An actuator for benchmarking fine-tuning of foundation models

Navigation

Verified details

These details have been verified by PyPI
Maintainers
Avatar for alessandropomponio-ibm from gravatar.com alessandropomponio-ibm

Unverified details

These details have not been verified by PyPI
Report project as malware

Project description

Currently supported experiments

Table of contents

Overview

The SFTTrainer actuator provides a flexible and scalable interface for running supervised fine-tuning (SFT) experiments on large language and vision-language models. It supports a variety of fine-tuning strategies including full fine-tuning, LoRA, QPTQ-LoRA, and prompt-tuning across both text-to-text and image-to-text datasets.

Designed for high-performance and distributed environments, SFTTrainer supports:

  • Single-GPU, multi-GPU, and multi-node training
  • Distributed Data Parallel (DDP) and Fully Sharded Data Parallel (FSDP) strategies
  • RDMA over Converged Ethernet (RoCE) for optimized multi-node communication
  • Ray-based task scheduling, enabling execution on both Kubernetes clusters and bare-metal infrastructure

Under the hood, this actuator wraps the fms-hf-tuning library, which itself builds on the SFTTrainer API from Hugging Face Transformers. This layered design allows users to leverage the robustness of the Hugging Face ecosystem while benefiting from ado’s orchestration and reproducibility features.

Requirements

fms-hf-tuning imports packages like flash-attn and mamba-ssm, which import torch during their
build phase. This means the base virtual environment of your Ray workers must
already include the appropriate version of torch:

  • fms-hf-tuning <= 2.8.2

    • Install torch==2.4.1
    • For RayClusters on Kubernetes, use: quay.io/ado/ado:1.0.1-py310-cu121-ofed2410v1140

  • fms-hf-tuning > 2.8.2

    • Install torch==2.6.0
      • Requires Python 3.11
    • For RayClusters on Kubernetes, use: quay.io/ado/ado:c6ba952ad79a2d86d1174fd9aaebddd8953c78cf-py311-cu121-ofed2410v1140

Full Fine-Tuning Experiments

finetune_full_benchmark-v1.0.0

An experiment instance:

  • performs full fine tuning
    • You may notice that even large-memory GPUs like the 80GB variant of the NVIDIA A100 chip need at least 2 GPUs to train models as big as 13B parameters.
  • the training data is artificial
  • use_flash_attn is set to True
  • packing is set to False
  • torch_dtype is set to bfloat16 by default, can also be float16
  • uses the FSDP distributed backend for multi-gpu runs by default, can also be DDP
  • multi-gpu runs with FSDP and DDP backends use 1 process per GPU (via accelerate)
  • runs 1 epoch by default, can also run a custom number of steps
  • does not save checkpoint
  • loads weights from a PVC
  • request 2 CPU cores per GPU device (with a minimum of 2 cores)

For FSDP runs we use the following accelerate_config.yml YAML file:

compute_environment: LOCAL_MACHINE
distributed_type: FSDP
downcast_bf16: "no"
fsdp_config:
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_backward_prefetch: BACKWARD_PRE
  fsdp_forward_prefetch: false
  fsdp_offload_params: false
  fsdp_sharding_strategy: ${fsdp_sharding_strategy}
  fsdp_state_dict_type: ${fsdp_state_dict_type}
  fsdp_cpu_ram_efficient_loading: true
  fsdp_sync_module_states: true
  fsdp_transformer_layer_cls_to_wrap: ${accelerate_config_fsdp_transformer_layer_cls_to_wrap}
machine_rank: { $THE MACHINE RANK - always 0 for single-node runs }
main_training_function: main
mixed_precision: ${accelerate_config_mixed_precision}
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: ${SOME_PORT}
num_processes: ${NUM_GPUS}

For DDP runs we use this instead:

compute_environment: LOCAL_MACHINE
debug: False
downcast_bf16: no
distributed_type: MULTI_GPU
machine_rank: { $THE MACHINE RANK - always 0 for single-node runs }
main_training_function: main
mixed_precision: ${accelerate_config_mixed_precision}
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: ${SOME_PORT}
num_processes: ${NUM_GPUS}

Commandline:

accelerate launch --config_file ${PATH_ACCELERATE_CONFIG} --num_processes ${NUMBER_GPUS} \
  ${PATH_TO_OUR_WRAPPER_OF_FMS_HF_TUNING_SFT_TRAINER} --model_name_or_path ${MODEL} \
  --torch_dtype bfloat16 --use_flash_attn True --training_data_path ${DATASET_PATH} \
  --response_template "\n### Response:" --dataset_text_field output --log_level debug \
  --num_train_epochs 1 --per_device_train_batch_size ${BATCH_SIZE/NUM_GPUS} \
  --max_seq_length ${MODEL_MAX_LENGTH} --eval_strategy no --output_dir ${RANDOM_DIR} \
  --gradient_accumulation_steps ${GRADIENT_ACCUMULATION_STEPS} --save_strategy no \
  --learning_rate 1e-05 --weight_decay 0.0 --warmup_ratio 0.03 --lr_scheduler_type cosine \
  --logging_steps 1 --include_tokens_per_second True --gradient_checkpointing True \
  --packing False --peft_method none --optim ${OPTIM} --bf16 ${BF16} \
  --gradient_checkpointing_kwargs='{"use_reentrant": ${GRADIENT_CHECKPOINTING_USE_REENTRANT}}' \
  --fast_moe ${FAST_MOE}

Note: --fast_moe is only supported for fms-hf-tuning v2.4.0+

We use a thin wrapper of sft_trainer.py which injects a custom Callback that exports the metrics collected by AIM. You can repeat our experiments by just pointing the above command-line to sft_trainer.py from the fms-hf-tuning package.

Versioning:

Requirements

  • The S3 bucket watson.runtime.wisdom.model.us-south mounted under /ibm-research-models (instructions).
  • The PVC hf-models-pvc mounted under /hf-models-pvc - should contain the models:
    • LLaMa/models/hf/13B/
    • LLaMa/models/hf/7B/
    • LLaMa/models/hf/llama2-70b/
    • LLaMa/models/hf/llama3-70b/
    • LLaMa/models/hf/llama3-8b/
    • LLaMa/models/hf/llama3.1-405b/
    • LLaMa/models/hf/llama3.1-70b/
    • LLaMa/models/hf/llama3.1-8b/
    • Mixtral-8x7B-Instruct-v0.1/
    • allam-1-13b-instruct-20240607/
    • granite-13b-base-v2/step_300000_ckpt/
    • granite-20b-code-base-v2/step_280000_ckpt/
    • granite-34b-code-base/
    • granite-8b-code-base/
    • granite-8b-japanese-base-v1-llama/
    • mistralai-mistral-7b-v0.1/
    • mistral-large/fp16_240620
  • The PVC ray-disorch-storage mounted under /data with the preprocessed artificial-dataset files (https://github.ibm.com/ai-foundation/watson-fm-stack-tracker/issues/550) under /data/fms-hf-tuning/artificial-dataset

Entity space

Required:

  • model_name: Supported models: ["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]
  • model_max_length: Maximum sequence length. Sequences will be right padded (and possibly truncated)
  • number_gpus: The effective number of GPUs (to be evenly distributed to number_nodes machines)
  • batch_size: the effective batch_size (will be evenly distributed to max(1, number_gpus) devices)
  • gpu_model: The value of the kubernetes node label nvidia.com/gpu.prod for example
    • NVIDIA-A100-80GB-PCIe
    • NVIDIA-A100-SXM4-80GB
    • NVIDIA-H100-PCIe

Optional:

