diff --git a/.nojekyll b/.nojekyll index afc349f6e..66c2f2e33 100644 --- a/.nojekyll +++ b/.nojekyll @@ -1 +1 @@ -ed280e0b \ No newline at end of file +3f6621ed \ No newline at end of file diff --git a/FAQS.html b/FAQS.html index a7903b7b4..616640792 100644 --- a/FAQS.html +++ b/FAQS.html @@ -210,6 +210,12 @@ ul.task-list li input[type="checkbox"] { Conversation + +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/TODO.html b/TODO.html index a6d9ab653..83b76b3f6 100644 --- a/TODO.html +++ b/TODO.html @@ -210,6 +210,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/amd_hpc.html b/docs/amd_hpc.html index a62226703..4a14f3a9d 100644 --- a/docs/amd_hpc.html +++ b/docs/amd_hpc.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/batch_vs_grad.html b/docs/batch_vs_grad.html index a7f1e52f2..c479e7201 100644 --- a/docs/batch_vs_grad.html +++ b/docs/batch_vs_grad.html @@ -211,6 +211,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/config.html b/docs/config.html index f3b29edcc..2e66a2c80 100644 --- a/docs/config.html +++ b/docs/config.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • @@ -501,352 +507,358 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin # whether to perform weighting if doing DPO training. Boolean. dpo_use_weighting: -# The name of the chat template to use for training, following values are supported: -# - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default value. -# - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py -# - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not available in the tokenizer. -# - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field. -# The selected chat template will be saved to the tokenizer_config.json for easier inferencing -# Note: It is recommended to set train_on_inputs to true when using a chat template that is different from the model's default chat template. -chat_template: tokenizer_default -# custom jinja template for chat template. This will be only used if chat_template is set to `jinja` or `null` (in which case chat_template is automatically set to `jinja`). Default is null. -chat_template_jinja: null -# Changes the default system message -default_system_message: You are a helpful assistant. Please give a long and detailed answer. # Currently only supports chatml. -# Axolotl attempts to save the dataset as an arrow after packing the data together so -# subsequent training attempts load faster, relative path -dataset_prepared_path: data/last_run_prepared -# Push prepared dataset to hub -push_dataset_to_hub: # repo path -# The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()` -# if not set. -dataset_processes: # defaults to os.cpu_count() if not set -# Keep dataset in memory while preprocessing -# Only needed if cached dataset is taking too much storage -dataset_keep_in_memory: -# push checkpoints to hub -hub_model_id: # private repo path to push finetuned model -# how to push checkpoints to hub -# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy -hub_strategy: -# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets -# Required to be true when used in combination with `push_dataset_to_hub` -hf_use_auth_token: # boolean -# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval. -val_set_size: 0.04 -# Num shards for whole dataset -dataset_shard_num: -# Index of shard to use for whole dataset -dataset_shard_idx: - -# The maximum length of an input to train with, this should typically be less than 2048 -# as most models have a token/context limit of 2048 -sequence_len: 2048 -# Pad inputs so each step uses constant sized buffers -# This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently -pad_to_sequence_len: -# Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true' -sample_packing: -# Set to 'false' if getting errors during eval with sample_packing on. -eval_sample_packing: -# You can set these packing optimizations AFTER starting a training at least once. -# The trainer will provide recommended values for these values. -sample_packing_eff_est: -total_num_tokens: -# Increasing the following values helps with packing, but usually only slightly (<%1.) -# The number of samples packed at a time. -sample_packing_group_size: 100000 -# The number of samples which can be packed into one sequence. Increase if using a large sequence_len with many short samples. -sample_packing_bin_size: 200 -# whether to concatenate samples during pretraining -pretraining_sample_concatenation: - -# Use batch flattening for speedups when not using sample_packing -batch_flattening: - -# Passed through to transformers when loading the model when launched without accelerate -# Use `sequential` when training w/ model parallelism to limit memory -device_map: -# Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model. -max_memory: +# reward modelling: `True` or `False` +reward_model: + +# process reward modelling: `True` or `False` +process_reward_model: + +# The name of the chat template to use for training, following values are supported: +# - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default value. +# - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py +# - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not available in the tokenizer. +# - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field. +# The selected chat template will be saved to the tokenizer_config.json for easier inferencing +# Note: It is recommended to set train_on_inputs to true when using a chat template that is different from the model's default chat template. +chat_template: tokenizer_default +# custom jinja template for chat template. This will be only used if chat_template is set to `jinja` or `null` (in which case chat_template is automatically set to `jinja`). Default is null. +chat_template_jinja: null +# Changes the default system message +default_system_message: You are a helpful assistant. Please give a long and detailed answer. # Currently only supports chatml. +# Axolotl attempts to save the dataset as an arrow after packing the data together so +# subsequent training attempts load faster, relative path +dataset_prepared_path: data/last_run_prepared +# Push prepared dataset to hub +push_dataset_to_hub: # repo path +# The maximum number of processes to use while preprocessing your input dataset. This defaults to `os.cpu_count()` +# if not set. +dataset_processes: # defaults to os.cpu_count() if not set +# Keep dataset in memory while preprocessing +# Only needed if cached dataset is taking too much storage +dataset_keep_in_memory: +# push checkpoints to hub +hub_model_id: # private repo path to push finetuned model +# how to push checkpoints to hub +# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy +hub_strategy: +# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets +# Required to be true when used in combination with `push_dataset_to_hub` +hf_use_auth_token: # boolean +# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc. 0 for no eval. +val_set_size: 0.04 +# Num shards for whole dataset +dataset_shard_num: +# Index of shard to use for whole dataset +dataset_shard_idx: + +# The maximum length of an input to train with, this should typically be less than 2048 +# as most models have a token/context limit of 2048 +sequence_len: 2048 +# Pad inputs so each step uses constant sized buffers +# This will reduce memory fragmentation and may prevent OOMs, by re-using memory more efficiently +pad_to_sequence_len: +# Use efficient multi-packing with block diagonal attention and per sequence position_ids. Recommend set to 'true' +sample_packing: +# Set to 'false' if getting errors during eval with sample_packing on. +eval_sample_packing: +# You can set these packing optimizations AFTER starting a training at least once. +# The trainer will provide recommended values for these values. +sample_packing_eff_est: +total_num_tokens: +# Increasing the following values helps with packing, but usually only slightly (<%1.) +# The number of samples packed at a time. +sample_packing_group_size: 100000 +# The number of samples which can be packed into one sequence. Increase if using a large sequence_len with many short samples. +sample_packing_bin_size: 200 +# whether to concatenate samples during pretraining +pretraining_sample_concatenation: + +# Use batch flattening for speedups when not using sample_packing +batch_flattening: -# If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model -adapter: lora -# If you already have a lora model trained that you want to load, put that here. -# This means after training, if you want to test the model, you should set this to the value of `output_dir`. -# Note that if you merge an adapter to the base model, a new subdirectory `merged` will be created under the `output_dir`. -lora_model_dir: - -# LoRA hyperparameters -# For more details about the following options, see: -# https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2 -lora_r: 8 -lora_alpha: 16 -lora_dropout: 0.05 -lora_target_modules: - - q_proj - - v_proj -# - k_proj -# - o_proj -# - gate_proj -# - down_proj -# - up_proj -lora_target_linear: # If true, will target all linear modules -peft_layers_to_transform: # The layer indices to transform, otherwise, apply to all layers - -# If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens. -# For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models. -# `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities. -# https://github.com/huggingface/peft/issues/334#issuecomment-1561727994 -lora_modules_to_save: -# - embed_tokens -# - lm_head - -lora_fan_in_fan_out: false - -# LoRA+ hyperparameters -# For more details about the following options, see: -# https://arxiv.org/abs/2402.12354 and `src/axolotl/core/train_builder.py` -loraplus_lr_ratio: # loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4. -loraplus_lr_embedding: # loraplus learning rate for lora embedding layers. Default value is 1e-6. +# Passed through to transformers when loading the model when launched without accelerate +# Use `sequential` when training w/ model parallelism to limit memory +device_map: +# Defines the max memory usage per gpu on the system. Passed through to transformers when loading the model. +max_memory: + +# If you want to use 'lora' or 'qlora' or leave blank to train all parameters in original model +adapter: lora +# If you already have a lora model trained that you want to load, put that here. +# This means after training, if you want to test the model, you should set this to the value of `output_dir`. +# Note that if you merge an adapter to the base model, a new subdirectory `merged` will be created under the `output_dir`. +lora_model_dir: + +# LoRA hyperparameters +# For more details about the following options, see: +# https://www.anyscale.com/blog/fine-tuning-llms-lora-or-full-parameter-an-in-depth-analysis-with-llama-2 +lora_r: 8 +lora_alpha: 16 +lora_dropout: 0.05 +lora_target_modules: + - q_proj + - v_proj +# - k_proj +# - o_proj +# - gate_proj +# - down_proj +# - up_proj +lora_target_linear: # If true, will target all linear modules +peft_layers_to_transform: # The layer indices to transform, otherwise, apply to all layers + +# If you added new tokens to the tokenizer, you may need to save some LoRA modules because they need to know the new tokens. +# For LLaMA and Mistral, you need to save `embed_tokens` and `lm_head`. It may vary for other models. +# `embed_tokens` converts tokens to embeddings, and `lm_head` converts embeddings to token probabilities. +# https://github.com/huggingface/peft/issues/334#issuecomment-1561727994 +lora_modules_to_save: +# - embed_tokens +# - lm_head + +lora_fan_in_fan_out: false -peft: - # Configuration options for loftq initialization for LoRA - # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization - loftq_config: - loftq_bits: # typically 4 bits +# LoRA+ hyperparameters +# For more details about the following options, see: +# https://arxiv.org/abs/2402.12354 and `src/axolotl/core/train_builder.py` +loraplus_lr_ratio: # loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4. +loraplus_lr_embedding: # loraplus learning rate for lora embedding layers. Default value is 1e-6. -# ReLoRA configuration -# Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed -relora_steps: # Number of steps per ReLoRA restart -relora_warmup_steps: # Number of per-restart warmup steps -relora_anneal_steps: # Number of anneal steps for each relora cycle -relora_prune_ratio: # threshold for optimizer magnitude when pruning -relora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings - -# wandb configuration if you're using it -# Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`. -wandb_mode: # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb -wandb_project: # Your wandb project name -wandb_entity: # A wandb Team name if using a Team -wandb_watch: -wandb_name: # Set the name of your wandb run -wandb_run_id: # Set the ID of your wandb run -wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training - -# mlflow configuration if you're using it -mlflow_tracking_uri: # URI to mlflow -mlflow_experiment_name: # Your experiment name -mlflow_run_name: # Your run name -hf_mlflow_log_artifacts: # set to true to copy each saved checkpoint on each save to mlflow artifact registry +peft: + # Configuration options for loftq initialization for LoRA + # https://huggingface.co/docs/peft/developer_guides/quantization#loftq-initialization + loftq_config: + loftq_bits: # typically 4 bits + +# ReLoRA configuration +# Must use either 'lora' or 'qlora' adapter, and does not support fsdp or deepspeed +relora_steps: # Number of steps per ReLoRA restart +relora_warmup_steps: # Number of per-restart warmup steps +relora_anneal_steps: # Number of anneal steps for each relora cycle +relora_prune_ratio: # threshold for optimizer magnitude when pruning +relora_cpu_offload: # True to perform lora weight merges on cpu during restarts, for modest gpu memory savings + +# wandb configuration if you're using it +# Make sure your `WANDB_API_KEY` environment variable is set (recommended) or you login to wandb with `wandb login`. +wandb_mode: # "offline" to save run metadata locally and not sync to the server, "disabled" to turn off wandb +wandb_project: # Your wandb project name +wandb_entity: # A wandb Team name if using a Team +wandb_watch: +wandb_name: # Set the name of your wandb run +wandb_run_id: # Set the ID of your wandb run +wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training -# Comet configuration if you're using it -# Make sure your `COMET_API_KEY` environment variable is set (recommended) or you login to Comet with `comet login`. -# Check out our documentation for more details https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/Experiment-Creation/#comet_ml.start -use_comet: # Enable or disable Comet integration. -comet_api_key: # API key for Comet. Recommended to set via `comet login`. -comet_workspace: # Workspace name in Comet. Defaults to the user's default workspace. -comet_project_name: # Project name in Comet. Defaults to Uncategorized. -comet_experiment_key: # Identifier for the experiment. Used to append data to an existing experiment or control the key of new experiments. Default to a random key. -comet_mode: # Create a new experiment ("create") or log to an existing one ("get"). Default ("get_or_create") auto-selects based on configuration. -comet_online: # Set to True to log data to Comet server, or False for offline storage. Default is True. -comet_experiment_config: # Dictionary for additional configuration settings, see the doc for more details. - -# Where to save the full-finetuned model to -output_dir: ./completed-model - -# Whether to use torch.compile and which backend to use -# setting to `auto` will enable torch compile when torch>=2.5.1 -torch_compile: # Optional[Union[Literal["auto"], bool]] -torch_compile_backend: # Optional[str] - -# Training hyperparameters - -# If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps. -gradient_accumulation_steps: 1 -# The number of samples to include in each batch. This is the number of samples sent to each GPU. -# Batch size per gpu = micro_batch_size * gradient_accumulation_steps -micro_batch_size: 2 -eval_batch_size: -num_epochs: 4 -warmup_steps: 100 # cannot use with warmup_ratio -warmup_ratio: 0.05 # cannot use with warmup_steps -learning_rate: 0.00003 -lr_quadratic_warmup: -logging_steps: -eval_steps: # Leave empty to eval at each epoch, integer for every N steps. float for fraction of total steps -evals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps -eval_strategy: # Set to `"no"` to skip evaluation, `"epoch"` at end of each epoch, leave empty to infer from `eval_steps`. -save_strategy: # Set to `"no"` to skip checkpoint saves, `"epoch"` at end of each epoch, `"best"` when better result is achieved, leave empty to infer from `save_steps`. -save_steps: # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps -saves_per_epoch: # number of times per epoch to save a checkpoint, mutually exclusive with save_steps -save_total_limit: # Checkpoints saved at a time -# Maximum number of iterations to train for. It precedes num_epochs which means that -# if both are set, num_epochs will not be guaranteed. -# e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps -max_steps: - -eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0 -eval_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128 -eval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is ["sacrebleu", "comet", "ter", "chrf", "perplexity"] - -profiler_steps: # enable the pytorch profiler to capture the first N steps of training to the output_dir. - # see https://pytorch.org/blog/understanding-gpu-memory-1/ for more information - # snapshots can be visualized @ https://pytorch.org/memory_viz - -loss_watchdog_threshold: # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training) -loss_watchdog_patience: # Number of high-loss steps in a row before the trainer aborts (default: 3) - -# Save model as safetensors (require safetensors package) -save_safetensors: +# mlflow configuration if you're using it +mlflow_tracking_uri: # URI to mlflow +mlflow_experiment_name: # Your experiment name +mlflow_run_name: # Your run name +hf_mlflow_log_artifacts: # set to true to copy each saved checkpoint on each save to mlflow artifact registry + +# Comet configuration if you're using it +# Make sure your `COMET_API_KEY` environment variable is set (recommended) or you login to Comet with `comet login`. +# Check out our documentation for more details https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/Experiment-Creation/#comet_ml.start +use_comet: # Enable or disable Comet integration. +comet_api_key: # API key for Comet. Recommended to set via `comet login`. +comet_workspace: # Workspace name in Comet. Defaults to the user's default workspace. +comet_project_name: # Project name in Comet. Defaults to Uncategorized. +comet_experiment_key: # Identifier for the experiment. Used to append data to an existing experiment or control the key of new experiments. Default to a random key. +comet_mode: # Create a new experiment ("create") or log to an existing one ("get"). Default ("get_or_create") auto-selects based on configuration. +comet_online: # Set to True to log data to Comet server, or False for offline storage. Default is True. +comet_experiment_config: # Dictionary for additional configuration settings, see the doc for more details. + +# Where to save the full-finetuned model to +output_dir: ./completed-model + +# Whether to use torch.compile and which backend to use +# setting to `auto` will enable torch compile when torch>=2.5.1 +torch_compile: # Optional[Union[Literal["auto"], bool]] +torch_compile_backend: # Optional[str] + +# Training hyperparameters + +# If greater than 1, backpropagation will be skipped and the gradients will be accumulated for the given number of steps. +gradient_accumulation_steps: 1 +# The number of samples to include in each batch. This is the number of samples sent to each GPU. +# Batch size per gpu = micro_batch_size * gradient_accumulation_steps +micro_batch_size: 2 +eval_batch_size: +num_epochs: 4 +warmup_steps: 100 # cannot use with warmup_ratio +warmup_ratio: 0.05 # cannot use with warmup_steps +learning_rate: 0.00003 +lr_quadratic_warmup: +logging_steps: +eval_steps: # Leave empty to eval at each epoch, integer for every N steps. float for fraction of total steps +evals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps +eval_strategy: # Set to `"no"` to skip evaluation, `"epoch"` at end of each epoch, leave empty to infer from `eval_steps`. +save_strategy: # Set to `"no"` to skip checkpoint saves, `"epoch"` at end of each epoch, `"best"` when better result is achieved, leave empty to infer from `save_steps`. +save_steps: # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps +saves_per_epoch: # number of times per epoch to save a checkpoint, mutually exclusive with save_steps +save_total_limit: # Checkpoints saved at a time +# Maximum number of iterations to train for. It precedes num_epochs which means that +# if both are set, num_epochs will not be guaranteed. +# e.g., when 1 epoch is 1000 steps => `num_epochs: 2` and `max_steps: 100` will train for 100 steps +max_steps: + +eval_table_size: # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0 +eval_max_new_tokens: # Total number of tokens generated for predictions sent to wandb. Default is 128 +eval_causal_lm_metrics: # HF evaluate metrics used during evaluation. Default is ["sacrebleu", "comet", "ter", "chrf", "perplexity"] + +profiler_steps: # enable the pytorch profiler to capture the first N steps of training to the output_dir. + # see https://pytorch.org/blog/understanding-gpu-memory-1/ for more information + # snapshots can be visualized @ https://pytorch.org/memory_viz -# Whether to mask out or include the human's prompt from the training labels -train_on_inputs: false -# Group similarly sized data to minimize padding. -# May be slower to start, as it must download and sort the entire dataset. -# Note that training loss may have an oscillating pattern with this enabled. -group_by_length: false - -# Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing -gradient_checkpointing: false -# additional kwargs to pass to the trainer for gradient checkpointing -# gradient_checkpointing_kwargs: -# use_reentrant: true +loss_watchdog_threshold: # High loss value, indicating the learning has broken down (a good estimate is ~2 times the loss at the start of training) +loss_watchdog_patience: # Number of high-loss steps in a row before the trainer aborts (default: 3) + +# Save model as safetensors (require safetensors package) +save_safetensors: + +# Whether to mask out or include the human's prompt from the training labels +train_on_inputs: false +# Group similarly sized data to minimize padding. +# May be slower to start, as it must download and sort the entire dataset. +# Note that training loss may have an oscillating pattern with this enabled. +group_by_length: false -# Stop training after this many evaluation losses have increased in a row -# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback -early_stopping_patience: 3 - -# Specify a scheduler and kwargs to use with the optimizer -lr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine -lr_scheduler_kwargs: -cosine_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 -cosine_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) +# Whether to use gradient checkpointing https://huggingface.co/docs/transformers/v4.18.0/en/performance#gradient-checkpointing +gradient_checkpointing: false +# additional kwargs to pass to the trainer for gradient checkpointing +# gradient_checkpointing_kwargs: +# use_reentrant: true + +# Stop training after this many evaluation losses have increased in a row +# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback +early_stopping_patience: 3 -# For one_cycle optim -lr_div_factor: # Learning rate div factor - -# Specify optimizer -# Valid values are driven by the Transformers OptimizerNames class, see: -# https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134 -# -# Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of -# torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used -# in the examples/ for your model and fine-tuning use case. -# -# Valid values for 'optimizer' include: -# - adamw_hf -# - adamw_torch -# - adamw_torch_fused -# - adamw_torch_xla -# - adamw_apex_fused -# - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version >= 2.5.1) -# - adafactor -# - adamw_anyprecision -# - sgd -# - adagrad -# - adamw_bnb_8bit -# - lion_8bit -# - lion_32bit -# - paged_adamw_32bit -# - paged_adamw_8bit -# - paged_lion_32bit -# - paged_lion_8bit -# - galore_adamw -# - galore_adamw_8bit -# - galore_adafactor -# - galore_adamw_layerwise -# - galore_adamw_8bit_layerwise -# - galore_adafactor_layerwise -optimizer: -# Dictionary of arguments to pass to the optimizer -optim_args: -# For Galore Optimizers the following optim_args are available -# rank: # type: int -# update_proj_gap # type: int -# scale # type: float -# proj_type: # type: str, default = std - -# 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 -optim_target_modules: -# - self_attn # for llama -# - mlp - -# Specify weight decay -weight_decay: -# adamw hyperparams -adam_beta1: -adam_beta2: -adam_epsilon: -# Gradient clipping max norm -max_grad_norm: - -# Augmentation techniques -# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings -# currently only supported on Llama and Mistral -neftune_noise_alpha: - -# Whether to bettertransformers -flash_optimum: -# Whether to use xformers attention patch https://github.com/facebookresearch/xformers: -xformers_attention: -# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention: -flash_attention: -flash_attn_cross_entropy: # Whether to use flash-attention cross entropy implementation - advanced use only -flash_attn_rms_norm: # Whether to use flash-attention rms norm implementation - advanced use only -flash_attn_fuse_qkv: # Whether to fuse QKV into a single operation -flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation -# Whether to use scaled-dot-product attention -# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html -sdp_attention: -# Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf -s2_attention: -# Resume from a specific checkpoint dir -resume_from_checkpoint: -# If resume_from_checkpoint isn't set and you simply want it to start where it left off. -# Be careful with this being turned on between different models. -auto_resume_from_checkpoints: false - -# Don't mess with this, it's here for accelerate and torchrun -local_rank: - -# Add or change special tokens. -# If you add tokens here, you don't need to add them to the `tokens` list. -special_tokens: - # bos_token: "<s>" - # eos_token: "</s>" - # unk_token: "<unk>" - # pad_token: "[PAD]" - -# Add extra tokens. -tokens: - -# FSDP -fsdp: -fsdp_config: - -# Deepspeed config path. e.g., deepspeed_configs/zero3.json -deepspeed: - -# Advanced DDP Arguments -ddp_timeout: -ddp_bucket_cap_mb: -ddp_broadcast_buffers: - -# Path to torch distx for optim 'adamw_anyprecision' -torchdistx_path: - -# Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize -pretraining_dataset: +# Specify a scheduler and kwargs to use with the optimizer +lr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine +lr_scheduler_kwargs: +cosine_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 +cosine_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) + +# For one_cycle optim +lr_div_factor: # Learning rate div factor + +# Specify optimizer +# Valid values are driven by the Transformers OptimizerNames class, see: +# https://github.com/huggingface/transformers/blob/95b374952dc27d8511541d6f5a4e22c9ec11fb24/src/transformers/training_args.py#L134 +# +# Note that not all optimizers may be available in your environment, ex: 'adamw_anyprecision' is part of +# torchdistx, 'adamw_bnb_8bit' is part of bnb.optim.Adam8bit, etc. When in doubt, it is recommended to start with the optimizer used +# in the examples/ for your model and fine-tuning use case. +# +# Valid values for 'optimizer' include: +# - adamw_hf +# - adamw_torch +# - adamw_torch_fused +# - adamw_torch_xla +# - adamw_apex_fused +# - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version >= 2.5.1) +# - adafactor +# - adamw_anyprecision +# - sgd +# - adagrad +# - adamw_bnb_8bit +# - lion_8bit +# - lion_32bit +# - paged_adamw_32bit +# - paged_adamw_8bit +# - paged_lion_32bit +# - paged_lion_8bit +# - galore_adamw +# - galore_adamw_8bit +# - galore_adafactor +# - galore_adamw_layerwise +# - galore_adamw_8bit_layerwise +# - galore_adafactor_layerwise +optimizer: +# Dictionary of arguments to pass to the optimizer +optim_args: +# For Galore Optimizers the following optim_args are available +# rank: # type: int +# update_proj_gap # type: int +# scale # type: float +# proj_type: # type: str, default = std + +# 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 +optim_target_modules: +# - self_attn # for llama +# - mlp + +# Specify weight decay +weight_decay: +# adamw hyperparams +adam_beta1: +adam_beta2: +adam_epsilon: +# Gradient clipping max norm +max_grad_norm: + +# Augmentation techniques +# NEFT https://arxiv.org/abs/2310.05914, set this to a number (paper default is 5) to add noise to embeddings +# currently only supported on Llama and Mistral +neftune_noise_alpha: + +# Whether to bettertransformers +flash_optimum: +# Whether to use xformers attention patch https://github.com/facebookresearch/xformers: +xformers_attention: +# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention: +flash_attention: +flash_attn_cross_entropy: # Whether to use flash-attention cross entropy implementation - advanced use only +flash_attn_rms_norm: # Whether to use flash-attention rms norm implementation - advanced use only +flash_attn_fuse_qkv: # Whether to fuse QKV into a single operation +flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation +# Whether to use scaled-dot-product attention +# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html +sdp_attention: +# Shifted-sparse attention (only llama) - https://arxiv.org/pdf/2309.12307.pdf +s2_attention: +# Resume from a specific checkpoint dir +resume_from_checkpoint: +# If resume_from_checkpoint isn't set and you simply want it to start where it left off. +# Be careful with this being turned on between different models. +auto_resume_from_checkpoints: false + +# Don't mess with this, it's here for accelerate and torchrun +local_rank: + +# Add or change special tokens. +# If you add tokens here, you don't need to add them to the `tokens` list. +special_tokens: + # bos_token: "<s>" + # eos_token: "</s>" + # unk_token: "<unk>" + # pad_token: "[PAD]" + +# Add extra tokens. +tokens: + +# FSDP +fsdp: +fsdp_config: + +# Deepspeed config path. e.g., deepspeed_configs/zero3.json +deepspeed: + +# Advanced DDP Arguments +ddp_timeout: +ddp_bucket_cap_mb: +ddp_broadcast_buffers: -# Debug mode -debug: +# Path to torch distx for optim 'adamw_anyprecision' +torchdistx_path: -# Seed -seed: +# Set to HF dataset for type: 'completion' for streaming instead of pre-tokenize +pretraining_dataset: -# Allow overwrite yml config using from cli -strict:
    +# Debug mode +debug: + +# Seed +seed: + +# Allow overwrite yml config using from cli +strict: diff --git a/docs/dataset-formats/conversation.html b/docs/dataset-formats/conversation.html index 124308b92..e7754df51 100644 --- a/docs/dataset-formats/conversation.html +++ b/docs/dataset-formats/conversation.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/dataset-formats/index.html b/docs/dataset-formats/index.html index c6781595a..42f3ffd5b 100644 --- a/docs/dataset-formats/index.html +++ b/docs/dataset-formats/index.html @@ -271,6 +271,12 @@ window.Quarto = { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • @@ -363,7 +369,7 @@ Description - + Pre-training @@ -371,7 +377,7 @@ Description Data format for a pre-training completion task. - + Instruction Tuning @@ -379,7 +385,7 @@ Description Instruction tuning formats for supervised fine-tuning. - + Conversation @@ -387,7 +393,15 @@ Description Conversation format for supervised fine-tuning. - + + +Stepwise Supervised Format + + +Format for datasets with stepwise completions and labels + + + Template-Free @@ -395,7 +409,7 @@ Description Construct prompts without a template. - + Custom Pre-Tokenized Dataset diff --git a/docs/dataset-formats/inst_tune.html b/docs/dataset-formats/inst_tune.html index f5d8246f9..7552abda8 100644 --- a/docs/dataset-formats/inst_tune.html +++ b/docs/dataset-formats/inst_tune.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/dataset-formats/pretraining.html b/docs/dataset-formats/pretraining.html index d35da48bd..85a03915a 100644 --- a/docs/dataset-formats/pretraining.html +++ b/docs/dataset-formats/pretraining.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/dataset-formats/stepwise_supervised.html b/docs/dataset-formats/stepwise_supervised.html new file mode 100644 index 000000000..a3fa20c28 --- /dev/null +++ b/docs/dataset-formats/stepwise_supervised.html @@ -0,0 +1,729 @@ + + + + + + + + + + +Stepwise Supervised Format – Axolotl + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    +
    + + +
    + +
    + + +
    + + + +
    + +
    +
    +

