install_requires.index(xformers_version))
@@ -149,9 +143,6 @@ extras_require = {
     "vllm": [
         "vllm==0.7.2",
     ],
-    "llmcompressor": [
-        "llmcompressor==0.5.1",
-    ],
 }
 
 install_requires, dependency_links, extras_require_build = parse_requirements(

src/axolotl/cli/args.py

@@ -39,16 +39,16 @@ class TrainerCliArgs:
 class VllmServeCliArgs:
     """Dataclass with CLI arguments for `axolotl vllm-serve` command."""
 
-    tensor_parallel_size: Optional[int] = field(
+    tensor_parallel_size: int = field(
-        default=None,
+        default=1,
         metadata={"help": "Number of tensor parallel workers to use."},
     )
-    host: Optional[str] = field(
+    host: str = field(
-        default=None,  # nosec B104
+        default="0.0.0.0",  # nosec B104
         metadata={"help": "Host address to run the server on."},
     )
-    port: Optional[int] = field(
+    port: int = field(
-        default=None,
+        default=8000,
         metadata={"help": "Port to run the server on."},
     )
     gpu_memory_utilization: Optional[float] = field(

src/axolotl/core/trainer_builder.py

@@ -1040,11 +1040,9 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
         if self.cfg.dataset_processes:
             training_args_kwargs["dataset_num_proc"] = self.cfg.dataset_processes
 
-        if self.cfg.trl and self.cfg.trl.beta is not None:
+        if (self.cfg.trl and self.cfg.trl.beta) or self.cfg.rl_beta:
-            training_args_kwargs["beta"] = self.cfg.trl.beta
+            training_args_kwargs["beta"] = self.cfg.trl.beta or self.cfg.rl_beta
-        elif self.cfg.rl_beta is not None:
+        if self.cfg.orpo_alpha:
-            training_args_kwargs["beta"] = self.cfg.rl_beta
-        elif self.cfg.orpo_alpha is not None:
             # trl does some odd mapping of alpha to beta to reuse the beta parameter ???
             training_args_kwargs["beta"] = self.cfg.orpo_alpha
 

src/axolotl/core/trainers/grpo/__init__.py

@@ -40,8 +40,8 @@ class GRPOStrategy:
 
         if trl.use_vllm:
             grpo_args_kwargs["use_vllm"] = trl.use_vllm
-            grpo_args_kwargs["vllm_server_host"] = trl.vllm_server_host or trl.vllm.host
+            grpo_args_kwargs["vllm_server_host"] = trl.vllm_server_host
-            grpo_args_kwargs["vllm_server_port"] = trl.vllm_server_port or trl.vllm.port
+            grpo_args_kwargs["vllm_server_port"] = trl.vllm_server_port
             if trl.vllm_server_timeout:
                 grpo_args_kwargs["vllm_server_timeout"] = trl.vllm_server_timeout
             if trl.vllm_guided_decoding_regex:

src/axolotl/integrations/cut_cross_entropy/README.md

@@ -47,8 +47,6 @@ cut_cross_entropy: true
 - qwen2
 - cohere
 - cohere2
-- glm
-- glm4
 
 ## Citation
 

src/axolotl/integrations/cut_cross_entropy/monkeypatch/glm4.py

@@ -1,57 +0,0 @@
-"""GLM 4 patch. GLM family inherits from Llama."""
-
-from types import MethodType
-
-import transformers
-from cut_cross_entropy.transformers.utils import (
-    PatchOptions,
-    TransformersModelT,
-)
-
-
-def patch_glm(
-    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
-    patch_options: PatchOptions,
-) -> TransformersModelT | None:
-
-    # Set the _PATCH_OPTS in the llama patch file
-    import cut_cross_entropy.transformers.llama as llama_patch
-
-    llama_patch._PATCH_OPTS = patch_options  # pylint: disable=protected-access
-
-    from cut_cross_entropy.transformers.llama import cce_forward
-    from transformers.models.glm import modeling_glm
-
-    if isinstance(maybe_model, transformers.PreTrainedModel):
-        assert isinstance(
-            maybe_model, modeling_glm.GlmForCausalLM
-        ), f"Expected a GlmForCausalLM model. Got {type(maybe_model)}."
-        maybe_model.forward = MethodType(cce_forward, maybe_model)
-        return maybe_model
-
-    modeling_glm.GlmForCausalLM.forward = cce_forward
-    return None
-
-
-def patch_glm4(
-    maybe_model: TransformersModelT | str | transformers.PretrainedConfig,
-    patch_options: PatchOptions,
-) -> TransformersModelT | None:
-
-    # Set the _PATCH_OPTS in the llama patch file
-    import cut_cross_entropy.transformers.llama as llama_patch
-
-    llama_patch._PATCH_OPTS = patch_options  # pylint: disable=protected-access
-
-    from cut_cross_entropy.transformers.llama import cce_forward
-    from transformers.models.glm4 import modeling_glm4
-
-    if isinstance(maybe_model, transformers.PreTrainedModel):
-        assert isinstance(
-            maybe_model, modeling_glm4.Glm4ForCausalLM
-        ), f"Expected a Glm4ForCausalLM model. Got {type(maybe_model)}."
-        maybe_model.forward = MethodType(cce_forward, maybe_model)
-        return maybe_model
-
-    modeling_glm4.Glm4ForCausalLM.forward = cce_forward
-    return None

src/axolotl/integrations/cut_cross_entropy/monkeypatch/patch.py

@@ -20,10 +20,6 @@ from axolotl.integrations.cut_cross_entropy.monkeypatch.gemma3 import (
     patch_gemma3,
     patch_gemma3_text,
 )
-from axolotl.integrations.cut_cross_entropy.monkeypatch.glm4 import (
-    patch_glm,
-    patch_glm4,
-)
 from axolotl.integrations.cut_cross_entropy.monkeypatch.llama4 import (
     patch_llama4,
     patch_llama4_text,
@@ -49,8 +45,6 @@ CUT_CROSS_ENTROPY_MODEL_MAPPING = {
     "qwen2": patch_qwen2,
     "cohere": patch_cohere,
     "cohere2": patch_cohere2,
-    "glm": patch_glm,
-    "glm4": patch_glm4,
 }
 