  • dataset_id: Default is news-tokens-16384plus-entries-4096. Available options are:
    • news-chars-512-entries-4096: 4096 entries with samples of 512 + 127 (prompt) + 512 characters
    • news-chars-1024-entries-4096: 4096 entries with samples of 1024 + 127 (prompt) + 1024 characters
    • news-chars-2048-entries-4096: 4096 entries with samples of 2048 + 127 (prompt) + 2048 characters
    • news-tokens-16384plus-entries-4096: 4096 entries, each entry has least 16384 tokens when tokenized with any of the granite-13b-v2, llama-13b-v2, llama-7b, or granite-20b-v2 tokenizers
    • vision-384x384-16384plus-entries-4096: A vision dataset containing 4096 entries. Each entry includes at least 16384 tokens when tokenized with granite-vision-3.2-2b, and consists of repeated copies of a single image with dimensions 384×384.
    • vision-384x768-16384plus-entries-4096: Similar to the above, this dataset also contains 4096 entries with a minimum of 16384 tokens per entry (tokenized using granite-vision-3.2-2b). Each entry uses repeated copies of a single image sized 384×768.
  • gradient_checkpointing: Default is True. If True, use gradient checkpointing to save memory (i.e. higher batchsizes) at the expense of slower backward pass
  • gradient_accumulation_steps: Default is 4. Number of update steps to accumulate before performing a backward/update pass. Only takes effect when gradient_checkpointing is True
  • torch_dtype: Default is bfloat16. One of bfloat16, float32, float16
  • max_steps: Default is -1. The number of optimization steps to perform. Set to -1 to respect num_train_epochs instead.
  • num_train_epochs: Default is 1.0. How many epochs to run. Ignored if max_steps is greater than 0.
  • stop_after_seconds: Default is -1.0. If set, the optimizer will be asked to stop after the specified time elapses. The check is performed after the end of each training step.
  • auto_stop_method: The default value is None. This parameter defines the method used to automatically stop the fine-tuning job. Supported values are WARMUP_60S_STABLE_120S_OR_10_STEPS and None. If set to WARMUP_60S_STABLE_120S_OR_10_STEPS, the job stops after spending at least 60 seconds in the warmup phase plus the longer of 120 seconds or the duration of 10 optimization steps. This method excludes the first 60 seconds of training when calculating throughput and system metrics.
  • distributed_backend: Default is FSDP for multi-gpu measurements, None (i.e. Data Parallel (DP)) for single-gpu measurements. Which pytorch backend to use when training with multiple GPU devices.
  • number_nodes: Default is 1. If set, actuator distributes tasks on multiple nodes. Each Node will use number_gpus/number_nodes GPUs. Each Node will use 1 process for each GPU it uses
  • fms_hf_tuning_version: Default is 2.1.2. Which version of fms-hf-tuning to use. Available options are: 3.0.0, 2.8.2, 2.7.1, 2.6.0, 2.5.0, 2.4.0, 2.3.1, 2.2.1, 2.1.2, 2.1.0, 2.0.1
  • enable_roce: Default is False. This setting is only in effect for multi-node runs. It controls whether RDMA over Converged Ethernet (RoCE) is switched on or not.
  • fast_moe: Default is 0. Configures the amount of expert parallel sharding. number_gpus must be divisible by it
  • fast_kernels: Default is None. Switches on fast kernels, the value is a list with strings of boolean values for [fast_loss, fast_rms_layernorm, fast_rope_embeddings]
  • optim: Default is adamw_torch. The optimizer to use. Available options are adamw_hf, adamw_torch, adamw_torch_fused, adamw_torch_xla, adamw_torch_npu_fused, adamw_apex_fused, adafactor, adamw_anyprecision, adamw_torch_4bit, ademamix, sgd, adagrad, adamw_bnb_8bit, adamw_8bit, ademamix_8bit, lion_8bit, lion_32bit, paged_adamw_32bit, paged_adamw_8bit, paged_ademamix_32bit, paged_ademamix_8bit, paged_lion_32bit, paged_lion_8bit, rmsprop, rmsprop_bnb, rmsprop_bnb_8bit, rmsprop_bnb_32bit, galore_adamw, galore_adamw_8bit, galore_adafactor, galore_adamw_layerwise, galore_adamw_8bit_layerwise, galore_adafactor_layerwise, lomo, adalomo, grokadamw, schedule_free_adamw, schedule_free_sgd
  • bf16: Default is False. Whether to use bf16 (mixed) precision instead of 32-bit. Requires Ampere or higher NVIDIA add bf16 mixed precision support for NPU architecture or using CPU (use_cpu) or Ascend NPU. This is an experimental API and it may change. Can be True, False.
  • gradient_checkpointing_use_reentrant: Default is False Specify whether to use the activation checkpoint variant that requires reentrant autograd. This parameter should be passed explicitly. Torch version 2.5 will raise an exception if use_reentrant is not passed. If use_reentrant=False, checkpoint will use an implementation that does not require reentrant autograd. This allows checkpoint to support additional functionality, such as working as expected with torch.autograd.grad and support for keyword arguments input into the checkpointed function. Can be True, False.
  • fsdp_sharding_strategy: Default is FULL_SHARD. [1] FULL_SHARD (shards optimizer states, gradients and parameters), " [2] SHARD_GRAD_OP (shards optimizer states and gradients), [3] NO_SHARD (DDP), [4] HYBRID_SHARD (shards optimizer states, gradients and parameters within each node while each node has full copy), [5] HYBRID_SHARD_ZERO2 (shards optimizer states and gradients within each node while each node has full copy). For more information, please refer the official PyTorch docs.
  • fsdp_state_dict_type: Default is FULL_STATE_DICT. [1] FULL_STATE_DICT, [2] LOCAL_STATE_DICT, [3] SHARDED_STATE_DICT
  • fsdp_use_orig_params: Default is True. If True, allows non-uniform requires_grad during init, which means support for interspersed frozen and trainable parameters. (useful only when use_fsdp flag is passed).
  • accelerate_config_mixed_precision: Default is no. Whether to use mixed precision training or not. Choose from no,fp16,bf16 or fp8. fp8 requires the installation of transformers-engine.
  • accelerate_config_fsdp_transformer_layer_cls_to_wrap: Default is None. List of transformer layer class names (case-sensitive) to wrap, e.g, GraniteDecoderLayer, LlamaDecoderLayer, MistralDecoderLayer, BertLayer, GPTJBlock, T5Block ... (useful only when using FSDP)
  • dataset_text_field: Default is None. Training dataset text field containing single sequence. Either the dataset_text_field or data_formatter_template need to be supplied. For running vision language model tuning pass the column name for text data.
  • dataset_image_field: Default is None. For running vision language model tuning pass the column name of the image data in the dataset.
  • remove_unused_columns: Default is True. Remove columns not required by the model when using an nlp.Dataset.
  • dataset_kwargs_skip_prepare_dataset: Default is False. When True, configures trl to skip preparing the dataset

NOTE: Because running accelerate with a single gpu is unsupported, when setting number_gpus to 1 this experiment actually runs the tuning.sft_trainer script directly (i.e. a DataParallel (DP) run).

Measured properties

We use AIM to collect profiling metadata. Then we convert the timeseries that AIM collects into the metrics below.

  • gpu_compute_utilization_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_compute_utilization_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_compute_utilization_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_watts_min (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_power_watts_avg (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_power_watts_max (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_memory_utilization_peak (0.0 if not using any GPUs): peak GPU memory utilization percentage across all devices
  • cpu_compute_utilization: Measured in Percentages (0 to 100 where 100 means 1 full core) (see note 2)
  • cpu_memory_utilization: Measured in Percentages (0 to 100) taken from AIM (see note 2)
  • train_runtime: Measured in seconds
  • train_samples_per_second
  • train_steps_per_second
  • train_tokens_per_second: How many tokens (including padding tokens) the run processed every second (for FSDP this is estimated from num_gpus * rank_0_train_tokens_per_second). Omitted when stop_after_seconds is greater than 0
  • train_tokens_per_gpu_per_second (will be equal to train_tokens_per_second when number_gpus <= 1, for FSDP this is reported using just rank 0). Omitted when stop_after_seconds is greater than 0
  • dataset_tokens_per_second: How many tokens from the dataset the run processed every second (see note 3)
  • dataset_tokens_per_second_per_gpu: How many tokens from the dataset the run processed every second per GPU (see note 3)
  • is_valid (see is_valid logic)

Notes:

  • (1) They are reported as the min/max/avg of the average in the timeseries metrics that AIM collects for the in-use GPUs of a run.
  • (2) CPU compute and memory utilization are percentages. They are reported as the min/max/avg of the metrics that AIM collects for the sft_trainer.py process. The total memory capacity of the nodes varies from 100 GB to 400 GB. Currently, we do not store this information in our database.
  • (3) dataset_tokens_per_second and dataset_tokens_per_second_per_gpu take into account the tokenizer.model_max_length and max_seq_length (i.e. for each entry, we report min(len(tokens(entry["output"])), tokenizer.model_max_length, max_seq_length)).

is_valid logic

A run for an entity is invalid if:

  1. batch_size cannot be evenly divided by number_gpus (i.e. batch_size % number_gpus != 0)
  2. number_gpus cannot be evenly divided by number_nodes (i.e. number_gpus % number_nodes != 0)
  3. number_nodes must be greater than 0
  4. batch_size must be greater than 0
  5. if number_gpus is greater than 0 then gpu_model must be a non-empty string
  6. if fast_moe is set and number_gpus is not divisible by it
  7. if fast_moe is set and the num_local_experts of the Mixture of Experts (MoE) model is not divisible by fast_moe (which is interpreted as ep_degrees by fms-hf-tuning)

Runs raising the following errors are considered invalid due to running out of GPU memory:

  • torch.cuda.OutOfMemoryError
  • RuntimeError: CUDA error: an illegal memory access was encountered

Measurements raising any other exception (including for example RuntimeError containing the string NCCL Error) are considered to have Failed. They will not contain the is_valid measured property, or any other property for that matter. Failed measurements do not record any properties and can be repeated.

Full Fine-Tuning Experiments for exploring GPU Out Of Memory and Transient Errors

finetune_full_stability-v1.0.0

An experiment instance:

  • performs full fine-tuning 5 times and reports the fraction of tasks that ran out of GPU memory, exhibited some unknown error, or completed successfully
    • You may notice that even large-memory GPUs like the 80GB variant of the NVIDIA A100 chip need at least 2 GPUs to train models as big as 13B parameters.
  • the training data is artificial
  • use_flash_attn is set to True
  • packing is set to False
  • torch_dtype is set to bfloat16
  • uses the FSDP distributed backend
  • runs 5 optimization steps
  • does not save checkpoint
  • loads weights from a PVC
  • request 2 CPU cores per GPU device (with a minimum of 2 cores)

We use the following accelerate_config.yml YAML file for all models:

compute_environment: LOCAL_MACHINE
debug: False
distributed_type: FSDP
downcast_bf16: "no"
fsdp_config:
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_backward_prefetch: BACKWARD_PRE
  fsdp_forward_prefetch: false
  fsdp_offload_params: false
  fsdp_sharding_strategy: 1
  fsdp_state_dict_type: FULL_STATE_DICT
  fsdp_cpu_ram_efficient_loading: true
  fsdp_sync_module_states: true
machine_rank: 0
main_training_function: main
mixed_precision: "no"
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: { $SOME_PORT }
num_processes: { $NUM_GPUS }

Commandline:

accelerate launch --config_file ${PATH_ACCELERATE_CONFIG} --num_processes ${NUMBER_GPUS} \
  ${PATH_TO_OUR_WRAPPER_OF_FMS_HF_TUNING_SFT_TRAINER} --model_name_or_path ${MODEL} \
  --torch_dtype bfloat16 --use_flash_attn True --training_data_path ${DATASET_PATH} \
  --response_template "\n### Response:" --dataset_text_field output --log_level debug \
  --max_steps -1 --per_device_train_batch_size ${BATCH_SIZE/NUM_GPUS} \
  --max_seq_length ${MODEL_MAX_LENGTH} --eval_strategy no --output_dir ${RANDOM_DIR} \
  --gradient_accumulation_steps ${GRADIENT_ACCUMULATION_STEPS} --save_strategy no \
  --learning_rate 1e-05 --weight_decay 0.0 --warmup_ratio 0.03 --lr_scheduler_type cosine \
  --logging_steps 1 --include_tokens_per_second True --gradient_checkpointing True \
  --packing False --peft_method none --optim ${OPTIM} --bf16 ${BF16} \
  --gradient_checkpointing_kwargs='{"use_reentrant": ${GRADIENT_CHECKPOINTING_USE_REENTRANT}}' \
  --fast_moe ${FAST_MOE}

Note: --fast_moe is only supported for fms-hf-tuning v2.4.0+

We use a thin wrapper of sft_trainer.py which injects a custom Callback that exports the metrics collected by AIM. You can repeat our experiments by just pointing the above command-line to sft_trainer.py from the fms-hf-tuning package.

Versioning:

Requirements

  • The S3 bucket watson.runtime.wisdom.model.us-south mounted under /ibm-research-models (instructions).
  • The PVC hf-models-pvc mounted under /hf-models-pvc - should contain the models:
    • LLaMa/models/hf/13B/
    • LLaMa/models/hf/7B/
    • LLaMa/models/hf/llama2-70b/
    • LLaMa/models/hf/llama3-70b/
    • LLaMa/models/hf/llama3-8b/
    • LLaMa/models/hf/llama3.1-405b/
    • LLaMa/models/hf/llama3.1-70b/
    • LLaMa/models/hf/llama3.1-8b/
    • Mixtral-8x7B-Instruct-v0.1/
    • allam-1-13b-instruct-20240607/
    • granite-13b-base-v2/step_300000_ckpt/
    • granite-20b-code-base-v2/step_280000_ckpt/
    • granite-34b-code-base/
    • granite-8b-code-base/
    • granite-8b-japanese-base-v1-llama/
    • mistralai-mistral-7b-v0.1/
    • mistral-large/fp16_240620
  • The PVC ray-disorch-storage mounted under /data with the preprocessed artificial-dataset files (https://github.ibm.com/ai-foundation/watson-fm-stack-tracker/issues/550) under /data/fms-hf-tuning/artificial-dataset

Entity space

  • model_name: Supported models: ["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]
  • dataset_id: One of
    • news-chars-512-entries-4096: 4096 entries with samples of 512 + 127 (prompt) + 512 characters
    • news-chars-1024-entries-4096: 4096 entries with samples of 1024 + 127 (prompt) + 1024 characters
    • news-chars-2048-entries-4096: 4096 entries with samples of 2048 + 127 (prompt) + 2048 characters
    • news-tokens-16384plus-entries-4096: 4096 entries, each entry has least 16384 tokens when tokenized with any of the granite-13b-v2, llama-13b-v2, llama-7b, or granite-20b-v2 tokenizers
    • news-tokens-128kplus-entries-320 : 320 entries, each entry has at least 128*1024 tokens
    • vision-384x384-16384plus-entries-4096: A vision dataset containing 4096 entries. Each entry includes at least 16384 tokens when tokenized with granite-vision-3.2-2b, and consists of repeated copies of a single image with dimensions 384×384.
    • vision-384x768-16384plus-entries-4096: Similar to the above, this dataset also contains 4096 entries with a minimum of 16384 tokens per entry (tokenized using granite-vision-3.2-2b). Each entry uses repeated copies of a single image sized 384×768.
  • number_gpus: Can be 0 or more - no support for multi-node runs
  • model_max_length: Maximum sequence length. Sequences will be right padded (and possibly truncated)
  • torch_dtype: Here you can use any valid torch_dtype value e.g. float32, bfloat16, float16, etc
  • batch_size: the effective batch_size (will be evenly distributed to max(1, number_gpus) devices)
  • gpu_model: The value of the kubernetes node label nvidia.com/gpu.prod for example
    • NVIDIA-A100-80GB-PCIe
    • NVIDIA-A100-SXM4-80GB
    • NVIDIA-H100-PCIe
  • gradient_accumulation_steps: Number of update steps to accumulate before performing a backward/update pass. Defaults to 4 when not set.