    Stepwise Supervised Format

    +
    + +
    +
    + Format for datasets with stepwise completions and labels +
    +
    + + +
    + + + + +
    + + + +
    + + +
    +

    Stepwise Supervised

    +

    The stepwise supervised format is designed for chain-of-thought (COT) reasoning datasets where each example contains multiple completion steps and a preference label for each step. ### ExampleHere’s a simple example of a stepwise supervised dataset entry:```json { “prompt”: “Which number is larger, 9.8 or 9.11?”, “completions”: [ “The fractional part of 9.8 is 0.8, while the fractional part of 9.11 is 0.11.”, “Since 0.11 is greater than 0.8, the number 9.11 is larger than 9.8.” ], “labels”: [true, false] }

    + + +
    + +
    + +
    + + + + + \ No newline at end of file diff --git a/docs/dataset-formats/template_free.html b/docs/dataset-formats/template_free.html index f7ba80fad..28a3f0bd8 100644 --- a/docs/dataset-formats/template_free.html +++ b/docs/dataset-formats/template_free.html @@ -211,6 +211,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/dataset-formats/tokenized.html b/docs/dataset-formats/tokenized.html index def534b7e..aee0578ea 100644 --- a/docs/dataset-formats/tokenized.html +++ b/docs/dataset-formats/tokenized.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/dataset_preprocessing.html b/docs/dataset_preprocessing.html index 51823561e..9488f4616 100644 --- a/docs/dataset_preprocessing.html +++ b/docs/dataset_preprocessing.html @@ -211,6 +211,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/debugging.html b/docs/debugging.html index 3159b3ac1..27e08979a 100644 --- a/docs/debugging.html +++ b/docs/debugging.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/faq.html b/docs/faq.html index dac2ffc39..fee0770c3 100644 --- a/docs/faq.html +++ b/docs/faq.html @@ -211,6 +211,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/fsdp_qlora.html b/docs/fsdp_qlora.html index 2242f466d..5efe81c74 100644 --- a/docs/fsdp_qlora.html +++ b/docs/fsdp_qlora.html @@ -211,6 +211,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/input_output.html b/docs/input_output.html index 41fd0eb41..492542ac5 100644 --- a/docs/input_output.html +++ b/docs/input_output.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/lr_groups.html b/docs/lr_groups.html index 1684b6e0d..06c3e104e 100644 --- a/docs/lr_groups.html +++ b/docs/lr_groups.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/mac.html b/docs/mac.html index fee30038a..b3ffd3f11 100644 --- a/docs/mac.html +++ b/docs/mac.html @@ -211,6 +211,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/multi-node.html b/docs/multi-node.html index e1d954792..a373ea3f8 100644 --- a/docs/multi-node.html +++ b/docs/multi-node.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/multimodal.html b/docs/multimodal.html index a02f4ab5d..78338600b 100644 --- a/docs/multimodal.html +++ b/docs/multimodal.html @@ -244,6 +244,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/multipack.html b/docs/multipack.html index e5af96ff4..0464065fb 100644 --- a/docs/multipack.html +++ b/docs/multipack.html @@ -211,6 +211,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/nccl.html b/docs/nccl.html index 9e994e907..2a31d4dff 100644 --- a/docs/nccl.html +++ b/docs/nccl.html @@ -211,6 +211,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/reward_modelling.html b/docs/reward_modelling.html new file mode 100644 index 000000000..106ee2d6b --- /dev/null +++ b/docs/reward_modelling.html @@ -0,0 +1,798 @@ + + + + + + + + + + +Reward Modelling – Axolotl + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    +
    + + +
    + +
    + + +
    + + + +
    + +
    +
    +

    Reward Modelling

    +
    + +
    +
    + Reward models are used to guide models towards behaviors which is preferred by humans, by training over large datasets annotated with human preferences. +
    +
    + + +
    + + + + +
    + + + +
    + + +
    +

    Overview

    +

    Reward modelling is a technique used to train models to predict the reward or value of a given input. This is particularly useful in reinforcement learning scenarios where the model needs to evaluate the quality of its actions or predictions. We support the reward modelling techniques supported by trl.