 

src/axolotl/integrations/liger/README.md

@@ -25,7 +25,7 @@ liger_fused_linear_cross_entropy: true
 - deepseek_v2
 - gemma
 - gemma2
-- gemma3
+- gemma3 (partial support, no support for FLCE yet)
 - granite
 - jamba
 - llama

src/axolotl/integrations/liger/__init__.py

@@ -21,6 +21,7 @@ It is designed to be performant, correct, and light-weight.
 import inspect
 import logging
 import sys
+from functools import partial
 
 from axolotl.integrations.base import BasePlugin
 
@@ -54,6 +55,7 @@ class LigerPlugin(BasePlugin):
             )
         from liger_kernel.transformers.cross_entropy import LigerCrossEntropyLoss
         from liger_kernel.transformers.functional import liger_cross_entropy
+        from liger_kernel.transformers.geglu import LigerGEGLUMLP
         from liger_kernel.transformers.layer_norm import LigerLayerNorm
         from liger_kernel.transformers.monkey_patch import MODEL_TYPE_TO_APPLY_LIGER_FN
         from liger_kernel.transformers.rms_norm import LigerRMSNorm
@@ -139,6 +141,38 @@ class LigerPlugin(BasePlugin):
                 modeling_mod.CrossEntropyLoss = LigerCrossEntropyLoss
             if cfg.liger_fused_linear_cross_entropy:
                 modeling_mod.DeepseekV2ForCausalLM.forward = deepseekv2_lce_forward
+        elif cfg.model_config_type in ["gemma3", "gemma3_text"]:
+            from transformers.models.gemma3 import modeling_gemma3
+
+            if cfg.liger_rope:
+                modeling_gemma3.apply_rotary_pos_emb = liger_rotary_pos_emb
+            if cfg.liger_rms_norm:
+
+                def _liger_rms_norm_wrapper(dim, **kwargs):
+                    "Convert 'dim' keyword to 'hidden_size' to pass to LigerRMSNorm"
+                    return LigerRMSNorm(hidden_size=dim, **kwargs)
+
+                modeling_gemma3.Gemma3RMSNorm = partial(
+                    _liger_rms_norm_wrapper,
+                    offset=1.0,
+                    casting_mode="gemma",
+                    init_fn="zeros",
+                    in_place=False,
+                )
+            if cfg.liger_glu_activation:
+                modeling_gemma3.Gemma3MLP = LigerGEGLUMLP
+            if cfg.liger_layer_norm:
+                modeling_gemma3.nn.LayerNorm = LigerLayerNorm
+
+            if cfg.liger_cross_entropy:
+                from transformers.loss.loss_utils import nn
+
+                nn.functional.cross_entropy = liger_cross_entropy
+
+            if cfg.liger_fused_linear_cross_entropy:
+                raise NotImplementedError(
+                    "Fused linear cross entropy is not yet supported for Gemma3."
+                )
         elif cfg.model_config_type == "llama4":
             from axolotl.integrations.liger.models.llama4 import (
                 apply_liger_kernel_to_llama4,

src/axolotl/integrations/llm_compressor/README.md

@@ -1,108 +0,0 @@
-# LLMCompressor Integration
-
-Fine-tune sparsified models in Axolotl using Neural Magic's [LLMCompressor](https://github.com/vllm-project/llm-compressor).
-
-This integration enables fine-tuning of models sparsified using LLMCompressor within the Axolotl training framework. By combining LLMCompressor's model compression capabilities with Axolotl's distributed training pipelines, users can efficiently fine-tune sparse models at scale.
-
-It uses Axolotl’s plugin system to hook into the fine-tuning flows while maintaining sparsity throughout training.
-
----
-
-## Requirements
-
-- Axolotl with `llmcompressor` extras:
-
-  ```bash
-  pip install "axolotl[llmcompressor]"
-  ```
-
-- Requires `llmcompressor >= 0.5.1`
-
-This will install all necessary dependencies to fine-tune sparsified models using the integration.
-
----
-
-## Usage
-
-To enable sparse fine-tuning with this integration, include the plugin in your Axolotl config:
-
-```yaml
-plugins:
-  - axolotl.integrations.llm_compressor.LLMCompressorPlugin
-
-llmcompressor:
-  recipe:
-    finetuning_stage:
-      finetuning_modifiers:
-        ConstantPruningModifier:
-          targets: [
-            're:.*q_proj.weight',
-            're:.*k_proj.weight',
-            're:.*v_proj.weight',
-            're:.*o_proj.weight',
-            're:.*gate_proj.weight',
-            're:.*up_proj.weight',
-            're:.*down_proj.weight',
-          ]
-          start: 0
-  save_compressed: true
-# ... (other training arguments)
-```
-
-This plugin **does not apply pruning or sparsification itself** — it is intended for **fine-tuning models that have already been sparsified**.
-
-Pre-sparsified checkpoints can be:
-- Generated using [LLMCompressor](https://github.com/vllm-project/llm-compressor)
-- Downloaded from [Neural Magic's Hugging Face page](https://huggingface.co/neuralmagic)
-- Any custom LLM with compatible sparsity patterns that you've created yourself
-
-To learn more about writing and customizing LLMCompressor recipes, refer to the official documentation:
-[https://github.com/vllm-project/llm-compressor/blob/main/README.md](https://github.com/vllm-project/llm-compressor/blob/main/README.md)
-
-### Storage Optimization with save_compressed
-
-Setting `save_compressed: true` in your configuration enables saving models in a compressed format, which:
-- Reduces disk space usage by approximately 40%
-- Maintains compatibility with vLLM for accelerated inference
-- Maintains compatibility with llmcompressor for further optimization (example: quantization)
-
-This option is highly recommended when working with sparse models to maximize the benefits of model compression.
-
-### Example Config
-
-See [`examples/llama-3/sparse-finetuning.yaml`](examples/llama-3/sparse-finetuning.yaml) for a complete example.
-
----
-
-## Inference with vLLM
-
-After fine-tuning your sparse model, you can leverage vLLM for efficient inference.
-You can also use LLMCompressor to apply additional quantization to your fine-tuned
-sparse model before inference for even greater performance benefits.:
-
-```python
-from vllm import LLM, SamplingParams
-
-prompts = [
-    "Hello, my name is",
-    "The president of the United States is",
-    "The capital of France is",
-    "The future of AI is",
-]
-sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
-llm = LLM("path/to/your/sparse/model")
-outputs = llm.generate(prompts, sampling_params)
-
-for output in outputs:
-    prompt = output.prompt
-    generated_text = output.outputs[0].text
-    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
-```
-
-For more details on vLLM's capabilities and advanced configuration options, see the [official vLLM documentation](https://docs.vllm.ai/).
-
-## Learn More
-
-For details on available sparsity and quantization schemes, fine-tuning recipes, and usage examples, visit the official LLMCompressor repository:
-
-[https://github.com/vllm-project/llm-compressor](https://github.com/vllm-project/llm-compressor)