Measured properties

  • f_gpu_oom: fraction of tasks that ran out of GPU memory
  • f_other_error: fraction of tasks that ran into an unknown error
  • f_no_error: fraction of tasks that completed successfully
  • is_valid: whether this collection of tasks is a valid point to investigate

is_valid logic

A run for an entity is invalid if:

  1. batch_size cannot be evenly divided by number_gpus (i.e. batch_size % number_gpus != 0)
  2. number_gpus cannot be evenly divided by number_nodes (i.e. number_gpus % number_nodes != 0)
  3. number_nodes must be greater than 0
  4. batch_size must be greater than 0
  5. if number_gpus is greater than 0 then gpu_model must be a non-empty string
  6. if fast_moe is set and number_gpus is not divisible by it
  7. if fast_moe is set and the num_local_experts of the Mixture of Experts (MoE) model is not divisible by fast_moe (which is interpreted as ep_degrees by fms-hf-tuning)

Runs raising the following errors are considered invalid due to running out of GPU memory:

  • torch.cuda.OutOfMemoryError
  • RuntimeError: CUDA error: an illegal memory access was encountered

Measurements raising any other exception (including for example RuntimeError containing the string NCCL Error) are considered to have Failed. They will not contain the is_valid measured property, or any other property for that matter. Failed measurements do not record any properties and can be repeated.

LORA Fine-Tuning Experiments

finetune_lora_benchmark-v1.0.0

An experiment instance:

  • performs LORA fine tuning
  • the training data is artificial
  • use_flash_attn is set to True
  • packing is set to False
  • torch_dtype is set to bfloat16 by default, can also be float16
  • uses the FSDP distributed backend for multi-gpu runs by default, can also be DDP
  • multi-gpu runs with FSDP and DDP backends use 1 process per GPU (via accelerate)
  • runs 1 epoch by default, can also run a custom number of steps
  • does not save checkpoint
  • loads weights from a PVC
  • request 2 CPU cores per GPU device (with a minimum of 2 cores)

For FSDP runs we use the following accelerate_config.yml YAML file:

compute_environment: LOCAL_MACHINE
distributed_type: FSDP
downcast_bf16: "no"
fsdp_config:
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_backward_prefetch: BACKWARD_PRE
  fsdp_forward_prefetch: false
  fsdp_offload_params: false
  fsdp_sharding_strategy: ${fsdp_sharding_strategy}
  fsdp_state_dict_type: ${fsdp_state_dict_type}
  fsdp_cpu_ram_efficient_loading: true
  fsdp_sync_module_states: true
  fsdp_transformer_layer_cls_to_wrap: ${accelerate_config_fsdp_transformer_layer_cls_to_wrap}
machine_rank: { $THE MACHINE RANK - always 0 for single-node runs }
main_training_function: main
mixed_precision: ${accelerate_config_mixed_precision}
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: { $SOME_PORT }
num_processes: { $NUM_GPUS }

For DDP runs we use this instead:

compute_environment: LOCAL_MACHINE
debug: False
downcast_bf16: no
distributed_type: MULTI_GPU
machine_rank: { $THE MACHINE RANK - always 0 for single-node runs }
main_training_function: main
mixed_precision: ${accelerate_config_mixed_precision}
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: { $SOME_PORT }
num_processes: { $NUM_GPUS }

Commandline:

accelerate launch --config_file ${PATH_ACCELERATE_CONFIG} --num_processes ${NUMBER_GPUS} \
  ${PATH_TO_OUR_WRAPPER_OF_FMS_HF_TUNING_SFT_TRAINER} --model_name_or_path ${MODEL} \
  --torch_dtype bfloat16 --use_flash_attn True --training_data_path ${DATASET_PATH} \
  --response_template "\n### Response:" --dataset_text_field output --log_level debug \
  --num_train_epochs 1 --per_device_train_batch_size ${BATCH_SIZE/NUM_GPUS} \
  --max_seq_length ${MODEL_MAX_LENGTH} --eval_strategy no --output_dir ${RANDOM_DIR} \
  --gradient_accumulation_steps ${GRADIENT_ACCUMULATION_STEPS} --save_strategy no \
  --learning_rate 1e-05 --weight_decay 0.0 --warmup_ratio 0.03 --lr_scheduler_type cosine \
  --logging_steps 1 --include_tokens_per_second True --gradient_checkpointing True \
  --packing False --peft_method lora --target_modules ${SPACE SEPARATED LAYER NAMES} \
  --optim ${OPTIM} --bf16 ${BF16} \
  --gradient_checkpointing_kwargs='{"use_reentrant": ${GRADIENT_CHECKPOINTING_USE_REENTRANT}}' \
  --fast_moe ${FAST_MOE}

Note: --fast_moe is only supported for fms-hf-tuning v2.4.0+

We use a thin wrapper of sft_trainer.py which injects a custom Callback that exports the metrics collected by AIM. You can repeat our experiments by just pointing the above command-line to sft_trainer.py from the fms-hf-tuning package.

Versioning:

Requirements

  • The S3 bucket watson.runtime.wisdom.model.us-south mounted under /ibm-research-models (instructions).
  • The PVC hf-models-pvc mounted under /hf-models-pvc - should contain the models:
    • LLaMa/models/hf/13B/
    • LLaMa/models/hf/7B/
    • LLaMa/models/hf/llama2-70b/
    • LLaMa/models/hf/llama3-70b/
    • LLaMa/models/hf/llama3-8b/
    • LLaMa/models/hf/llama3.1-405b/
    • LLaMa/models/hf/llama3.1-70b/
    • LLaMa/models/hf/llama3.1-8b/
    • Mixtral-8x7B-Instruct-v0.1/
    • allam-1-13b-instruct-20240607/
    • granite-13b-base-v2/step_300000_ckpt/
    • granite-20b-code-base-v2/step_280000_ckpt/
    • granite-34b-code-base/
    • granite-8b-code-base/
    • granite-8b-japanese-base-v1-llama/
    • mistralai-mistral-7b-v0.1/
    • mistral-large/fp16_240620
  • The PVC ray-disorch-storage mounted under /data with the preprocessed artificial-dataset files (https://github.ibm.com/ai-foundation/watson-fm-stack-tracker/issues/550) under /data/fms-hf-tuning/artificial-dataset

Entity space

Required:

  • model_name: Supported models: ["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]
  • model_max_length: Maximum sequence length. Sequences will be right padded (and possibly truncated)
  • number_gpus: The effective number of GPUs (to be evenly distributed to number_nodes machines)
  • batch_size: the effective batch_size (will be evenly distributed to max(1, number_gpus) devices)
  • gpu_model: The value of the kubernetes node label nvidia.com/gpu.prod for example
    • NVIDIA-A100-80GB-PCIe
    • NVIDIA-A100-SXM4-80GB
    • NVIDIA-H100-PCIe

Optional:

  • dataset_id: Default is news-tokens-16384plus-entries-4096. Available options are:
    • news-chars-512-entries-4096: 4096 entries with samples of 512 + 127 (prompt) + 512 characters
    • news-chars-1024-entries-4096: 4096 entries with samples of 1024 + 127 (prompt) + 1024 characters
    • news-chars-2048-entries-4096: 4096 entries with samples of 2048 + 127 (prompt) + 2048 characters
    • news-tokens-16384plus-entries-4096: 4096 entries, each entry has least 16384 tokens when tokenized with any of the granite-13b-v2, llama-13b-v2, llama-7b, or granite-20b-v2 tokenizers
    • vision-384x384-16384plus-entries-4096: A vision dataset containing 4096 entries. Each entry includes at least 16384 tokens when tokenized with granite-vision-3.2-2b, and consists of repeated copies of a single image with dimensions 384×384.
    • vision-384x768-16384plus-entries-4096: Similar to the above, this dataset also contains 4096 entries with a minimum of 16384 tokens per entry (tokenized using granite-vision-3.2-2b). Each entry uses repeated copies of a single image sized 384×768.
  • gradient_checkpointing: Default is True. If True, use gradient checkpointing to save memory (i.e. higher batchsizes) at the expense of slower backward pass
  • gradient_accumulation_steps: Default is 4. Number of update steps to accumulate before performing a backward/update pass. Only takes effect when gradient_checkpointing is True
  • torch_dtype: Default is bfloat16. One of bfloat16, float32, float16
  • max_steps: Default is -1. The number of optimization steps to perform. Set to -1 to respect num_train_epochs instead.
  • num_train_epochs: Default is 1.0. How many epochs to run. Ignored if max_steps is greater than 0.
  • stop_after_seconds: Default is -1.0. If set, the optimizer will be asked to stop after the specified time elapses. The check is performed after the end of each training step.
  • auto_stop_method: The default value is None. This parameter defines the method used to automatically stop the fine-tuning job. Supported values are WARMUP_60S_STABLE_120S_OR_10_STEPS and None. If set to WARMUP_60S_STABLE_120S_OR_10_STEPS, the job stops after spending at least 60 seconds in the warmup phase plus the longer of 120 seconds or the duration of 10 optimization steps. This method excludes the first 60 seconds of training when calculating throughput and system metrics.
  • distributed_backend: Default is FSDP for multi-gpu measurements, None (i.e. Data Parallel (DP)) for single-gpu measurements. Which pytorch backend to use when training with multiple GPU devices.
  • number_nodes: Default is 1. If set, actuator distributes tasks on multiple nodes. Each Node will use number_gpus/number_nodes GPUs. Each Node will use 1 process for each GPU it uses
  • fms_hf_tuning_version: Default is 2.1.2. Which version of fms-hf-tuning to use. Available options are: 3.0.0, 2.8.2, 2.7.1, 2.6.0, 2.5.0, 2.4.0, 2.3.1, 2.2.1, 2.1.2, 2.1.0, 2.0.1
  • enable_roce: Default is False. This setting is only in effect for multi-node runs. It controls whether RDMA over Converged Ethernet (RoCE) is switched on or not.
  • fast_moe: Default is 0. Configures the amount of expert parallel sharding. number_gpus must be divisible by it
  • fast_kernels: Default is None. Switches on fast kernels, the value is a list with strings of boolean values for [fast_loss, fast_rms_layernorm, fast_rope_embeddings]
  • r: Default is 4. The LORA rank
  • lora_alpha: Default is 16. Scales the learning weights.
  • optim: Default is adamw_torch. The optimizer to use. Available options are adamw_hf, adamw_torch, adamw_torch_fused, adamw_torch_xla, adamw_torch_npu_fused, adamw_apex_fused, adafactor, adamw_anyprecision, adamw_torch_4bit, ademamix, sgd, adagrad, adamw_bnb_8bit, adamw_8bit, ademamix_8bit, lion_8bit, lion_32bit, paged_adamw_32bit, paged_adamw_8bit, paged_ademamix_32bit, paged_ademamix_8bit, paged_lion_32bit, paged_lion_8bit, rmsprop, rmsprop_bnb, rmsprop_bnb_8bit, rmsprop_bnb_32bit, galore_adamw, galore_adamw_8bit, galore_adafactor, galore_adamw_layerwise, galore_adamw_8bit_layerwise, galore_adafactor_layerwise, lomo, adalomo, grokadamw, schedule_free_adamw, schedule_free_sgd
  • bf16: Default is False. Whether to use bf16 (mixed) precision instead of 32-bit. Requires Ampere or higher NVIDIA add bf16 mixed precision support for NPU architecture or using CPU (use_cpu) or Ascend NPU. This is an experimental API and it may change. Can be True, False.
  • gradient_checkpointing_use_reentrant: Default is False Specify whether to use the activation checkpoint variant that requires reentrant autograd. This parameter should be passed explicitly. Torch version 2.5 will raise an exception if use_reentrant is not passed. If use_reentrant=False, checkpoint will use an implementation that does not require reentrant autograd. This allows checkpoint to support additional functionality, such as working as expected with torch.autograd.grad and support for keyword arguments input into the checkpointed function. Can be True, False.
  • fsdp_sharding_strategy: Default is FULL_SHARD. [1] FULL_SHARD (shards optimizer states, gradients and parameters), " [2] SHARD_GRAD_OP (shards optimizer states and gradients), [3] NO_SHARD (DDP), [4] HYBRID_SHARD (shards optimizer states, gradients and parameters within each node while each node has full copy), [5] HYBRID_SHARD_ZERO2 (shards optimizer states and gradients within each node while each node has full copy). For more information, please refer the official PyTorch docs.
  • fsdp_state_dict_type: Default is FULL_STATE_DICT. [1] FULL_STATE_DICT, [2] LOCAL_STATE_DICT, [3] SHARDED_STATE_DICT
  • fsdp_use_orig_params: Default is True. If True, allows non-uniform requires_grad during init, which means support for interspersed frozen and trainable parameters. (useful only when use_fsdp flag is passed).
  • accelerate_config_mixed_precision: Default is no. Whether to use mixed precision training or not. Choose from no,fp16,bf16 or fp8. fp8 requires the installation of transformers-engine.
  • accelerate_config_fsdp_transformer_layer_cls_to_wrap: Default is None. List of transformer layer class names (case-sensitive) to wrap, e.g, GraniteDecoderLayer, LlamaDecoderLayer, MistralDecoderLayer, BertLayer, GPTJBlock, T5Block ... (useful only when using FSDP)
  • dataset_text_field: Default is None. Training dataset text field containing single sequence. Either the dataset_text_field or data_formatter_template need to be supplied. For running vision language model tuning pass the column name for text data.
  • dataset_image_field: Default is None. For running vision language model tuning pass the column name of the image data in the dataset.
  • remove_unused_columns: Default is True. Remove columns not required by the model when using an nlp.Dataset.
  • dataset_kwargs_skip_prepare_dataset: Default is False. When True, configures trl to skip preparing the dataset

Hardcoded:

Sets the --target_modules layer names based on the model_name:

  • llama3.2-1b: ["q_proj", "v_proj"]
  • llama3.2-3b: ["q_proj", "v_proj"]
  • smollm2-135m: ["q_proj", "v_proj"]
  • granite-3.0-1b-a400m-base: ["q_proj", "v_proj"]
  • granite-3.1-3b-a800m-instruct: ["q_proj", "v_proj"]
  • granite-vision-3.2-2b: ["q_proj", "v_proj"]
  • granite-3b-code-base-128k: ["q_proj", "v_proj"]
  • granite-7b-base: ["q_proj", "v_proj"]
  • granite-8b-code-base-128k: ["q_proj", "v_proj"]
  • granite-8b-code-base: ["q_proj", "v_proj"]
  • granite-8b-japanese: ["q_proj", "v_proj"]
  • granite-13b-v2: ["c_attn", "c_proj"]
  • granite-20b-v2: ["c_attn", "c_proj"]
  • granite-34b-code-base: ["c_attn", "c_proj"]
  • llama-7b: ["q_proj", "k_proj"]
  • llama-13b: ["q_proj", "k_proj"]
  • llama2-70b: ["q_proj", "v_proj"]
  • llama3-8b: ["q_proj", "k_proj"]
  • llama3-70b: ["q_proj", "v_proj"]
  • llama3.1-8b: ["q_proj", "v_proj"]
  • llama3.1-70b: ["q_proj", "v_proj"]
  • llama3.1-405b: ["q_proj", "v_proj"]
  • granite-4.0-micro: ["q_proj", "v_proj"]
  • granite-4.0-h-1b: ["q_proj", "v_proj"]
  • granite-4.0-350m: ["q_proj", "v_proj"]
  • granite-4.0-h-small: ["q_proj", "v_proj"]
  • granite-4.0-h-micro: ["q_proj", "v_proj"]
  • granite-4.0-h-tiny: ["q_proj", "v_proj"]
  • allam-1-13b: ["q_proj", "v_proj"]
  • hf-tiny-model-private/tiny-random-BloomForCausalLM: ["dense_h_to_4h", "dense_4h_to_4h"]
  • mistral-7b-v0.1: ["q_proj", "v_proj"]
  • mistral-123b-v2: ["q_proj", "v_proj"]
  • mixtral-8x7b-instruct-v0.1: ["q_proj", "v_proj"]
  • granite-3-8b: ["q_proj", "v_proj"]
  • granite-3.1-2b: ["q_proj", "v_proj"]
  • granite-3.1-8b-instruct: ["q_proj", "v_proj"]
  • llava-v1.6-mistral-7b: ["q_proj", "v_proj"]

NOTE: Because running accelerate with a single gpu is unsupported, when setting number_gpus to 1 this experiment actually runs the tuning.sft_trainer script directly (i.e. a DataParallel (DP) run).