    +
    +
    +

    (Outcome) Reward Models

    +

    Outcome reward models are trained using data which contains preference annotations for an entire interaction between the user and model (e.g. rather than per-turn or per-step).

    +
    base_model: google/gemma-2-2b
+model_type: AutoModelForSequenceClassification
+num_labels: 1
+tokenizer_type: AutoTokenizer
+
+reward_model: true
+chat_template: gemma
+datasets:
+  - path: argilla/distilabel-intel-orca-dpo-pairs
+    type: bradley_terry.chat_template
+
+val_set_size: 0.1
+eval_steps: 100
    +
    +
    +

    Process Reward Models (PRM)

    +

    Process reward models are trained using data which contains preference annotations for each step in a series of interactions. Typically, PRMs are trained to provide reward signals over each step of a reasoning trace and are used for downstream reinforcement learning.

    +
    base_model: Qwen/Qwen2.5-3B
+model_type: AutoModelForTokenClassification
+num_labels: 2
+
+process_reward_model: true
+datasets:
+  - path: trl-lib/math_shepherd
+    type: stepwise_supervised
+    split: train
+
+val_set_size: 0.1
+eval_steps: 100
    + + +
    + +
    + +
    + + + + + \ No newline at end of file diff --git a/docs/rlhf.html b/docs/rlhf.html index 624d11df5..990e6f39c 100644 --- a/docs/rlhf.html +++ b/docs/rlhf.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/torchao.html b/docs/torchao.html index df1277cd4..1743ccd43 100644 --- a/docs/torchao.html +++ b/docs/torchao.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/docs/unsloth.html b/docs/unsloth.html index 8c26665ed..2c159e144 100644 --- a/docs/unsloth.html +++ b/docs/unsloth.html @@ -245,6 +245,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/examples/colab-notebooks/colab-axolotl-example.html b/examples/colab-notebooks/colab-axolotl-example.html index b6bf97b9d..b8d0b260c 100644 --- a/examples/colab-notebooks/colab-axolotl-example.html +++ b/examples/colab-notebooks/colab-axolotl-example.html @@ -273,6 +273,12 @@ window.Quarto = { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/index.html b/index.html index ea8534b49..b2dffcfa3 100644 --- a/index.html +++ b/index.html @@ -244,6 +244,12 @@ pre > code.sourceCode > span > a:first-child::before { text-decoration: underlin Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/listings.json b/listings.json index 00fbd1fdb..1350f1df7 100644 --- a/listings.json +++ b/listings.json @@ -5,6 +5,7 @@ "/docs/dataset-formats/pretraining.html", "/docs/dataset-formats/inst_tune.html", "/docs/dataset-formats/conversation.html", + "/docs/dataset-formats/stepwise_supervised.html", "/docs/dataset-formats/template_free.html", "/docs/dataset-formats/tokenized.html" ] diff --git a/search.json b/search.json index 13ce20ad9..f68132eb2 100644 --- a/search.json +++ b/search.json @@ -7,289 +7,62 @@ "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/dataset-formats/index.html", - "href": "docs/dataset-formats/index.html", - "title": "Dataset Formats", + "objectID": "docs/dataset-formats/stepwise_supervised.html", + "href": "docs/dataset-formats/stepwise_supervised.html", + "title": "Stepwise Supervised Format", "section": "", - "text": "Axolotl supports a variety of dataset formats. It is recommended to use a JSONL format. The schema of the JSONL depends upon the task and the prompt template you wish to use. Instead of a JSONL, you can also use a HuggingFace dataset with columns for each JSONL field.\nBelow are these various formats organized by task:\n\n\n\n\n\n\n\n\n\nTitle\n\n\nDescription\n\n\n\n\n\n\nPre-training\n\n\nData format for a pre-training completion task.\n\n\n\n\nInstruction Tuning\n\n\nInstruction tuning formats for supervised fine-tuning.\n\n\n\n\nConversation\n\n\nConversation format for supervised fine-tuning.\n\n\n\n\nTemplate-Free\n\n\nConstruct prompts without a template.\n\n\n\n\nCustom Pre-Tokenized Dataset\n\n\nHow to use a custom pre-tokenized dataset.\n\n\n\n\n\nNo matching items", + "text": "The stepwise supervised format is designed for chain-of-thought (COT) reasoning datasets where each example contains multiple completion steps and a preference label for each step. ### ExampleHere’s a simple example of a stepwise supervised dataset entry:```json { “prompt”: “Which number is larger, 9.8 or 9.11?”, “completions”: [ “The fractional part of 9.8 is 0.8, while the fractional part of 9.11 is 0.11.”, “Since 0.11 is greater than 0.8, the number 9.11 is larger than 9.8.” ], “labels”: [true, false] }", "crumbs": [ - "Dataset Formats" + "Dataset Formats", + "Stepwise Supervised Format" ] }, { - "objectID": "docs/dataset-formats/inst_tune.html", - "href": "docs/dataset-formats/inst_tune.html", - "title": "Instruction Tuning", + "objectID": "docs/dataset-formats/stepwise_supervised.html#stepwise-supervised", + "href": "docs/dataset-formats/stepwise_supervised.html#stepwise-supervised", + "title": "Stepwise Supervised Format", "section": "", - "text": "instruction; input(optional)\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\"}", + "text": "The stepwise supervised format is designed for chain-of-thought (COT) reasoning datasets where each example contains multiple completion steps and a preference label for each step. ### ExampleHere’s a simple example of a stepwise supervised dataset entry:```json { “prompt”: “Which number is larger, 9.8 or 9.11?”, “completions”: [ “The fractional part of 9.8 is 0.8, while the fractional part of 9.11 is 0.11.”, “Since 0.11 is greater than 0.8, the number 9.11 is larger than 9.8.” ], “labels”: [true, false] }", "crumbs": [ "Dataset Formats", - "Instruction Tuning" + "Stepwise Supervised Format" ] }, { - "objectID": "docs/dataset-formats/inst_tune.html#alpaca", - "href": "docs/dataset-formats/inst_tune.html#alpaca", - "title": "Instruction Tuning", + "objectID": "docs/dataset-formats/pretraining.html", + "href": "docs/dataset-formats/pretraining.html", + "title": "Pre-training", "section": "", - "text": "instruction; input(optional)\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\"}", + "text": "For pretraining, there is no prompt template or roles. The only required field is text:\n\n\ndata.jsonl\n\n{\"text\": \"first row\"}\n{\"text\": \"second row\"}\n...\n\n\n\n\n\n\n\nStreaming is recommended for large datasets\n\n\n\nAxolotl usually loads the entire dataset into memory. This will be challenging for large datasets. Use the following config to enable streaming:\n\n\nconfig.yaml\n\npretraining_dataset:\n - name:\n path:\n split:\n text_column: # column in dataset with the data, usually `text`\n type: pretrain\n trust_remote_code:\n skip: # number of rows of data to skip over from the beginning\n...", "crumbs": [ "Dataset Formats", - "Instruction Tuning" + "Pre-training" ] }, { - "objectID": "docs/dataset-formats/inst_tune.html#jeopardy", - "href": "docs/dataset-formats/inst_tune.html#jeopardy", - "title": "Instruction Tuning", - "section": "jeopardy", - "text": "jeopardy\nquestion and answer\n\n\ndata.jsonl\n\n{\"question\": \"...\", \"category\": \"...\", \"answer\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#oasst", - "href": "docs/dataset-formats/inst_tune.html#oasst", - "title": "Instruction Tuning", - "section": "oasst", - "text": "oasst\ninstruction\n\n\ndata.jsonl\n\n{\"INSTRUCTION\": \"...\", \"RESPONSE\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#gpteacher", - "href": "docs/dataset-formats/inst_tune.html#gpteacher", - "title": "Instruction Tuning", - "section": "gpteacher", - "text": "gpteacher\ninstruction; input(optional)\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#reflection", - "href": "docs/dataset-formats/inst_tune.html#reflection", - "title": "Instruction Tuning", - "section": "reflection", - "text": "reflection\ninstruction with reflect; input(optional)\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\", \"reflection\": \"...\", \"corrected\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#explainchoice", - "href": "docs/dataset-formats/inst_tune.html#explainchoice", - "title": "Instruction Tuning", - "section": "explainchoice", - "text": "explainchoice\nquestion, choices, (solution OR explanation)\n\n\ndata.jsonl\n\n{\"question\": \"...\", \"choices\": [\"...\"], \"solution\": \"...\", \"explanation\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#concisechoice", - "href": "docs/dataset-formats/inst_tune.html#concisechoice", - "title": "Instruction Tuning", - "section": "concisechoice", - "text": "concisechoice\nquestion, choices, (solution OR explanation)\n\n\ndata.jsonl\n\n{\"question\": \"...\", \"choices\": [\"...\"], \"solution\": \"...\", \"explanation\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#summarizetldr", - "href": "docs/dataset-formats/inst_tune.html#summarizetldr", - "title": "Instruction Tuning", - "section": "summarizetldr", - "text": "summarizetldr\narticle and summary\n\n\ndata.jsonl\n\n{\"article\": \"...\", \"summary\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#alpaca_chat", - "href": "docs/dataset-formats/inst_tune.html#alpaca_chat", - "title": "Instruction Tuning", - "section": "alpaca_chat", - "text": "alpaca_chat\nbasic instruct for alpaca chat\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_qa", - "href": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_qa", - "title": "Instruction Tuning", - "section": "alpaca_chat.load_qa", - "text": "alpaca_chat.load_qa\nquestion and answer for alpaca chat\n\n\ndata.jsonl\n\n{\"question\": \"...\", \"answer\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_concise", - "href": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_concise", - "title": "Instruction Tuning", - "section": "alpaca_chat.load_concise", - "text": "alpaca_chat.load_concise\nquestion and answer for alpaca chat, for concise answers\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_camel_ai", - "href": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_camel_ai", - "title": "Instruction Tuning", - "section": "alpaca_chat.load_camel_ai", - "text": "alpaca_chat.load_camel_ai\nquestion and answer for alpaca chat, for load_camel_ai\n\n\ndata.jsonl\n\n{\"message_1\": \"...\", \"message_2\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#alpaca_w_system.load_open_orca", - "href": "docs/dataset-formats/inst_tune.html#alpaca_w_system.load_open_orca", - "title": "Instruction Tuning", - "section": "alpaca_w_system.load_open_orca", - "text": "alpaca_w_system.load_open_orca\nsupport for open orca datasets with included system prompts, instruct\n\n\ndata.jsonl\n\n{\"system_prompt\": \"...\", \"question\": \"...\", \"response\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#context_qa", - "href": "docs/dataset-formats/inst_tune.html#context_qa", - "title": "Instruction Tuning", - "section": "context_qa", - "text": "context_qa\nin context question answering from an article\n\n\ndata.jsonl\n\n{\"article\": \"...\", \"question\": \"...\", \"answer\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#context_qa.load_v2", - "href": "docs/dataset-formats/inst_tune.html#context_qa.load_v2", - "title": "Instruction Tuning", - "section": "context_qa.load_v2", - "text": "context_qa.load_v2\nin context question answering (alternate)\n\n\ndata.jsonl\n\n{\"context\": \"...\", \"question\": \"...