src/axolotl/integrations/llm_compressor/__init__.py

@@ -1,5 +0,0 @@
-"""Integration entry point for the LLMCompressor plugin."""
-
-from .plugin import LLMCompressorPlugin
-
-__all__ = ["LLMCompressorPlugin"]

src/axolotl/integrations/llm_compressor/args.py

@@ -1,40 +0,0 @@
-"""
-LLMCompressor and Sparse Finetuning config models.
-"""
-
-from typing import Any
-
-from pydantic import BaseModel, Field
-from typing_extensions import Annotated
-
-
-class CompressionArgs(BaseModel):
-    """Sparse Finetuning config for LLMCompressor."""
-
-    # Typing for recipe is set to Any due to:
-    # https://github.com/vllm-project/llm-compressor/issues/1319
-    recipe: Annotated[
-        Any,
-        Field(
-            description="The recipe containing the compression algorithms and hyperparameters to apply."
-        ),
-    ]
-
-    save_compressed: Annotated[
-        bool,
-        Field(
-            default=False,
-            description="Whether to save the compressed model after training.",
-        ),
-    ]
-
-
-class LLMCompressorArgs(BaseModel):
-    """LLMCompressor configuration BaseModel."""
-
-    llmcompressor: Annotated[
-        CompressionArgs,
-        Field(
-            description="Arguments enabling compression pathways through the LLM Compressor plugins"
-        ),
-    ]

src/axolotl/integrations/llm_compressor/plugin.py

@@ -1,171 +0,0 @@
-"""
-Sparse Finetuning plugin for Axolotl — enables handling of sparse neural networks
-by maintaining masks for zero weights during training.
-"""
-
-import logging
-from functools import wraps
-from typing import Any, Callable, Concatenate, ParamSpec, TypeVar
-
-from llmcompressor import active_session, create_session
-from llmcompressor.core import callbacks as session_callbacks
-from llmcompressor.recipe import Recipe
-from torch.nn import Module
-from transformers.trainer import Trainer
-from transformers.trainer_callback import TrainerCallback, TrainerControl, TrainerState
-from transformers.training_args import TrainingArguments
-
-from axolotl.integrations.base import BasePlugin
-
-P = ParamSpec("P")  # Params for generic function signatures
-R = TypeVar("R")  # Return type for generic function signatures
-
-LOG = logging.getLogger("axolotl.integrations.llm_compressor")
-
-
-class LLMCompressorCallbackHandler(TrainerCallback):
-    """
-    Trainer callback for Sparse Finetuning.
-    Maintains sparsity patterns during training by applying masks after optimization steps,
-    ensuring zero-weight updates are canceled out.
-    """
-
-    def __init__(self, trainer: Trainer, recipe: Any):
-        """
-        Initialize the Sparse Finetuning callback handler.
-
-        Args:
-            trainer (Trainer): Huggingface Trainer instance.
-            recipe (Recipe | dict): Sparse finetuning recipe to apply.
-        """
-        super().__init__()
-        self.trainer = trainer
-        self.recipe = (
-            Recipe.model_validate(recipe) if not isinstance(recipe, Recipe) else recipe
-        )
-        self.original_compute_loss = trainer.compute_loss
-        self.trainer.compute_loss = compute_loss_wrapper(self.trainer.compute_loss)
-        create_session()
-
-    def on_train_begin(
-        self,
-        args: TrainingArguments,
-        state: TrainerState,
-        control: TrainerControl,
-        **kwargs,
-    ) -> None:
-        """
-        Called at the beginning of training. Initializes the compression session.
-
-        Args:
-            args (TrainingArguments): Training arguments.
-            state (TrainerState): Trainer state.
-            control (TrainerControl): Trainer control.
-        """
-        super().on_train_begin(args, state, control, **kwargs)
-        self.trainer.accelerator.wait_for_everyone()
-        active_session().initialize(
-            model=self.trainer.model,
-            optimizer=self.trainer.optimizer,
-            start=state.epoch,
-            recipe=self.recipe,
-        )
-        self.trainer.accelerator.wait_for_everyone()
-
-    def on_step_begin(
-        self,
-        args: TrainingArguments,
-        state: TrainerState,
-        control: TrainerControl,
-        **kwargs,
-    ) -> None:
-        """
-        Called at the beginning of a training step. Triggers batch_start callback.
-        """
-        super().on_step_begin(args, state, control, **kwargs)
-        session_callbacks.batch_start()
-
-    def on_step_end(
-        self,
-        args: TrainingArguments,
-        state: TrainerState,
-        control: TrainerControl,
-        **kwargs,
-    ) -> None:
-        """
-        Called at the end of a training step. Triggers optimizer and batch_end callbacks.
-        """
-        super().on_step_end(args, state, control, **kwargs)
-        session_callbacks.optim_pre_step()
-        session_callbacks.optim_post_step()
-        session_callbacks.batch_end()
-
-    def on_train_end(
-        self,
-        args: TrainingArguments,
-        state: TrainerState,
-        control: TrainerControl,
-        **kwargs,
-    ) -> None:
-        """
-        Called at the end of training. Finalizes the compression session.
-        """
-        super().on_train_end(args, state, control, **kwargs)
-        active_session().finalize()
-        self.trainer.compute_loss_func = self.original_compute_loss
-
-
-class LLMCompressorPlugin(BasePlugin):
-    """
-    Sparse Finetuning plugin for Axolotl integration.
-    """
-
-    def get_input_args(self) -> str:
-        """
-        Returns the path to the plugin's argument definition.
-
-        Returns:
-            str: Dotted path to the LLMCompressorArgs class.
-        """
-        return "axolotl.integrations.llm_compressor.args.LLMCompressorArgs"
-
-    def add_callbacks_post_trainer(self, cfg: Any, trainer: Trainer) -> list:
-        """
-        Adds Sparse Finetuning callback to the Trainer instance.
-
-        Args:
-            cfg (Any): Configuration object containing the sparse recipe.
-            trainer (Trainer): Huggingface Trainer instance.
-
-        Returns:
-            list: List containing the configured callback instances.
-        """
-        LOG.info("Adding Sparse Finetuning callback to the trainer")
-        callback = LLMCompressorCallbackHandler(
-            trainer=trainer,
-            recipe=cfg.llmcompressor.recipe,
-        )
-        return [callback]
-
-
-def compute_loss_wrapper(
-    compute_loss_func: Callable[Concatenate[Module, P], R],
-) -> Callable[Concatenate[Module, P], R]:
-    """
-    Wraps the loss computation function to trigger the loss_calculated callback.
-
-    Args:
-        compute_loss_func (Callable): Original loss computation function.
-
-    Returns:
-        Callable: Wrapped function that also invokes the loss_calculated callback.
-    """
-
-    @wraps(compute_loss_func)
-    def compute_and_notify(model: Module, *args: P.args, **kwargs: P.kwargs) -> R:
-        loss = compute_loss_func(model, *args, **kwargs)
-        if active_session().lifecycle.initialized_ and model.training:
-            session_callbacks.loss_calculated(loss=loss)
-        return loss
-
-    return compute_and_notify