Measured properties

We use AIM to collect profiling metadata. Then we convert the timeseries that AIM collects into the metrics below.

  • gpu_compute_utilization_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_compute_utilization_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_compute_utilization_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_watts_min (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_power_watts_avg (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_power_watts_max (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_memory_utilization_peak (0.0 if not using any GPUs): peak GPU memory utilization percentage across all devices
  • cpu_compute_utilization: Measured in Percentages (0 to 100 where 100 means 1 full core) (see note 2)
  • cpu_memory_utilization: Measured in Percentages (0 to 100) taken from AIM (see note 2)
  • train_runtime: Measured in seconds
  • train_samples_per_second
  • train_steps_per_second
  • train_tokens_per_second: How many tokens (including padding tokens) the run processed every second (for FSDP this is estimated from num_gpus * rank_0_train_tokens_per_second). Omitted when stop_after_seconds is greater than 0
  • train_tokens_per_gpu_per_second (will be equal to train_tokens_per_second when number_gpus <= 1, for FSDP this is reported using just rank 0). Omitted when stop_after_seconds is greater than 0
  • dataset_tokens_per_second: How many tokens from the dataset the run processed every second (see note 3)
  • dataset_tokens_per_second_per_gpu: How many tokens from the dataset the run processed every second per GPU (see note 3)
  • is_valid (see is_valid logic)

Notes:

  • (1) They are reported as the min/max/avg of the average in the timeseries metrics that AIM collects for the in-use GPUs of a run.
  • (2) CPU compute and memory utilization are percentages. They are reported as the min/max/avg of the metrics that AIM collects for the sft_trainer.py process. The total memory capacity of the nodes varies from 100 GB to 400 GB. Currently, we do not store this information in our database.
  • (3) dataset_tokens_per_second and dataset_tokens_per_second_per_gpu take into account the tokenizer.model_max_length and max_seq_length (i.e. for each entry, we report min(len(tokens(entry["output"])), tokenizer.model_max_length, max_seq_length)).

is_valid logic

A run for an entity is invalid if:

  1. batch_size cannot be evenly divided by number_gpus (i.e. batch_size % number_gpus != 0)
  2. number_gpus cannot be evenly divided by number_nodes (i.e. number_gpus % number_nodes != 0)
  3. number_nodes must be greater than 0
  4. batch_size must be greater than 0
  5. if number_gpus is greater than 0 then gpu_model must be a non-empty string
  6. if fast_moe is set and number_gpus is not divisible by it
  7. if fast_moe is set and the num_local_experts of the Mixture of Experts (MoE) model is not divisible by fast_moe (which is interpreted as ep_degrees by fms-hf-tuning)

Runs raising the following errors are considered invalid due to running out of GPU memory:

  • torch.cuda.OutOfMemoryError
  • RuntimeError: CUDA error: an illegal memory access was encountered

Measurements raising any other exception (including for example RuntimeError containing the string NCCL Error) are considered to have Failed. They will not contain the is_valid measured property, or any other property for that matter. Failed measurements do not record any properties and can be repeated.

GPTQ-LORA Fine-Tuning Experiments

finetune_gtpq-lora_benchmark-v1.0.0

An experiment instance:

  • performs LORA fine tuning
  • the training data is artificial
  • use_flash_attn is set to True
  • packing is set to False
  • torch_dtype is set to float16, cannot be a different value
  • uses the FSDP distributed backend for multi-gpu runs by default, can also be DDP
  • multi-gpu runs with FSDP and DDP backends use 1 process per GPU (via accelerate)
  • runs 1 epoch by default, can also run a custom number of steps
  • does not save checkpoint
  • loads weights from a PVC
  • request 2 CPU cores per GPU device (with a minimum of 2 cores)
  • uses fms-acceleration plugins to perform GPTQ LoRA. Specifically:
    • auto_gptq is set to triton_v2
    • fast_kernels is set to True True True
    • fused_lora is set to auto_gptq True
    • torch_dtype is set to float16
    • loads GPTQ compatible pre-quantized weights from a PVC

For FSDP runs we use the following accelerate_config.yml YAML file:

compute_environment: LOCAL_MACHINE
distributed_type: FSDP
downcast_bf16: "no"
fsdp_config:
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_backward_prefetch: BACKWARD_PRE
  fsdp_forward_prefetch: false
  fsdp_offload_params: false
  fsdp_sharding_strategy: ${fsdp_sharding_strategy}
  fsdp_state_dict_type: ${fsdp_state_dict_type}
  fsdp_cpu_ram_efficient_loading: true
  fsdp_sync_module_states: true
  fsdp_transformer_layer_cls_to_wrap: ${accelerate_config_fsdp_transformer_layer_cls_to_wrap}
machine_rank: { $THE MACHINE RANK - always 0 for single-node runs }
main_training_function: main
mixed_precision: ${accelerate_config_mixed_precision}
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: { $SOME_PORT }
num_processes: { $NUM_GPUS }

For DDP runs we use this instead:

compute_environment: LOCAL_MACHINE
debug: False
downcast_bf16: no
distributed_type: MULTI_GPU
machine_rank: { $THE MACHINE RANK - always 0 for single-node runs }
main_training_function: main
mixed_precision: ${accelerate_config_mixed_precision}
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: { $SOME_PORT }
num_processes: { $NUM_GPUS }

Commandline:

accelerate launch --config_file ${PATH_ACCELERATE_CONFIG} --num_processes ${NUMBER_GPUS} \
  ${PATH_TO_OUR_WRAPPER_OF_FMS_HF_TUNING_SFT_TRAINER} --model_name_or_path ${MODEL} \
  --torch_dtype float16 --use_flash_attn True --training_data_path ${DATASET_PATH} \
  --response_template "\n### Response:" --dataset_text_field output --log_level debug \
  --num_train_epochs 1 --per_device_train_batch_size ${BATCH_SIZE/NUM_GPUS} \
  --max_seq_length ${MODEL_MAX_LENGTH} --eval_strategy no --output_dir ${RANDOM_DIR} \
  --gradient_accumulation_steps ${GRADIENT_ACCUMULATION_STEPS} --save_strategy no \
  --learning_rate 1e-05 --weight_decay 0.0 --warmup_ratio 0.03 --lr_scheduler_type cosine \
  --logging_steps 1 --include_tokens_per_second True --gradient_checkpointing True \
  --packing False --peft_method lora --target_modules ${SPACE SEPARATED LAYER NAMES} \
  --fp16 true --fast_kernels true true true --fused_lora auto_gptq true --auto_gptq triton_v2 \
  --optim ${OPTIM} --bf16 ${BF16} \
  --gradient_checkpointing_kwargs='{"use_reentrant": ${GRADIENT_CHECKPOINTING_USE_REENTRANT}}' \
  --fast_moe ${FAST_MOE}

Note: --fast_moe is only supported for fms-hf-tuning v2.4.0+

We use a thin wrapper of sft_trainer.py which injects a custom Callback that exports the metrics collected by AIM. You can repeat our experiments by just pointing the above command-line to sft_trainer.py from the fms-hf-tuning package.