\", \"answer\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#context_qa.load_404", - "href": "docs/dataset-formats/inst_tune.html#context_qa.load_404", - "title": "Instruction Tuning", - "section": "context_qa.load_404", - "text": "context_qa.load_404\nin context question answering from an article, with default response for no answer from context\n\n\ndata.jsonl\n\n{\"article\": \"...\", \"unanswerable_question\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#creative_acr.load_answer", - "href": "docs/dataset-formats/inst_tune.html#creative_acr.load_answer", - "title": "Instruction Tuning", - "section": "creative_acr.load_answer", - "text": "creative_acr.load_answer\ninstruction and revision\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"revision\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#creative_acr.load_critique", - "href": "docs/dataset-formats/inst_tune.html#creative_acr.load_critique", - "title": "Instruction Tuning", - "section": "creative_acr.load_critique", - "text": "creative_acr.load_critique\ncritique\n\n\ndata.jsonl\n\n{\"scores\": \"...\", \"critiques\": \"...\", \"instruction\": \"...\", \"answer\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#creative_acr.load_revise", - "href": "docs/dataset-formats/inst_tune.html#creative_acr.load_revise", - "title": "Instruction Tuning", - "section": "creative_acr.load_revise", - "text": "creative_acr.load_revise\ncritique and revise\n\n\ndata.jsonl\n\n{\"scores\": \"...\", \"critiques\": \"...\", \"instruction\": \"...\", \"answer\": \"...\", \"revision\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#metharme", - "href": "docs/dataset-formats/inst_tune.html#metharme", - "title": "Instruction Tuning", - "section": "metharme", - "text": "metharme\ninstruction, adds additional eos tokens\n\n\ndata.jsonl\n\n{\"prompt\": \"...\", \"generation\": \"...\"}", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/inst_tune.html#how-to-add-custom-prompt-format", - "href": "docs/dataset-formats/inst_tune.html#how-to-add-custom-prompt-format", - "title": "Instruction Tuning", - "section": "How to add custom prompt format", - "text": "How to add custom prompt format\nFor a dataset that is preprocessed for instruction purposes:\n\n\ndata.jsonl\n\n{\"input\": \"...\", \"output\": \"...\"}\n\nYou can use this example in your YAML config:\n\n\nconfig.yaml\n\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]\"\n\nSee full config options under here.", - "crumbs": [ - "Dataset Formats", - "Instruction Tuning" - ] - }, - { - "objectID": "docs/dataset-formats/template_free.html", - "href": "docs/dataset-formats/template_free.html", - "title": "Template-Free", + "objectID": "docs/dataset-formats/tokenized.html", + "href": "docs/dataset-formats/tokenized.html", + "title": "Custom Pre-Tokenized Dataset", "section": "", - "text": "See these docs.", + "text": "Pass an empty type: in your axolotl config.\nColumns in Dataset must be exactly input_ids, attention_mask, labels\nTo indicate that a token should be ignored during training, set its corresponding label to -100.\nYou must add BOS and EOS, and make sure that you are training on EOS by not setting its label to -100.\nFor pretraining, do not truncate/pad documents to the context window length.\nFor instruction training, documents must be truncated/padded as desired.\n\nSample config:\n\n\nconfig.yml\n\ndatasets:\n - path: /path/to/your/file.jsonl\n ds_type: json\n type:\n\nSample jsonl:\n{\"input_ids\":[271,299,99],\"attention_mask\":[1,1,1],\"labels\":[271,-100,99]}\n{\"input_ids\":[87,227,8383,12],\"attention_mask\":[1,1,1,1],\"labels\":[87,227,8383,12]}", "crumbs": [ "Dataset Formats", - "Template-Free" + "Custom Pre-Tokenized Dataset" ] }, { - "objectID": "docs/amd_hpc.html", - "href": "docs/amd_hpc.html", - "title": "Training with AMD GPUs on HPC Systems", + "objectID": "docs/torchao.html", + "href": "docs/torchao.html", + "title": "PyTorch ao", "section": "", - "text": "This guide provides step-by-step instructions for installing and configuring Axolotl on a High-Performance Computing (HPC) environment equipped with AMD GPUs.", - "crumbs": [ - "How-To Guides", - "Training with AMD GPUs on HPC Systems" - ] + "text": "Installation\nStable Release from the PyTorch index\npip install torchao --extra-index-url https://download.pytorch.org/whl/cu121 # full options are cpu/cu118/cu121/cu124\nNightly release\npip install --pre torchao-nightly --index-url https://download.pytorch.org/whl/nightly/cu121 # full options are cpu/cu118/cu121/cu124" }, { - "objectID": "docs/amd_hpc.html#setup", - "href": "docs/amd_hpc.html#setup", - "title": "Training with AMD GPUs on HPC Systems", - "section": "Setup", - "text": "Setup\n\n1. Install Python\nWe recommend using Miniforge, a minimal conda-based Python distribution:\ncurl -L -O \"https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh\"\nbash Miniforge3-$(uname)-$(uname -m).sh\n\n\n2. Configure Python Environment\nAdd Python to your PATH and ensure it’s available at login:\necho 'export PATH=~/miniforge3/bin:$PATH' >> ~/.bashrc\necho 'if [ -f ~/.bashrc ]; then . ~/.bashrc; fi' >> ~/.bash_profile\n\n\n3. Load AMD GPU Software\nLoad the ROCm module:\nmodule load rocm/5.7.1\nNote: The specific module name and version may vary depending on your HPC system. Consult your system documentation for the correct module name.\n\n\n4. Install PyTorch\nInstall PyTorch with ROCm support:\npip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.7 --force-reinstall\n\n\n5. Install Flash Attention\nClone and install the Flash Attention repository:\ngit clone --recursive https://github.com/ROCmSoftwarePlatform/flash-attention.git\nexport GPU_ARCHS=\"gfx90a\"\ncd flash-attention\nexport PYTHON_SITE_PACKAGES=$(python -c 'import site; print(site.getsitepackages()[0])')\npatch \"${PYTHON_SITE_PACKAGES}/torch/utils/hipify/hipify_python.py\" hipify_patch.patch\npip install --no-build-isolation .\n\n\n6. Install Axolotl\nClone and install Axolotl:\ngit clone https://github.com/axolotl-ai-cloud/axolotl\ncd axolotl\npip install packaging ninja\npip install --no-build-isolation -e .\n\n\n7. Apply xformers Workaround\nxformers appears to be incompatible with ROCm. Apply the following workarounds: - Edit $HOME/packages/axolotl/src/axolotl/monkeypatch/llama_attn_hijack_flash.py modifying the code to always return False for SwiGLU availability from xformers. - Edit $HOME/miniforge3/lib/python3.10/site-packages/xformers/ops/swiglu_op.py replacing the “SwiGLU” function with a pass statement.\n\n\n8. Prepare Job Submission Script\nCreate a script for job submission using your HPC’s particular software (e.g. Slurm, PBS). Include necessary environment setup and the command to run Axolotl training. If the compute node(s) do(es) not have internet access, it is recommended to include\nexport TRANSFORMERS_OFFLINE=1\nexport HF_DATASETS_OFFLINE=1\n\n\n9. Download Base Model\nDownload a base model using the Hugging Face CLI:\nhuggingface-cli download meta-llama/Meta-Llama-3.1-8B --local-dir ~/hfdata/llama3.1-8B\n\n\n10. Create Axolotl Configuration\nCreate an Axolotl configuration file (YAML format) tailored to your specific training requirements and dataset. Use FSDP for multi-node training.\nNote: Deepspeed did not work at the time of testing. However, if anyone managed to get it working, please let us know.\n\n\n11. Preprocess Data\nRun preprocessing on the login node:\nCUDA_VISIBLE_DEVICES=\"\" python -m axolotl.cli.preprocess /path/to/your/config.yaml\n\n\n12. Train\nYou are now ready to submit your previously prepared job script. 🚂", - "crumbs": [ - "How-To Guides", - "Training with AMD GPUs on HPC Systems" - ] + "objectID": "docs/reward_modelling.html", + "href": "docs/reward_modelling.html", + "title": "Reward Modelling", + "section": "", + "text": "Overview\nReward modelling is a technique used to train models to predict the reward or value of a given input. This is particularly useful in reinforcement learning scenarios where the model needs to evaluate the quality of its actions or predictions. We support the reward modelling techniques supported by trl.\n\n\n(Outcome) Reward Models\nOutcome reward models are trained using data which contains preference annotations for an entire interaction between the user and model (e.g. rather than per-turn or per-step).\nbase_model: google/gemma-2-2b\nmodel_type: AutoModelForSequenceClassification\nnum_labels: 1\ntokenizer_type: AutoTokenizer\n\nreward_model: true\nchat_template: gemma\ndatasets:\n - path: argilla/distilabel-intel-orca-dpo-pairs\n type: bradley_terry.chat_template\n\nval_set_size: 0.1\neval_steps: 100\n\n\nProcess Reward Models (PRM)\nProcess reward models are trained using data which contains preference annotations for each step in a series of interactions. Typically, PRMs are trained to provide reward signals over each step of a reasoning trace and are used for downstream reinforcement learning.\nbase_model: Qwen/Qwen2.5-3B\nmodel_type: AutoModelForTokenClassification\nnum_labels: 2\n\nprocess_reward_model: true\ndatasets:\n - path: trl-lib/math_shepherd\n type: stepwise_supervised\n split: train\n\nval_set_size: 0.1\neval_steps: 100" }, { "objectID": "docs/input_output.html", @@ -329,7 +102,7 @@ "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, 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 revision: # Optional[str] The specific revision of the dataset to use when loading from the Hugging Face Hub. This can be a commit hash, tag, or branch name. If not specified, the latest version will be used. This parameter is ignored for local datasets.\n trust_remote_code: # Optional[bool] Trust remote code for untrusted source\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 # Using chat template\n - path: ...\n # Set type to `chat_template` to use this strategy\n type: chat_template\n # Specify the name of the chat template to use\n # The name of the chat template to use for training, following values are supported:\n # - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default.\n # - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py\n # - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to if the tokenizer does not have a chat template else default to tokenizer. E.g. tokenizer_default_fallback_chatml.\n # - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field.\n chat_template: tokenizer_default\n\n # Custom jinja chat template. Used only if `chat_template: jinja` or empty.\n chat_template_jinja:\n\n # Key containing the messages (default: \"messages\")\n field_messages: messages\n # Key for role in each message (default: \"role\")\n message_field_role: role\n # Key for content in each message (default: \"content\")\n message_field_content: content\n\n # Optional[Dict[str, List]]. Roles mapping in the messages. The default is:\n roles:\n user: [\"human\", \"user\"]\n assistant: [\"gpt\", \"assistant\"]\n system: [\"system\"]\n tool: [\"tool\"]\n\n # IMPORTANT: The following fields determine which parts of the conversation to train on.\n # Priority order: message_field_training > message_field_training_detail > train_on_inputs or role in roles_to_train\n # See examples at `docs/dataset-formats/conversation.qmd`\n # Note: If the below 4 fields are empty, defaults to training only on the last message.\n\n # Optional[List[str]]. Roles to train on. The tokens from these roles will be considered for the loss.