src/axolotl/integrations/llm_compressor/utils.py

@@ -1,40 +0,0 @@
-"""Utilities for llmcompressor integration with axolotl."""
-
-from typing import Union
-
-from llmcompressor.transformers.sparsification.compressed_tensors_utils import (
-    modify_save_pretrained,
-)
-from transformers import PreTrainedModel, Trainer
-
-
-def save_compressed_model(
-    model: PreTrainedModel,
-    output_dir: Union[str, bytes],
-    trainer: Trainer,
-    safe_serialization: bool = False,
-    save_compressed: bool = False,
-) -> None:
-    """
-    Synchronize processes, apply compression hooks, and save the model.
-
-    Args:
-        model (PreTrainedModel): The model to be saved.
-        output_dir (str or bytes): Path where the model files will be written.
-        trainer (Trainer): Hugging Face Trainer for process synchronization.
-        safe_serialization (bool): Use safe serialization if True.
-        save_compressed (bool): Write compressed tensors if True.
-    """
-    trainer.accelerator.wait_for_everyone()
-
-    # Only the main process writes the files
-    if not trainer.accelerator.is_main_process:
-        return
-
-    modify_save_pretrained(model)
-    model.save_pretrained(
-        output_dir,
-        safe_serialization=safe_serialization,
-        save_compressed=save_compressed,
-        skip_sparsity_compression_stats=not save_compressed,
-    )

src/axolotl/monkeypatch/multipack.py

@@ -31,8 +31,6 @@ SUPPORTED_MULTIPACK_MODEL_TYPES = [
     "starcoder2",
     "deepseek_v2",
     "deepseek_v3",
-    "glm",
-    "glm4",
 ]
 
 

src/axolotl/monkeypatch/relora.py

@@ -272,7 +272,7 @@ class ReLoRAScheduler(LRScheduler):
         self.warmup_steps = warmup_steps
         self.anneal_steps = anneal_steps
         self.min_lr_scale = min_lr_scale
-        super().__init__(optimizer, inner_schedule.last_epoch)
+        super().__init__(optimizer, inner_schedule.last_epoch, inner_schedule.verbose)
 
     def get_lr(self) -> float:
         self.inner_schedule.last_epoch = self.last_epoch

src/axolotl/train.py

@@ -271,19 +271,6 @@ def save_trained_model(
                 os.remove(os.path.join(cfg.output_dir, "model.safetensors"))
             except FileNotFoundError:
                 pass
-    elif hasattr(cfg, "llmcompressor") and cfg.llmcompressor:
-        from axolotl.integrations.llm_compressor.utils import (
-            save_compressed_model,
-        )
-
-        save_compressed_model(
-            model=model,
-            output_dir=cfg.output_dir,
-            trainer=trainer,
-            safe_serialization=safe_serialization,
-            save_compressed=cfg.llmcompressor.save_compressed,
-        )
-
     elif cfg.local_rank == 0:
         if cfg.flash_optimum and BetterTransformer:
             model = BetterTransformer.reverse(model)
@@ -292,7 +279,6 @@ def save_trained_model(
             trainer.model.save_pretrained(
                 cfg.output_dir, safe_serialization=safe_serialization
             )
-
         model.save_pretrained(cfg.output_dir, safe_serialization=safe_serialization)
 
 

src/axolotl/utils/data/sft.py

@@ -3,7 +3,6 @@
 import functools
 import logging
 import os
-import tempfile
 from pathlib import Path
 from typing import List, Optional, Tuple, Union
 
@@ -118,27 +117,9 @@ def prepare_dataset(cfg, tokenizer, processor=None, preprocess_iterable=None):
             cfg.pretraining_dataset[0]["type"] or "pretrain",
         )
 
-        # when letting accelerator dispatch batches from the main process, we don't need to load the dataset from
+        iter_ds = load_dataset(
-        # other ranks, we just need to present a fake dataset
+            path, streaming=True, split=split, name=name, data_files=data_files
-        if (
+        )
-            cfg.accelerator_config
-            and cfg.accelerator_config.dispatch_batches
-            and not is_local_main_process()
-        ):
-            with tempfile.NamedTemporaryFile(mode="w+", delete=False) as f:
-                f.write("text\n")
-                f.write("lorem ipsum dolor sit amet\n")
-                # rewind the file pointer to the beginning so we can read it again
-                f.seek(0)
-                iter_ds = load_dataset(
-                    "csv", data_files=f.name, split="train", streaming=True
-                )
-        else:
-            if is_local_main_process():
-                iter_ds = load_dataset(
-                    path, streaming=True, split=split, name=name, data_files=data_files
-                )
-
         if skip:
             LOG.info(f"Skipping {skip} samples from the dataset")
             iter_ds = iter_ds.skip(skip)