Versioning:

Requirements

  • The S3 bucket watson.runtime.wisdom.model.us-south mounted under /ibm-research-models (instructions).
  • The PVC hf-models-pvc mounted under /hf-models-pvc - should contain the models:
    • LLaMa/models/hf/7B-gptq/
    • LLaMa/models/hf/llama3-70b-gptq/
    • LLaMa/models/hf/llama3.1-405b-gptq/
    • granite-20b-code-base-v2/step_280000_ckpt-gptq/
    • granite-34b-gptq/
    • granite-7b-base-gtpq/
    • granite-8b-code-instruct-gptq/
    • mistral-7B-v0.3-gptq/
    • mixtral_8x7b_instruct_v0.1_gptq/
  • The PVC ray-disorch-storage mounted under /data with the preprocessed artificial-dataset files (https://github.ibm.com/ai-foundation/watson-fm-stack-tracker/issues/550) under /data/fms-hf-tuning/artificial-dataset

Entity space

Required:

  • model_name: Supported models: ["llama-7b", "granite-20b-v2", "granite-7b-base", "granite-8b-code-instruct", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama3.1-405b"]
  • model_max_length: Maximum sequence length. Sequences will be right padded (and possibly truncated)
  • number_gpus: The effective number of GPUs (to be evenly distributed to number_nodes machines)
  • batch_size: the effective batch_size (will be evenly distributed to max(1, number_gpus) devices)
  • gpu_model: The value of the kubernetes node label nvidia.com/gpu.prod for example
    • NVIDIA-A100-80GB-PCIe
    • NVIDIA-A100-SXM4-80GB
    • NVIDIA-H100-PCIe

Optional:

  • dataset_id: Default is news-tokens-16384plus-entries-4096. Available options are:
    • news-chars-512-entries-4096: 4096 entries with samples of 512 + 127 (prompt) + 512 characters
    • news-chars-1024-entries-4096: 4096 entries with samples of 1024 + 127 (prompt) + 1024 characters
    • news-chars-2048-entries-4096: 4096 entries with samples of 2048 + 127 (prompt) + 2048 characters
    • news-tokens-16384plus-entries-4096: 4096 entries, each entry has least 16384 tokens when tokenized with any of the granite-13b-v2, llama-13b-v2, llama-7b, or granite-20b-v2 tokenizers
    • vision-384x384-16384plus-entries-4096: A vision dataset containing 4096 entries. Each entry includes at least 16384 tokens when tokenized with granite-vision-3.2-2b, and consists of repeated copies of a single image with dimensions 384×384.
    • vision-384x768-16384plus-entries-4096: Similar to the above, this dataset also contains 4096 entries with a minimum of 16384 tokens per entry (tokenized using granite-vision-3.2-2b). Each entry uses repeated copies of a single image sized 384×768.
  • gradient_checkpointing: Default is True. If True, use gradient checkpointing to save memory (i.e. higher batchsizes) at the expense of slower backward pass
  • gradient_accumulation_steps: Default is 4. Number of update steps to accumulate before performing a backward/update pass. Only takes effect when gradient_checkpointing is True
  • torch_dtype: Default is float16. One of float16
  • max_steps: Default is -1. The number of optimization steps to perform. Set to -1 to respect num_train_epochs instead.
  • num_train_epochs: Default is 1.0. How many epochs to run. Ignored if max_steps is greater than 0.
  • stop_after_seconds: Default is -1.0. If set, the optimizer will be asked to stop after the specified time elapses. The check is performed after the end of each training step.
  • auto_stop_method: The default value is None. This parameter defines the method used to automatically stop the fine-tuning job. Supported values are WARMUP_60S_STABLE_120S_OR_10_STEPS and None. If set to WARMUP_60S_STABLE_120S_OR_10_STEPS, the job stops after spending at least 60 seconds in the warmup phase plus the longer of 120 seconds or the duration of 10 optimization steps. This method excludes the first 60 seconds of training when calculating throughput and system metrics.
  • distributed_backend: Default is FSDP for multi-gpu measurements, None (i.e. Data Parallel (DP)) for single-gpu measurements. Which pytorch backend to use when training with multiple GPU devices.
  • number_nodes: Default is 1. If set, actuator distributes tasks on multiple nodes. Each Node will use number_gpus/number_nodes GPUs. Each Node will use 1 process for each GPU it uses
  • fms_hf_tuning_version: Default is 2.1.2. Which version of fms-hf-tuning to use. Available options are: 3.0.0, 2.8.2, 2.7.1, 2.6.0, 2.5.0, 2.4.0, 2.3.1, 2.2.1, 2.1.2, 2.1.0, 2.0.1
  • enable_roce: Default is False. This setting is only in effect for multi-node runs. It controls whether RDMA over Converged Ethernet (RoCE) is switched on or not.
  • fast_moe: Default is 0. Configures the amount of expert parallel sharding. number_gpus must be divisible by it
  • fast_kernels: Default is None. Switches on fast kernels, the value is a list with strings of boolean values for [fast_loss, fast_rms_layernorm, fast_rope_embeddings]
  • r: Default is 4. The LORA rank
  • lora_alpha: Default is 16. Scales the learning weights.
  • optim: Default is adamw_torch. The optimizer to use. Available options are adamw_hf, adamw_torch, adamw_torch_fused, adamw_torch_xla, adamw_torch_npu_fused, adamw_apex_fused, adafactor, adamw_anyprecision, adamw_torch_4bit, ademamix, sgd, adagrad, adamw_bnb_8bit, adamw_8bit, ademamix_8bit, lion_8bit, lion_32bit, paged_adamw_32bit, paged_adamw_8bit, paged_ademamix_32bit, paged_ademamix_8bit, paged_lion_32bit, paged_lion_8bit, rmsprop, rmsprop_bnb, rmsprop_bnb_8bit, rmsprop_bnb_32bit, galore_adamw, galore_adamw_8bit, galore_adafactor, galore_adamw_layerwise, galore_adamw_8bit_layerwise, galore_adafactor_layerwise, lomo, adalomo, grokadamw, schedule_free_adamw, schedule_free_sgd
  • bf16: Default is False. Whether to use bf16 (mixed) precision instead of 32-bit. Requires Ampere or higher NVIDIA add bf16 mixed precision support for NPU architecture or using CPU (use_cpu) or Ascend NPU. This is an experimental API and it may change. Can be True, False.
  • gradient_checkpointing_use_reentrant: Default is False Specify whether to use the activation checkpoint variant that requires reentrant autograd. This parameter should be passed explicitly. Torch version 2.5 will raise an exception if use_reentrant is not passed. If use_reentrant=False, checkpoint will use an implementation that does not require reentrant autograd. This allows checkpoint to support additional functionality, such as working as expected with torch.autograd.grad and support for keyword arguments input into the checkpointed function. Can be True, False.
  • fsdp_sharding_strategy: Default is FULL_SHARD. [1] FULL_SHARD (shards optimizer states, gradients and parameters), " [2] SHARD_GRAD_OP (shards optimizer states and gradients), [3] NO_SHARD (DDP), [4] HYBRID_SHARD (shards optimizer states, gradients and parameters within each node while each node has full copy), [5] HYBRID_SHARD_ZERO2 (shards optimizer states and gradients within each node while each node has full copy). For more information, please refer the official PyTorch docs.
  • fsdp_state_dict_type: Default is FULL_STATE_DICT. [1] FULL_STATE_DICT, [2] LOCAL_STATE_DICT, [3] SHARDED_STATE_DICT
  • fsdp_use_orig_params: Default is True. If True, allows non-uniform requires_grad during init, which means support for interspersed frozen and trainable parameters. (useful only when use_fsdp flag is passed).
  • accelerate_config_mixed_precision: Default is no. Whether to use mixed precision training or not. Choose from no,fp16,bf16 or fp8. fp8 requires the installation of transformers-engine.
  • accelerate_config_fsdp_transformer_layer_cls_to_wrap: Default is None. List of transformer layer class names (case-sensitive) to wrap, e.g, GraniteDecoderLayer, LlamaDecoderLayer, MistralDecoderLayer, BertLayer, GPTJBlock, T5Block ... (useful only when using FSDP)
  • dataset_text_field: Default is None. Training dataset text field containing single sequence. Either the dataset_text_field or data_formatter_template need to be supplied. For running vision language model tuning pass the column name for text data.
  • dataset_image_field: Default is None. For running vision language model tuning pass the column name of the image data in the dataset.
  • remove_unused_columns: Default is True. Remove columns not required by the model when using an nlp.Dataset.
  • dataset_kwargs_skip_prepare_dataset: Default is False. When True, configures trl to skip preparing the dataset

Hardcoded:

Sets the --target_modules layer names based on the model_name:

  • granite-8b-code-instruct: ["q_proj", "v_proj"]
  • granite-7b-base: ["q_proj", "v_proj"]
  • granite-20b-v2: ["c_attn", "c_proj"]
  • granite-34b-code-base: ["c_attn", "c_proj"]
  • llama-7b: ["q_proj", "k_proj"]
  • llama3-70b: ["q_proj", "v_proj"]
  • mistral-7b-v0.1: ["q_proj", "v_proj"]
  • mixtral-8x7b-instruct-v0.1: ["q_proj", "v_proj"]
  • llama3.1-405b: ["q_proj", "v_proj"]
  • allam-1-13b: ["q_proj", "v_proj"]

NOTE: Because running accelerate with a single gpu is unsupported, when setting number_gpus to 1 this experiment actually runs the tuning.sft_trainer script directly (i.e. a DataParallel (DP) run).