\n roles_to_train: [\"assistant\"] # default\n # Optional[str]. Which EOS tokens to train on in the conversation. Possible values are:\n # - all: train on all EOS tokens\n # - turn (default): train on the EOS token at the end of each trainable turn\n # - last: train on the last EOS token in the conversation\n train_on_eos: last\n # The key in the message turn that indicates via boolean whether tokens of a turn should be considered for training. Useful to selectively train on certain turns besides the `roles_to_train`.\n message_field_training: training\n # The key in the message turn that contains the training details. Useful to selectively train on certain tokens in a turn.\n # The value of the key is a List[Dict] containing `begin_offset` (start character index in content), `end_offset` (end character index in content), and `train` (boolean whether to train).\n message_field_training_detail: train_detail\n\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\nDeduplicates datasets and test_datasets with identical entries.\ndataset_exact_deduplication: 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'\nrl:\n# whether to perform weighting if doing DPO training. Boolean.\ndpo_use_weighting:\n\n# The name of the chat template to use for training, following values are supported:\n# - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default value.\n# - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py\n# - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not available in the tokenizer.\n# - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field.\n# The selected chat template will be saved to the tokenizer_config.json for easier inferencing\n# Note: It is recommended to set train_on_inputs to true when using a chat template that is different from the model's default chat template.\nchat_template: tokenizer_default\n# custom jinja template for chat template. This will be only used if chat_template is set to `jinja` or `null` (in which case chat_template is automatically set to `jinja`). Default is null.\nchat_template_jinja: null\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# Increasing the following values helps with packing, but usually only slightly (<%1.)\n# The number of samples packed at a time.\nsample_packing_group_size: 100000\n# The number of samples which can be packed into one sequence. Increase if using a large sequence_len with many short samples.\nsample_packing_bin_size: 200\n# whether to concatenate samples during pretraining\npretraining_sample_concatenation:\n\n# Use batch flattening for speedups when not using sample_packing\nbatch_flattening:\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\n# LoRA+ hyperparameters\n# For more details about the following options, see:\n# https://arxiv.org/abs/2402.12354 and `src/axolotl/core/train_builder.py`\nloraplus_lr_ratio: # loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4.\nloraplus_lr_embedding: # loraplus learning rate for lora embedding layers. Default value is 1e-6.\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\nmlflow_run_name: # Your run name\nhf_mlflow_log_artifacts: # set to true to copy each saved checkpoint on each save to mlflow artifact registry\n\n# Comet configuration if you're using it\n# Make sure your `COMET_API_KEY` environment variable is set (recommended) or you login to Comet with `comet login`.\n# Check out our documentation for more details https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/Experiment-Creation/#comet_ml.start\nuse_comet: # Enable or disable Comet integration.\ncomet_api_key: # API key for Comet. Recommended to set via `comet login`.\ncomet_workspace: # Workspace name in Comet. Defaults to the user's default workspace.\ncomet_project_name: # Project name in Comet. Defaults to Uncategorized.\ncomet_experiment_key: # Identifier for the experiment. Used to append data to an existing experiment or control the key of new experiments. Default to a random key.\ncomet_mode: # Create a new experiment (\"create\") or log to an existing one (\"get\"). Default (\"get_or_create\") auto-selects based on configuration.\ncomet_online: # Set to True to log data to Comet server, or False for offline storage. Default is True.\ncomet_experiment_config: # Dictionary for additional configuration settings, see the doc for more details.\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\n# setting to `auto` will enable torch compile when torch>=2.5.1\ntorch_compile: # Optional[Union[Literal[\"auto\"], 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.\n# Batch size per gpu = micro_batch_size * gradient_accumulation_steps\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, integer for every N steps. float for fraction of total steps\nevals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps\neval_strategy: # Set to `\"no\"` to skip evaluation, `\"epoch\"` at end of each epoch, leave empty to infer from `eval_steps`.\nsave_strategy: # Set to `\"no\"` to skip checkpoint saves, `\"epoch\"` at end of each epoch, `\"best\"` when better result is achieved, leave empty to infer from `save_steps`.\nsave_steps: # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps\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\", \"perplexity\"]\n\nprofiler_steps: # enable the pytorch profiler to capture the first N steps of training to the output_dir.\n # see https://pytorch.org/blog/understanding-gpu-memory-1/ for more information\n # snapshots can be visualized @ https://pytorch.org/memory_viz\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# - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version >= 2.5.1)\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 # pad_token: \"[PAD]\"\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:", + "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, 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 revision: # Optional[str] The specific revision of the dataset to use when loading from the Hugging Face Hub. This can be a commit hash, tag, or branch name. If not specified, the latest version will be used. This parameter is ignored for local datasets.\n trust_remote_code: # Optional[bool] Trust remote code for untrusted source\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 # Using chat template\n - path: ...\n # Set type to `chat_template` to use this strategy\n type: chat_template\n # Specify the name of the chat template to use\n # The name of the chat template to use for training, following values are supported:\n # - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default.\n # - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py\n # - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to if the tokenizer does not have a chat template else default to tokenizer. E.g. tokenizer_default_fallback_chatml.\n # - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field.\n chat_template: tokenizer_default\n\n # Custom jinja chat template. Used only if `chat_template: jinja` or empty.\n chat_template_jinja:\n\n # Key containing the messages (default: \"messages\")\n field_messages: messages\n # Key for role in each message (default: \"role\")\n message_field_role: role\n # Key for content in each message (default: \"content\")\n message_field_content: content\n\n # Optional[Dict[str, List]]. Roles mapping in the messages. The default is:\n roles:\n user: [\"human\", \"user\"]\n assistant: [\"gpt\", \"assistant\"]\n system: [\"system\"]\n tool: [\"tool\"]\n\n # IMPORTANT: The following fields determine which parts of the conversation to train on.\n # Priority order: message_field_training > message_field_training_detail > train_on_inputs or role in roles_to_train\n # See examples at `docs/dataset-formats/conversation.qmd`\n # Note: If the below 4 fields are empty, defaults to training only on the last message.\n\n # Optional[List[str]]. Roles to train on. The tokens from these roles will be considered for the loss.\n roles_to_train: [\"assistant\"] # default\n # Optional[str]. Which EOS tokens to train on in the conversation. Possible values are:\n # - all: train on all EOS tokens\n # - turn (default): train on the EOS token at the end of each trainable turn\n # - last: train on the last EOS token in the conversation\n train_on_eos: last\n # The key in the message turn that indicates via boolean whether tokens of a turn should be considered for training. Useful to selectively train on certain turns besides the `roles_to_train`.\n message_field_training: training\n # The key in the message turn that contains the training details. Useful to selectively train on certain tokens in a turn.\n # The value of the key is a List[Dict] containing `begin_offset` (start character index in content), `end_offset` (end character index in content), and `train` (boolean whether to train).\n message_field_training_detail: train_detail\n\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\nDeduplicates datasets and test_datasets with identical entries.\ndataset_exact_deduplication: 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'\nrl:\n# whether to perform weighting if doing DPO training. Boolean.\ndpo_use_weighting:\n\n# reward modelling: `True` or `False`\nreward_model:\n\n# process reward modelling: `True` or `False`\nprocess_reward_model:\n\n# The name of the chat template to use for training, following values are supported:\n# - tokenizer_default: Uses the chat template that is available in the tokenizer_config.json. If the chat template is not available in the tokenizer, it will raise an error. This is the default value.\n# - alpaca/inst/chatml/gemma/cohere/llama3/phi_3/deepseek_v2/jamba: These chat templates are available in the axolotl codebase at src/axolotl/utils/chat_templates.py\n# - tokenizer_default_fallback_*: where * is the name of the chat template to fallback to. E.g. tokenizer_default_fallback_chatml. This is useful when the chat template is not available in the tokenizer.\n# - jinja: Uses a custom jinja template for the chat template. The custom jinja template should be provided in the chat_template_jinja field.\n# The selected chat template will be saved to the tokenizer_config.json for easier inferencing\n# Note: It is recommended to set train_on_inputs to true when using a chat template that is different from the model's default chat template.\nchat_template: tokenizer_default\n# custom jinja template for chat template. This will be only used if chat_template is set to `jinja` or `null` (in which case chat_template is automatically set to `jinja`). Default is null.\nchat_template_jinja: null\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# Increasing the following values helps with packing, but usually only slightly (<%1.)\n# The number of samples packed at a time.\nsample_packing_group_size: 100000\n# The number of samples which can be packed into one sequence. Increase if using a large sequence_len with many short samples.\nsample_packing_bin_size: 200\n# whether to concatenate samples during pretraining\npretraining_sample_concatenation:\n\n# Use batch flattening for speedups when not using sample_packing\nbatch_flattening:\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\n# LoRA+ hyperparameters\n# For more details about the following options, see:\n# https://arxiv.org/abs/2402.12354 and `src/axolotl/core/train_builder.py`\nloraplus_lr_ratio: # loraplus learning rate ratio lr_B / lr_A. Recommended value is 2^4.\nloraplus_lr_embedding: # loraplus learning rate for lora embedding layers. Default value is 1e-6.