src/axolotl/utils/models.py

@@ -139,22 +139,6 @@ def check_model_config(cfg: DictDefault, model_config: PretrainedConfig):
         hasattr(model_config, "quantization_config")
         and model_config.quantization_config
     )
-
-    # Detect compressed-tensors config
-    is_compressed_tensors_config = (
-        quant_config_exists
-        and model_config.quantization_config.get("quant_method") == "compressed-tensors"
-    )
-
-    if is_compressed_tensors_config:
-        if model_config.quantization_config.get("config_groups"):
-            LOG.warning(
-                "Found `config_groups` in a compressed-tensors config. "
-                "QAT integration with llmcompressor is not tested."
-            )
-        # Skip further quant checks for compressed-tensors
-        return
-
     quant_config_method_is_gptq = (
         quant_config_exists
         and "quant_method" in model_config.quantization_config

src/axolotl/utils/schedulers.py

@@ -40,7 +40,7 @@ class RexLR(LRScheduler):
         self.max_lr = max_lr
         self.total_steps = total_steps
         self.num_warmup_steps = num_warmup_steps
-        self.last_step = max(last_step - 1, 0)
+        self.last_step = last_step - 1
 
         # Ensure each parameter group has an "initial_lr" key to avoid issues when resuming.
         for group in optimizer.param_groups:

src/axolotl/utils/schemas/config.py

@@ -660,7 +660,6 @@ class AxolotlInputConfig(
             data.get("val_set_size") == 0
             and (data.get("eval_steps") or data.get("eval_strategy"))
             and not data.get("test_datasets")
-            and data.get("eval_strategy") != "no"
         ):
             raise ValueError(
                 "eval_steps and eval_strategy are not supported with val_set_size == 0"

src/axolotl/utils/schemas/vllm.py

@@ -36,11 +36,3 @@ class VllmConfig(BaseModel):
         default=None,
         json_schema_extra={"description": "Enable prefix caching for VLLM"},
     )
-    host: str | None = Field(
-        default="0.0.0.0",  # nosec B104
-        json_schema_extra={"description": "Host for the vLLM server to start on"},
-    )
-    port: int | None = Field(
-        default=8000,
-        json_schema_extra={"description": "Port of the vLLM server to start on"},
-    )

tests/conftest.py

@@ -193,14 +193,6 @@ def download_tiny_shakespeare_dataset():
     snapshot_download_w_retry("winglian/tiny-shakespeare", repo_type="dataset")
 
 
-@pytest.fixture(scope="session", autouse=True)
-def download_evolkit_kd_sample_dataset():
-    # download the dataset
-    snapshot_download_w_retry(
-        "axolotl-ai-co/evolkit-logprobs-pipeline-75k-v2-sample", repo_type="dataset"
-    )
-
-
 @pytest.fixture(scope="session", autouse=True)
 def download_deepseek_model_fixture():
     snapshot_download_w_retry("axolotl-ai-co/DeepSeek-V3-11M", repo_type="model")
@@ -216,16 +208,6 @@ def download_huggyllama_model_fixture():
     )
 
 
-@pytest.fixture(scope="session", autouse=True)
-def download_llama33_70b_model_fixture():
-    # download the tokenizer only
-    snapshot_download_w_retry(
-        "axolotl-ai-co/Llama-3.3-70B-Instruct-tokenizer",
-        repo_type="model",
-        allow_patterns=["*token*", "config.json"],
-    )
-
-
 @pytest.fixture(scope="session", autouse=True)
 def download_llama_1b_model_fixture():
     # download the tokenizer only
@@ -333,14 +315,6 @@ def download_llama2_model_fixture():
     )
 
 
-@pytest.fixture(scope="session", autouse=True)
-def download_llama32_1b_model_fixture():
-    snapshot_download_w_retry(
-        "osllmai-community/Llama-3.2-1B",
-        repo_type="model",
-    )
-
-
 @pytest.fixture
 @enable_hf_offline
 def tokenizer_huggyllama(
@@ -522,6 +496,12 @@ def dataset_fozziethebeat_alpaca_messages_2k_dpo_test_rev_ea82cff(
     return datasets.load_from_disk(ds_path)["train"]
 
 
+@pytest.fixture(scope="session", autouse=True)
+def download_tiny_llama_7m_model():
+    # download the model
+    return snapshot_download_w_retry("axolotl-ai-internal/llama-7m", repo_type="model")
+
+
 # # pylint: disable=redefined-outer-name,unused-argument
 # def test_load_fixtures(
 #     download_smollm2_135m_model,

tests/e2e/integrations/test_cut_cross_entropy.py

@@ -8,7 +8,7 @@ from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
 from axolotl.utils import get_pytorch_version
-from axolotl.utils.config import normalize_config, prepare_plugins, validate_config
+from axolotl.utils.config import normalize_config, prepare_plugins
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists
@@ -56,7 +56,6 @@ class TestCutCrossEntropyIntegration:
     # pylint: disable=redefined-outer-name
     def test_llama_w_cce(self, min_cfg, temp_dir):
         cfg = DictDefault(min_cfg)
-        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
@@ -102,7 +101,6 @@ class TestCutCrossEntropyIntegration:
                 "bf16": "auto",
             }
         )
-        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
@@ -131,7 +129,6 @@ class TestCutCrossEntropyIntegration:
                 attention_type: True,
             }
         )
-        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()

tests/e2e/integrations/test_kd.py

@@ -90,7 +90,7 @@ class TestKnowledgeDistillation:
         train(cfg=cfg, dataset_meta=dataset_meta)
         assert (Path(temp_dir) / "model.safetensors").exists()
         check_tensorboard(
-            temp_dir + "/runs", "train/loss", 1.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/loss", 1.0, "Train loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -121,5 +121,5 @@ class TestKnowledgeDistillation:
         train(cfg=cfg, dataset_meta=dataset_meta)
         assert (Path(temp_dir) / "adapter_model.safetensors").exists()
         check_tensorboard(
-            temp_dir + "/runs", "train/loss", 1.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/loss", 1.0, "Train loss (%s) is too high"
         )

tests/e2e/integrations/test_liger.py

@@ -5,7 +5,7 @@ Simple end-to-end test for Liger integration
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, prepare_plugins, validate_config
+from axolotl.utils.config import normalize_config, prepare_plugins
 from axolotl.utils.dict import DictDefault
 
 from tests.e2e.utils import check_model_output_exists, require_torch_2_4_1
@@ -54,7 +54,6 @@ class LigerIntegrationTestCase:
             }
         )
         # pylint: disable=duplicate-code
-        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
@@ -101,7 +100,6 @@ class LigerIntegrationTestCase:
             }
         )
         # pylint: disable=duplicate-code
-        cfg = validate_config(cfg)
         prepare_plugins(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()

tests/e2e/integrations/test_llm_compressor.py

@@ -1,104 +0,0 @@
-"""
-E2E smoke tests for LLMCompressorPlugin integration
-"""
-
-from pathlib import Path
-
-import pytest
-
-from axolotl.cli.args import TrainerCliArgs
-from axolotl.common.datasets import load_datasets
-from axolotl.train import train
-from axolotl.utils.config import normalize_config, prepare_plugins, validate_config
-from axolotl.utils.dict import DictDefault
-
-from tests.e2e.utils import check_model_output_exists, require_torch_2_4_1
-
-MODELS = [
-    "nm-testing/llama2.c-stories42M-pruned2.4-compressed",
-    "nm-testing/llama2.c-stories42M-gsm8k-sparse-only-compressed",
-]
-
-
-@pytest.mark.parametrize(
-    "base_model", MODELS, ids=["no-checkpoint-recipe", "with-checkpoint-recipe"]
-)
-@pytest.mark.parametrize(
-    "save_compressed", [True, False], ids=["save_compressed", "save_uncompressed"]
-)
-class TestLLMCompressorIntegration:
-    """
-    e2e tests for axolotl.integrations.llm_compressor.LLMCompressorPlugin
-    """
-
-    @require_torch_2_4_1
-    def test_llmcompressor_plugin(
-        self, temp_dir, base_model: str, save_compressed: bool
-    ):
-        # core cfg
-        cfg = DictDefault(
-            {
-                "base_model": base_model,
-                "plugins": ["axolotl.integrations.llm_compressor.LLMCompressorPlugin"],
-                "sequence_len": 1024,
-                "val_set_size": 0.05,
-                "special_tokens": {"pad_token": "<|endoftext|>"},
-                "datasets": [{"path": "mhenrichsen/alpaca_2k_test", "type": "alpaca"}],
-                "num_epochs": 1,
-                "micro_batch_size": 2,
-                "gradient_accumulation_steps": 2,
-                "output_dir": temp_dir,
-                "learning_rate": 1e-5,
-                "optimizer": "adamw_torch_fused",
-                "lr_scheduler": "cosine",
-                "save_safetensors": True,
-                "bf16": "auto",
-                "max_steps": 5,
-                "llmcompressor": {
-                    "recipe": {
-                        "finetuning_stage": {
-                            "finetuning_modifiers": {
-                                "ConstantPruningModifier": {
-                                    "targets": [
-                                        "re:.*q_proj.weight",
-                                        "re:.*k_proj.weight",
-                                        "re:.*v_proj.weight",
-                                        "re:.*o_proj.weight",
-                                        "re:.*gate_proj.weight",
-                                        "re:.*up_proj.weight",
-                                        "re:.*down_proj.weight",
-                                    ],
-                                    "start": 0,
-                                },
-                            },
-                        },
-                    },
-                    "save_compressed": save_compressed,
-                },
-            }
-        )
-
-        prepare_plugins(cfg)
-        cfg = validate_config(cfg)
-        normalize_config(cfg)
-        cli_args = TrainerCliArgs()
-        dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
-
-        train(cfg=cfg, dataset_meta=dataset_meta)
-        check_model_output_exists(temp_dir, cfg)
-        _check_llmcompressor_model_outputs(temp_dir, save_compressed)
-
-
-def _check_llmcompressor_model_outputs(temp_dir, save_compressed):
-
-    # recipe.yaml should exist
-    assert (Path(temp_dir) / "recipe.yaml").exists()
-
-    # sparsity config exists if save_compressed
-    if save_compressed:
-        from compressed_tensors import ModelCompressor
-        from compressed_tensors.config import Sparse24BitMaskConfig
-
-        compressor = ModelCompressor.from_pretrained(temp_dir)
-        assert compressor is not None
-        assert isinstance(compressor.sparsity_config, Sparse24BitMaskConfig)

tests/e2e/multigpu/solo/__init__.py

@@ -1,2 +0,0 @@
-# Tests under this directory should get run "solo" on their own as they
-# seem to cause issues when run in the same batch as other tests.

tests/e2e/multigpu/solo/test_flex.py

@@ -49,9 +49,8 @@ class TestPackedFlex:
                 },
                 "datasets": [
                     {
-                        "path": "tatsu-lab/alpaca",
+                        "path": "vicgalle/alpaca-gpt4",
                         "type": "alpaca",
-                        "split": "train[:10%]",
                     },
                 ],
                 "num_epochs": 1,
@@ -90,5 +89,5 @@ class TestPackedFlex:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.0, "Train loss (%s) is too high"
         )

tests/e2e/multigpu/test_gemma3.py

@@ -96,5 +96,5 @@ class TestMultiGPUGemma3:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 1.8, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 1.8, "Train loss (%s) is too high"
         )

tests/e2e/multigpu/test_llama.py

@@ -43,7 +43,7 @@ class TestMultiGPULlama:
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sequence_len": 2048,
                 "adapter": "lora",
                 "lora_r": 8,
@@ -94,7 +94,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -105,7 +105,7 @@ class TestMultiGPULlama:
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sequence_len": 2048,
                 "sample_packing": True,
                 "eval_sample_packing": False,
@@ -159,14 +159,14 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     def test_dpo_lora_ddp(self, temp_dir):
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sequence_len": 2048,
                 "sample_packing": False,
                 "eval_sample_packing": False,
@@ -244,7 +244,7 @@ class TestMultiGPULlama:
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sequence_len": 2048,
                 "sample_packing": False,
                 "eval_sample_packing": False,
@@ -326,7 +326,7 @@ class TestMultiGPULlama:
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sequence_len": 2048,
                 "val_set_size": 0.01,
                 "special_tokens": {
@@ -385,7 +385,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -396,7 +396,7 @@ class TestMultiGPULlama:
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sample_packing": True,
                 "pad_to_sequence_len": True,
                 "sequence_len": 1024,
@@ -457,7 +457,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     @require_torch_2_6_0
@@ -475,7 +475,7 @@ class TestMultiGPULlama:
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sample_packing": True,
                 "pad_to_sequence_len": True,
                 "sequence_len": 2048,
@@ -538,7 +538,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.1, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.1, "Train loss (%s) is too high"
         )
 