\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\nmlflow_run_name: # Your run name\nhf_mlflow_log_artifacts: # set to true to copy each saved checkpoint on each save to mlflow artifact registry\n\n# Comet configuration if you're using it\n# Make sure your `COMET_API_KEY` environment variable is set (recommended) or you login to Comet with `comet login`.\n# Check out our documentation for more details https://www.comet.com/docs/v2/api-and-sdk/python-sdk/reference/Experiment-Creation/#comet_ml.start\nuse_comet: # Enable or disable Comet integration.\ncomet_api_key: # API key for Comet. Recommended to set via `comet login`.\ncomet_workspace: # Workspace name in Comet. Defaults to the user's default workspace.\ncomet_project_name: # Project name in Comet. Defaults to Uncategorized.\ncomet_experiment_key: # Identifier for the experiment. Used to append data to an existing experiment or control the key of new experiments. Default to a random key.\ncomet_mode: # Create a new experiment (\"create\") or log to an existing one (\"get\"). Default (\"get_or_create\") auto-selects based on configuration.\ncomet_online: # Set to True to log data to Comet server, or False for offline storage. Default is True.\ncomet_experiment_config: # Dictionary for additional configuration settings, see the doc for more details.\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\n# setting to `auto` will enable torch compile when torch>=2.5.1\ntorch_compile: # Optional[Union[Literal[\"auto\"], 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.\n# Batch size per gpu = micro_batch_size * gradient_accumulation_steps\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, integer for every N steps. float for fraction of total steps\nevals_per_epoch: # number of times per epoch to run evals, mutually exclusive with eval_steps\neval_strategy: # Set to `\"no\"` to skip evaluation, `\"epoch\"` at end of each epoch, leave empty to infer from `eval_steps`.\nsave_strategy: # Set to `\"no\"` to skip checkpoint saves, `\"epoch\"` at end of each epoch, `\"best\"` when better result is achieved, leave empty to infer from `save_steps`.\nsave_steps: # Leave empty to save at each epoch, integer for every N steps. float for fraction of total steps\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\", \"perplexity\"]\n\nprofiler_steps: # enable the pytorch profiler to capture the first N steps of training to the output_dir.\n # see https://pytorch.org/blog/understanding-gpu-memory-1/ for more information\n # snapshots can be visualized @ https://pytorch.org/memory_viz\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# - adopt_adamw (an EXPERIMENTAL optimizer, only for torch version >= 2.5.1)\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 # pad_token: \"[PAD]\"\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" @@ -796,32 +569,288 @@ ] }, { - "objectID": "docs/torchao.html", - "href": "docs/torchao.html", - "title": "PyTorch ao", + "objectID": "docs/amd_hpc.html", + "href": "docs/amd_hpc.html", + "title": "Training with AMD GPUs on HPC Systems", "section": "", - "text": "Installation\nStable Release from the PyTorch index\npip install torchao --extra-index-url https://download.pytorch.org/whl/cu121 # full options are cpu/cu118/cu121/cu124\nNightly release\npip install --pre torchao-nightly --index-url https://download.pytorch.org/whl/nightly/cu121 # full options are cpu/cu118/cu121/cu124" - }, - { - "objectID": "docs/dataset-formats/tokenized.html", - "href": "docs/dataset-formats/tokenized.html", - "title": "Custom Pre-Tokenized Dataset", - "section": "", - "text": "Pass an empty type: in your axolotl config.\nColumns in Dataset must be exactly input_ids, attention_mask, labels\nTo indicate that a token should be ignored during training, set its corresponding label to -100.\nYou must add BOS and EOS, and make sure that you are training on EOS by not setting its label to -100.\nFor pretraining, do not truncate/pad documents to the context window length.\nFor instruction training, documents must be truncated/padded as desired.\n\nSample config:\n\n\nconfig.yml\n\ndatasets:\n - path: /path/to/your/file.jsonl\n ds_type: json\n type:\n\nSample jsonl:\n{\"input_ids\":[271,299,99],\"attention_mask\":[1,1,1],\"labels\":[271,-100,99]}\n{\"input_ids\":[87,227,8383,12],\"attention_mask\":[1,1,1,1],\"labels\":[87,227,8383,12]}", + "text": "This guide provides step-by-step instructions for installing and configuring Axolotl on a High-Performance Computing (HPC) environment equipped with AMD GPUs.", "crumbs": [ - "Dataset Formats", - "Custom Pre-Tokenized Dataset" + "How-To Guides", + "Training with AMD GPUs on HPC Systems" ] }, { - "objectID": "docs/dataset-formats/pretraining.html", - "href": "docs/dataset-formats/pretraining.html", - "title": "Pre-training", + "objectID": "docs/amd_hpc.html#setup", + "href": "docs/amd_hpc.html#setup", + "title": "Training with AMD GPUs on HPC Systems", + "section": "Setup", + "text": "Setup\n\n1. Install Python\nWe recommend using Miniforge, a minimal conda-based Python distribution:\ncurl -L -O \"https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh\"\nbash Miniforge3-$(uname)-$(uname -m).sh\n\n\n2. Configure Python Environment\nAdd Python to your PATH and ensure it’s available at login:\necho 'export PATH=~/miniforge3/bin:$PATH' >> ~/.bashrc\necho 'if [ -f ~/.bashrc ]; then . ~/.bashrc; fi' >> ~/.bash_profile\n\n\n3. Load AMD GPU Software\nLoad the ROCm module:\nmodule load rocm/5.7.1\nNote: The specific module name and version may vary depending on your HPC system. Consult your system documentation for the correct module name.\n\n\n4. Install PyTorch\nInstall PyTorch with ROCm support:\npip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.7 --force-reinstall\n\n\n5. Install Flash Attention\nClone and install the Flash Attention repository:\ngit clone --recursive https://github.com/ROCmSoftwarePlatform/flash-attention.git\nexport GPU_ARCHS=\"gfx90a\"\ncd flash-attention\nexport PYTHON_SITE_PACKAGES=$(python -c 'import site; print(site.getsitepackages()[0])')\npatch \"${PYTHON_SITE_PACKAGES}/torch/utils/hipify/hipify_python.py\" hipify_patch.patch\npip install --no-build-isolation .\n\n\n6. Install Axolotl\nClone and install Axolotl:\ngit clone https://github.com/axolotl-ai-cloud/axolotl\ncd axolotl\npip install packaging ninja\npip install --no-build-isolation -e .\n\n\n7. Apply xformers Workaround\nxformers appears to be incompatible with ROCm. Apply the following workarounds: - Edit $HOME/packages/axolotl/src/axolotl/monkeypatch/llama_attn_hijack_flash.py modifying the code to always return False for SwiGLU availability from xformers. - Edit $HOME/miniforge3/lib/python3.10/site-packages/xformers/ops/swiglu_op.py replacing the “SwiGLU” function with a pass statement.\n\n\n8. Prepare Job Submission Script\nCreate a script for job submission using your HPC’s particular software (e.g. Slurm, PBS). Include necessary environment setup and the command to run Axolotl training. If the compute node(s) do(es) not have internet access, it is recommended to include\nexport TRANSFORMERS_OFFLINE=1\nexport HF_DATASETS_OFFLINE=1\n\n\n9. Download Base Model\nDownload a base model using the Hugging Face CLI:\nhuggingface-cli download meta-llama/Meta-Llama-3.1-8B --local-dir ~/hfdata/llama3.1-8B\n\n\n10. Create Axolotl Configuration\nCreate an Axolotl configuration file (YAML format) tailored to your specific training requirements and dataset. Use FSDP for multi-node training.\nNote: Deepspeed did not work at the time of testing. However, if anyone managed to get it working, please let us know.\n\n\n11. Preprocess Data\nRun preprocessing on the login node:\nCUDA_VISIBLE_DEVICES=\"\" python -m axolotl.cli.preprocess /path/to/your/config.yaml\n\n\n12. Train\nYou are now ready to submit your previously prepared job script. 🚂", + "crumbs": [ + "How-To Guides", + "Training with AMD GPUs on HPC Systems" + ] + }, + { + "objectID": "docs/dataset-formats/template_free.html", + "href": "docs/dataset-formats/template_free.html", + "title": "Template-Free", "section": "", - "text": "For pretraining, there is no prompt template or roles. The only required field is text:\n\n\ndata.jsonl\n\n{\"text\": \"first row\"}\n{\"text\": \"second row\"}\n...\n\n\n\n\n\n\n\nStreaming is recommended for large datasets\n\n\n\nAxolotl usually loads the entire dataset into memory. This will be challenging for large datasets. Use the following config to enable streaming:\n\n\nconfig.yaml\n\npretraining_dataset:\n - name:\n path:\n split:\n text_column: # column in dataset with the data, usually `text`\n type: pretrain\n trust_remote_code:\n skip: # number of rows of data to skip over from the beginning\n...", + "text": "See these docs.", "crumbs": [ "Dataset Formats", - "Pre-training" + "Template-Free" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html", + "href": "docs/dataset-formats/inst_tune.html", + "title": "Instruction Tuning", + "section": "", + "text": "instruction; input(optional)\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#alpaca", + "href": "docs/dataset-formats/inst_tune.html#alpaca", + "title": "Instruction Tuning", + "section": "", + "text": "instruction; input(optional)\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#jeopardy", + "href": "docs/dataset-formats/inst_tune.html#jeopardy", + "title": "Instruction Tuning", + "section": "jeopardy", + "text": "jeopardy\nquestion and answer\n\n\ndata.jsonl\n\n{\"question\": \"...\", \"category\": \"...\", \"answer\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#oasst", + "href": "docs/dataset-formats/inst_tune.html#oasst", + "title": "Instruction Tuning", + "section": "oasst", + "text": "oasst\ninstruction\n\n\ndata.jsonl\n\n{\"INSTRUCTION\": \"...\", \"RESPONSE\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#gpteacher", + "href": "docs/dataset-formats/inst_tune.html#gpteacher", + "title": "Instruction Tuning", + "section": "gpteacher", + "text": "gpteacher\ninstruction; input(optional)\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#reflection", + "href": "docs/dataset-formats/inst_tune.html#reflection", + "title": "Instruction Tuning", + "section": "reflection", + "text": "reflection\ninstruction with reflect; input(optional)\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"output\": \"...\", \"reflection\": \"...\", \"corrected\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#explainchoice", + "href": "docs/dataset-formats/inst_tune.html#explainchoice", + "title": "Instruction Tuning", + "section": "explainchoice", + "text": "explainchoice\nquestion, choices, (solution OR explanation)\n\n\ndata.jsonl\n\n{\"question\": \"...\", \"choices\": [\"...\"], \"solution\": \"...\", \"explanation\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#concisechoice", + "href": "docs/dataset-formats/inst_tune.html#concisechoice", + "title": "Instruction Tuning", + "section": "concisechoice", + "text": "concisechoice\nquestion, choices, (solution OR explanation)\n\n\ndata.jsonl\n\n{\"question\": \"...\", \"choices\": [\"...\"], \"solution\": \"...\", \"explanation\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#summarizetldr", + "href": "docs/dataset-formats/inst_tune.html#summarizetldr", + "title": "Instruction Tuning", + "section": "summarizetldr", + "text": "summarizetldr\narticle and summary\n\n\ndata.jsonl\n\n{\"article\": \"...