     def test_fsdp_qlora_prequant_packed(self, temp_dir):
@@ -618,7 +618,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -654,7 +654,7 @@ class TestMultiGPULlama:
             adapter = {}
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sample_packing": True,
                 "pad_to_sequence_len": True,
                 "sequence_len": 1024,
@@ -702,7 +702,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -728,7 +728,7 @@ class TestMultiGPULlama:
             adapter = {}
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sample_packing": True,
                 "pad_to_sequence_len": True,
                 "sequence_len": 1024,
@@ -776,7 +776,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -802,7 +802,7 @@ class TestMultiGPULlama:
             adapter = {}
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "sample_packing": True,
                 "pad_to_sequence_len": True,
                 "sequence_len": 1024,
@@ -850,7 +850,7 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     @pytest.mark.skip(
@@ -860,7 +860,7 @@ class TestMultiGPULlama:
         # pylint: disable=duplicate-code
         cfg = DictDefault(
             {
-                "base_model": "HuggingFaceTB/SmolLM2-135M",
+                "base_model": "axolotl-ai-internal/llama-7m",
                 "fix_untrained_tokens": True,
                 "sequence_len": 512,
                 "val_set_size": 0.0,
@@ -917,5 +917,5 @@ class TestMultiGPULlama:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 4.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 4.0, "Train loss (%s) is too high"
         )

tests/e2e/multigpu/test_ray.py

@@ -80,7 +80,7 @@ class TestMultiGPURay:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )
 
     @require_torch_lt_2_6_0
@@ -138,5 +138,5 @@ class TestMultiGPURay:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.3, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.3, "Train loss (%s) is too high"
         )

tests/e2e/multigpu/test_sp.py

@@ -10,7 +10,7 @@ from transformers.testing_utils import get_torch_dist_unique_port
 
 from axolotl.utils.dict import DictDefault
 
-from ...utils import check_tensorboard
+from ..utils import check_tensorboard
 
 os.environ["WANDB_DISABLED"] = "true"
 
@@ -93,7 +93,7 @@ class TestSequenceParallelism:
         )
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.6, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.6, "Train loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(

tests/e2e/patched/test_4d_multipack_llama.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -60,7 +60,6 @@ class Test4dMultipackLlama(unittest.TestCase):
                 "fp16": True,
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -105,7 +104,6 @@ class Test4dMultipackLlama(unittest.TestCase):
                 "fp16": True,
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_fa_xentropy.py

@@ -86,5 +86,5 @@ class TestFAXentropyLlama:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 1.5, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 1.5, "Train loss (%s) is too high"
         )

tests/e2e/patched/test_falcon_samplepack.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -63,7 +63,6 @@ class TestFalconPatched(unittest.TestCase):
                 "bf16": "auto",
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -104,7 +103,6 @@ class TestFalconPatched(unittest.TestCase):
                 "bf16": "auto",
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_fused_llama.py

@@ -12,7 +12,7 @@ from transformers.utils import is_torch_bf16_gpu_available
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -67,7 +67,6 @@ class TestFusedLlama(unittest.TestCase):
             cfg.bf16 = True
         else:
             cfg.fp16 = True
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_llama_s2_attention.py

@@ -11,7 +11,7 @@ import pytest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -65,7 +65,6 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
             }
         )
 
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -106,7 +105,6 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
             }
         )
 
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_lora_llama_multipack.py

@@ -12,7 +12,7 @@ from transformers.utils import is_auto_gptq_available, is_torch_bf16_gpu_availab
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -70,7 +70,6 @@ class TestLoraLlama(unittest.TestCase):
         else:
             cfg.fp16 = True
 
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -121,7 +120,6 @@ class TestLoraLlama(unittest.TestCase):
                 "lr_scheduler": "cosine",
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_mistral_samplepack.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -63,7 +63,6 @@ class TestMistral(unittest.TestCase):
                 "bf16": "auto",
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -105,7 +104,6 @@ class TestMistral(unittest.TestCase):
                 "bf16": "auto",
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_mixtral_samplepack.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -60,7 +60,6 @@ class TestMixtral(unittest.TestCase):
                 "bf16": "auto",
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_model_patches.py

@@ -6,7 +6,7 @@ import unittest
 
 import transformers
 
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 from axolotl.utils.models import load_model, load_tokenizer
 
@@ -47,7 +47,6 @@ class TestModelPatches(unittest.TestCase):
                 "eval_steps": 10,
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         tokenizer = load_tokenizer(cfg)
         load_model(cfg, tokenizer, inference=False)
@@ -80,7 +79,6 @@ class TestModelPatches(unittest.TestCase):
                 "eval_steps": 10,
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         tokenizer = load_tokenizer(cfg)
         load_model(cfg, tokenizer, inference=False)

tests/e2e/patched/test_phi_multipack.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, with_temp_dir
@@ -63,7 +63,6 @@ class TestPhiMultipack(unittest.TestCase):
             }
         )
 
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -83,7 +82,7 @@ class TestPhiMultipack(unittest.TestCase):
                 "sample_packing": True,
                 "flash_attention": True,
                 "pad_to_sequence_len": True,
-                "load_in_4bit": True,
+                "load_in_8bit": False,
                 "adapter": "qlora",
                 "lora_r": 64,
                 "lora_alpha": 32,
@@ -115,7 +114,6 @@ class TestPhiMultipack(unittest.TestCase):
             }
         )
 
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_resume.py

@@ -12,7 +12,7 @@ from transformers.utils import is_torch_bf16_gpu_available
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, most_recent_subdir
@@ -46,9 +46,8 @@ class TestResumeLlama:
                 },
                 "datasets": [
                     {
-                        "path": "tatsu-lab/alpaca",
+                        "path": "vicgalle/alpaca-gpt4",
                         "type": "alpaca",
-                        "split": "train[:10%]",
                     },
                 ],
                 "num_epochs": 2,
@@ -68,7 +67,6 @@ class TestResumeLlama:
             cfg.bf16 = True
         else:
             cfg.fp16 = True
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/patched/test_unsloth_qlora.py

@@ -10,7 +10,7 @@ import pytest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from ..utils import check_model_output_exists, check_tensorboard
@@ -72,7 +72,6 @@ class TestUnslothQLoRA:
             }
         )
 
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -81,7 +80,7 @@ class TestUnslothQLoRA:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.0, "Train loss (%s) is too high"
         )
 
     def test_unsloth_llama_qlora_unpacked(self, temp_dir):
@@ -123,7 +122,6 @@ class TestUnslothQLoRA:
             }
         )
 
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -132,7 +130,7 @@ class TestUnslothQLoRA:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.0, "Train loss (%s) is too high"
         )
 
     @pytest.mark.parametrize(
@@ -179,7 +177,6 @@ class TestUnslothQLoRA:
             }
         )
 
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
@@ -188,5 +185,5 @@ class TestUnslothQLoRA:
         check_model_output_exists(temp_dir, cfg)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.0, "Train loss (%s) is too high"
         )

tests/e2e/solo/test_flex.py

@@ -41,9 +41,8 @@ class TestPackedFlex(unittest.TestCase):
                 },
                 "datasets": [
                     {
-                        "path": "tatsu-lab/alpaca",
+                        "path": "vicgalle/alpaca-gpt4",
                         "type": "alpaca",
-                        "split": "train[:10%]",
                     },
                 ],
                 "num_epochs": 1,
@@ -70,5 +69,5 @@ class TestPackedFlex(unittest.TestCase):
         train(cfg=cfg, dataset_meta=dataset_meta)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.0, "Train loss (%s) is too high"
         )

tests/e2e/test_embeddings_lr.py

@@ -102,7 +102,6 @@ class TestEmbeddingsLrScale(unittest.TestCase):
                 "use_tensorboard": True,
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/test_llama_pretrain.py

@@ -84,5 +84,5 @@ class TestPretrainLlama:
             temp_dir + "/runs",
             "train/train_loss",
             loss_threshold,
-            "Train Loss is too high",
+            "Train Loss (%s) is too high",
         )

tests/e2e/test_llama_vision.py

@@ -109,7 +109,6 @@ class TestLlamaVision(unittest.TestCase):
                 "bf16": True,
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/test_packing_loss.py

@@ -40,9 +40,8 @@ class TestPackedLlama(unittest.TestCase):
                 },
                 "datasets": [
                     {
-                        "path": "tatsu-lab/alpaca",
+                        "path": "vicgalle/alpaca-gpt4",
                         "type": "alpaca",
-                        "split": "train[:10%]",
                     },
                 ],
                 "num_epochs": 1,
@@ -69,5 +68,5 @@ class TestPackedLlama(unittest.TestCase):
         train(cfg=cfg, dataset_meta=dataset_meta)
 
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.0, "Train loss (%s) is too high"
         )

tests/e2e/test_phi.py

@@ -79,7 +79,7 @@ class TestPhi(unittest.TestCase):
                 "tokenizer_type": "AutoTokenizer",
                 "sequence_len": 2048,
                 "sample_packing": False,
-                "load_in_4bit": True,
+                "load_in_8bit": False,
                 "adapter": "qlora",
                 "lora_r": 64,
                 "lora_alpha": 32,
@@ -111,7 +111,6 @@ class TestPhi(unittest.TestCase):
                 "bf16": "auto",
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/test_process_reward_model_smollm2.py

@@ -9,7 +9,7 @@ import unittest
 from axolotl.cli.args import TrainerCliArgs
 from axolotl.common.datasets import load_datasets
 from axolotl.train import train
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.dict import DictDefault
 
 from .utils import check_model_output_exists, check_tensorboard, with_temp_dir
@@ -57,7 +57,6 @@ class TestProcessRewardSmolLM2(unittest.TestCase):
                 "seed": 42,
             }
         )
-        cfg = validate_config(cfg)
         normalize_config(cfg)
         cli_args = TrainerCliArgs()
         dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)

tests/e2e/test_reward_model_smollm2.py

@@ -73,6 +73,6 @@ class TestRewardModelLoraSmolLM2(unittest.TestCase):
 
         train(cfg=cfg, dataset_meta=dataset_meta)
         check_tensorboard(
-            temp_dir + "/runs", "train/train_loss", 2.5, "Train Loss is too high"
+            temp_dir + "/runs", "train/train_loss", 2.5, "Train loss (%s) is too high"
         )
         check_model_output_exists(temp_dir, cfg)

tests/test_exact_deduplication.py

@@ -11,7 +11,7 @@ from unittest.mock import patch
 import pytest
 from datasets import Dataset
 
-from axolotl.utils.config import normalize_config, validate_config
+from axolotl.utils.config import normalize_config
 from axolotl.utils.data import prepare_dataset
 from axolotl.utils.data.rl import load_prepare_preference_datasets
 from axolotl.utils.data.utils import deduplicate_and_log_datasets
@@ -319,7 +319,6 @@ class TestDeduplicateNonRL(unittest.TestCase):
                 "num_epochs": 1,
             }
         )
-        self.cfg_1 = validate_config(self.cfg_1)
         normalize_config(self.cfg_1)
 
     @pytest.mark.skip(reason="TODO: fix hf hub offline to work with HF rate limits")

tests/test_perplexity.py

@@ -8,7 +8,7 @@ from transformers.models.auto.tokenization_auto import AutoTokenizer
 
 from axolotl.utils.callbacks.perplexity import Perplexity
 
-MODEL_NAME = "HuggingFaceTB/SmolLM2-135M"
+MODEL_NAME = "axolotl-ai-internal/llama-7m"
 
 
 @fixture()
@@ -36,7 +36,7 @@ One day, a little fish named Fin was swimming near the shore. He saw a big crab
 """
     result = metric.compute(model, [sample_text])
     ppl = result["score"]
-    assert round(ppl, 2) == 7.41
+    assert round(ppl, 2) == 75.14
 
 
 def test_perplexity_short(model, metric):
@@ -44,4 +44,4 @@ def test_perplexity_short(model, metric):
     sample_text = "Once upon a time, there was a little car named Beep. Beep loved to go fast and play in the sun."
     result = metric.compute(model, [sample_text])
     ppl = result["score"]
-    assert round(ppl, 2) == 10.33
+    assert round(ppl, 2) == 70.54