\", \"summary\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#alpaca_chat", + "href": "docs/dataset-formats/inst_tune.html#alpaca_chat", + "title": "Instruction Tuning", + "section": "alpaca_chat", + "text": "alpaca_chat\nbasic instruct for alpaca chat\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_qa", + "href": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_qa", + "title": "Instruction Tuning", + "section": "alpaca_chat.load_qa", + "text": "alpaca_chat.load_qa\nquestion and answer for alpaca chat\n\n\ndata.jsonl\n\n{\"question\": \"...\", \"answer\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_concise", + "href": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_concise", + "title": "Instruction Tuning", + "section": "alpaca_chat.load_concise", + "text": "alpaca_chat.load_concise\nquestion and answer for alpaca chat, for concise answers\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"input\": \"...\", \"response\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_camel_ai", + "href": "docs/dataset-formats/inst_tune.html#alpaca_chat.load_camel_ai", + "title": "Instruction Tuning", + "section": "alpaca_chat.load_camel_ai", + "text": "alpaca_chat.load_camel_ai\nquestion and answer for alpaca chat, for load_camel_ai\n\n\ndata.jsonl\n\n{\"message_1\": \"...\", \"message_2\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#alpaca_w_system.load_open_orca", + "href": "docs/dataset-formats/inst_tune.html#alpaca_w_system.load_open_orca", + "title": "Instruction Tuning", + "section": "alpaca_w_system.load_open_orca", + "text": "alpaca_w_system.load_open_orca\nsupport for open orca datasets with included system prompts, instruct\n\n\ndata.jsonl\n\n{\"system_prompt\": \"...\", \"question\": \"...\", \"response\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#context_qa", + "href": "docs/dataset-formats/inst_tune.html#context_qa", + "title": "Instruction Tuning", + "section": "context_qa", + "text": "context_qa\nin context question answering from an article\n\n\ndata.jsonl\n\n{\"article\": \"...\", \"question\": \"...\", \"answer\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#context_qa.load_v2", + "href": "docs/dataset-formats/inst_tune.html#context_qa.load_v2", + "title": "Instruction Tuning", + "section": "context_qa.load_v2", + "text": "context_qa.load_v2\nin context question answering (alternate)\n\n\ndata.jsonl\n\n{\"context\": \"...\", \"question\": \"...\", \"answer\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#context_qa.load_404", + "href": "docs/dataset-formats/inst_tune.html#context_qa.load_404", + "title": "Instruction Tuning", + "section": "context_qa.load_404", + "text": "context_qa.load_404\nin context question answering from an article, with default response for no answer from context\n\n\ndata.jsonl\n\n{\"article\": \"...\", \"unanswerable_question\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#creative_acr.load_answer", + "href": "docs/dataset-formats/inst_tune.html#creative_acr.load_answer", + "title": "Instruction Tuning", + "section": "creative_acr.load_answer", + "text": "creative_acr.load_answer\ninstruction and revision\n\n\ndata.jsonl\n\n{\"instruction\": \"...\", \"revision\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#creative_acr.load_critique", + "href": "docs/dataset-formats/inst_tune.html#creative_acr.load_critique", + "title": "Instruction Tuning", + "section": "creative_acr.load_critique", + "text": "creative_acr.load_critique\ncritique\n\n\ndata.jsonl\n\n{\"scores\": \"...\", \"critiques\": \"...\", \"instruction\": \"...\", \"answer\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#creative_acr.load_revise", + "href": "docs/dataset-formats/inst_tune.html#creative_acr.load_revise", + "title": "Instruction Tuning", + "section": "creative_acr.load_revise", + "text": "creative_acr.load_revise\ncritique and revise\n\n\ndata.jsonl\n\n{\"scores\": \"...\", \"critiques\": \"...\", \"instruction\": \"...\", \"answer\": \"...\", \"revision\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#metharme", + "href": "docs/dataset-formats/inst_tune.html#metharme", + "title": "Instruction Tuning", + "section": "metharme", + "text": "metharme\ninstruction, adds additional eos tokens\n\n\ndata.jsonl\n\n{\"prompt\": \"...\", \"generation\": \"...\"}", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/inst_tune.html#how-to-add-custom-prompt-format", + "href": "docs/dataset-formats/inst_tune.html#how-to-add-custom-prompt-format", + "title": "Instruction Tuning", + "section": "How to add custom prompt format", + "text": "How to add custom prompt format\nFor a dataset that is preprocessed for instruction purposes:\n\n\ndata.jsonl\n\n{\"input\": \"...\", \"output\": \"...\"}\n\nYou can use this example in your YAML config:\n\n\nconfig.yaml\n\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]\"\n\nSee full config options under here.", + "crumbs": [ + "Dataset Formats", + "Instruction Tuning" + ] + }, + { + "objectID": "docs/dataset-formats/index.html", + "href": "docs/dataset-formats/index.html", + "title": "Dataset Formats", + "section": "", + "text": "Axolotl supports a variety of dataset formats. It is recommended to use a JSONL format. The schema of the JSONL depends upon the task and the prompt template you wish to use. Instead of a JSONL, you can also use a HuggingFace dataset with columns for each JSONL field.\nBelow are these various formats organized by task:\n\n\n\n\n\n\n\n\n\nTitle\n\n\nDescription\n\n\n\n\n\n\nPre-training\n\n\nData format for a pre-training completion task.\n\n\n\n\nInstruction Tuning\n\n\nInstruction tuning formats for supervised fine-tuning.\n\n\n\n\nConversation\n\n\nConversation format for supervised fine-tuning.\n\n\n\n\nStepwise Supervised Format\n\n\nFormat for datasets with stepwise completions and labels\n\n\n\n\nTemplate-Free\n\n\nConstruct prompts without a template.\n\n\n\n\nCustom Pre-Tokenized Dataset\n\n\nHow to use a custom pre-tokenized dataset.\n\n\n\n\n\nNo matching items", + "crumbs": [ + "Dataset Formats" ] }, { diff --git a/sitemap.xml b/sitemap.xml index aaa4c2db4..f9065e73d 100644 --- a/sitemap.xml +++ b/sitemap.xml @@ -2,118 +2,126 @@ https://axolotl-ai-cloud.github.io/axolotl/FAQS.html - 2025-01-29T04:23:56.857Z + 2025-01-29T05:08:44.746Z - https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/index.html - 2025-01-29T04:23:56.858Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/inst_tune.html - 2025-01-29T04:23:56.858Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/template_free.html - 2025-01-29T04:23:56.858Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/amd_hpc.html - 2025-01-29T04:23:56.858Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/input_output.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/config.html - 2025-01-29T04:23:56.858Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/lr_groups.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/debugging.html - 2025-01-29T04:23:56.859Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/faq.html - 2025-01-29T04:23:56.859Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/multi-node.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/unsloth.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/examples/colab-notebooks/colab-axolotl-example.html - 2025-01-29T04:23:56.861Z - - - https://axolotl-ai-cloud.github.io/axolotl/TODO.html - 2025-01-29T04:23:56.857Z - - - https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html - 2025-01-29T04:23:56.874Z - - - https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/LICENSE.html - 2025-01-29T04:23:56.874Z - - - https://axolotl-ai-cloud.github.io/axolotl/index.html - 2025-01-29T04:23:56.871Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/mac.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/multimodal.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/batch_vs_grad.html - 2025-01-29T04:23:56.858Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/dataset_preprocessing.html - 2025-01-29T04:23:56.858Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/fsdp_qlora.html - 2025-01-29T04:23:56.859Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/rlhf.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/multipack.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/nccl.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/torchao.html - 2025-01-29T04:23:56.860Z - - - https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/tokenized.html - 2025-01-29T04:23:56.858Z + https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/stepwise_supervised.html + 2025-01-29T05:08:44.747Z https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/pretraining.html - 2025-01-29T04:23:56.858Z + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/tokenized.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/torchao.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/reward_modelling.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/input_output.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/config.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/lr_groups.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/debugging.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/faq.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/multi-node.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/unsloth.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/examples/colab-notebooks/colab-axolotl-example.html + 2025-01-29T05:08:44.750Z + + + https://axolotl-ai-cloud.github.io/axolotl/TODO.html + 2025-01-29T05:08:44.746Z + + + https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html + 2025-01-29T05:08:44.763Z + + + https://axolotl-ai-cloud.github.io/axolotl/src/axolotl/integrations/LICENSE.html + 2025-01-29T05:08:44.763Z + + + https://axolotl-ai-cloud.github.io/axolotl/index.html + 2025-01-29T05:08:44.760Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/mac.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/multimodal.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/batch_vs_grad.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/dataset_preprocessing.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/fsdp_qlora.html + 2025-01-29T05:08:44.748Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/rlhf.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/multipack.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/nccl.html + 2025-01-29T05:08:44.749Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/amd_hpc.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/template_free.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/inst_tune.html + 2025-01-29T05:08:44.747Z + + + https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/index.html + 2025-01-29T05:08:44.747Z https://axolotl-ai-cloud.github.io/axolotl/docs/dataset-formats/conversation.html - 2025-01-29T04:23:56.858Z + 2025-01-29T05:08:44.747Z diff --git a/src/axolotl/integrations/LICENSE.html b/src/axolotl/integrations/LICENSE.html index 1358af7ef..e22db7249 100644 --- a/src/axolotl/integrations/LICENSE.html +++ b/src/axolotl/integrations/LICENSE.html @@ -210,6 +210,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +
  • diff --git a/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html b/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html index 78dfac229..b3e97505e 100644 --- a/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html +++ b/src/axolotl/integrations/cut_cross_entropy/ACKNOWLEDGEMENTS.html @@ -210,6 +210,12 @@ ul.task-list li input[type="checkbox"] { Conversation
    +
    • +
  • +
    + + Stepwise Supervised Format +