Measured properties

We use AIM to collect profiling metadata. Then we convert the timeseries that AIM collects into the metrics below.

  • gpu_compute_utilization_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_compute_utilization_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_compute_utilization_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_watts_min (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_power_watts_avg (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_power_watts_max (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_memory_utilization_peak (0.0 if not using any GPUs): peak GPU memory utilization percentage across all devices
  • cpu_compute_utilization: Measured in Percentages (0 to 100 where 100 means 1 full core) (see note 2)
  • cpu_memory_utilization: Measured in Percentages (0 to 100) taken from AIM (see note 2)
  • train_runtime: Measured in seconds
  • train_samples_per_second
  • train_steps_per_second
  • train_tokens_per_second: How many tokens (including padding tokens) the run processed every second (for FSDP this is estimated from num_gpus * rank_0_train_tokens_per_second). Omitted when stop_after_seconds is greater than 0
  • train_tokens_per_gpu_per_second (will be equal to train_tokens_per_second when number_gpus <= 1, for FSDP this is reported using just rank 0). Omitted when stop_after_seconds is greater than 0
  • dataset_tokens_per_second: How many tokens from the dataset the run processed every second (see note 3)
  • dataset_tokens_per_second_per_gpu: How many tokens from the dataset the run processed every second per GPU (see note 3)
  • is_valid (see is_valid logic)

Notes:

  • (1) They are reported as the min/max/avg of the average in the timeseries metrics that AIM collects for the in-use GPUs of a run.
  • (2) CPU compute and memory utilization are percentages. They are reported as the min/max/avg of the metrics that AIM collects for the sft_trainer.py process. The total memory capacity of the nodes varies from 100 GB to 400 GB. Currently, we do not store this information in our database.
  • (3) dataset_tokens_per_second and dataset_tokens_per_second_per_gpu take into account the tokenizer.model_max_length and max_seq_length (i.e. for each entry, we report min(len(tokens(entry["output"])), tokenizer.model_max_length, max_seq_length)).

is_valid logic

A run for an entity is invalid if:

  1. batch_size cannot be evenly divided by number_gpus (i.e. batch_size % number_gpus != 0)
  2. number_gpus cannot be evenly divided by number_nodes (i.e. number_gpus % number_nodes != 0)
  3. number_nodes must be greater than 0
  4. batch_size must be greater than 0
  5. if number_gpus is greater than 0 then gpu_model must be a non-empty string
  6. if fast_moe is set and number_gpus is not divisible by it
  7. if fast_moe is set and the num_local_experts of the Mixture of Experts (MoE) model is not divisible by fast_moe (which is interpreted as ep_degrees by fms-hf-tuning)

Runs raising the following errors are considered invalid due to running out of GPU memory:

  • torch.cuda.OutOfMemoryError
  • RuntimeError: CUDA error: an illegal memory access was encountered

Measurements raising any other exception (including for example RuntimeError containing the string NCCL Error) are considered to have Failed. They will not contain the is_valid measured property, or any other property for that matter. Failed measurements do not record any properties and can be repeated.

PT Fine-Tuning Experiments

finetune_pt_benchmark-v1.0.0

An experiment instance:

  • performs prompt-tuning fine tuning
  • the training data is artificial
  • use_flash_attn is set to True
  • packing is set to False
  • torch_dtype is set to bfloat16 by default, can also be float16
  • uses the FSDP distributed backend for multi-gpu runs by default, can also be DDP
  • multi-gpu runs with FSDP and DDP backends use 1 process per GPU (via accelerate)
  • runs 1 epoch by default, can also run a custom number of steps
  • does not save checkpoint
  • loads weights from a PVC
  • request 2 CPU cores per GPU device (with a minimum of 2 cores)

For FSDP runs we use the following accelerate_config.yml YAML file:

compute_environment: LOCAL_MACHINE
distributed_type: FSDP
downcast_bf16: "no"
fsdp_config:
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_backward_prefetch: BACKWARD_PRE
  fsdp_forward_prefetch: false
  fsdp_offload_params: false
  fsdp_sharding_strategy: ${fsdp_sharding_strategy}
  fsdp_state_dict_type: ${fsdp_state_dict_type}
  fsdp_cpu_ram_efficient_loading: true
  fsdp_sync_module_states: true
  fsdp_transformer_layer_cls_to_wrap: ${accelerate_config_fsdp_transformer_layer_cls_to_wrap}
machine_rank: { $THE MACHINE RANK - always 0 for single-node runs }
main_training_function: main
mixed_precision: ${accelerate_config_mixed_precision}
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: { $SOME_PORT }
num_processes: { $NUM_GPUS }

For DDP runs we use this instead:

compute_environment: LOCAL_MACHINE
debug: False
downcast_bf16: no
distributed_type: MULTI_GPU
machine_rank: { $THE MACHINE RANK - always 0 for single-node runs }
main_training_function: main
mixed_precision: ${accelerate_config_mixed_precision}
num_machines: 1
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false
main_process_port: { $SOME_PORT }
num_processes: { $NUM_GPUS }

Commandline:

accelerate launch --config_file ${PATH_ACCELERATE_CONFIG} --num_processes ${NUMBER_GPUS} \
  ${PATH_TO_OUR_WRAPPER_OF_FMS_HF_TUNING_SFT_TRAINER} --model_name_or_path ${MODEL} \
  --torch_dtype bfloat16 --use_flash_attn True --training_data_path ${DATASET_PATH} \
  --response_template "\n### Response:" --dataset_text_field output --log_level debug \
  --num_train_epochs 1 --per_device_train_batch_size ${BATCH_SIZE/NUM_GPUS} \
  --max_seq_length ${MODEL_MAX_LENGTH} --eval_strategy no --output_dir ${RANDOM_DIR} \
  --gradient_accumulation_steps ${GRADIENT_ACCUMULATION_STEPS} --save_strategy no \
  --learning_rate 1e-05 --weight_decay 0.0 --warmup_ratio 0.03 --lr_scheduler_type cosine \
  --logging_steps 1 --include_tokens_per_second True --gradient_checkpointing True \
  --packing False --peft_method none \
  --fast_moe ${FAST_MOE}

Note: --fast_moe is only supported for fms-hf-tuning v2.4.0+

We use a thin wrapper of sft_trainer.py which injects a custom Callback that exports the metrics collected by AIM. You can repeat our experiments by just pointing the above command-line to sft_trainer.py from the fms-hf-tuning package.

Versioning:

Requirements

  • The S3 bucket watson.runtime.wisdom.model.us-south mounted under /ibm-research-models (instructions).
  • The PVC hf-models-pvc mounted under /hf-models-pvc - should contain the models:
    • LLaMa/models/hf/13B/
    • LLaMa/models/hf/7B/
    • LLaMa/models/hf/llama2-70b/
    • LLaMa/models/hf/llama3-70b/
    • LLaMa/models/hf/llama3-8b/
    • LLaMa/models/hf/llama3.1-405b/
    • LLaMa/models/hf/llama3.1-70b/
    • LLaMa/models/hf/llama3.1-8b/
    • Mixtral-8x7B-Instruct-v0.1/
    • allam-1-13b-instruct-20240607/
    • granite-13b-base-v2/step_300000_ckpt/
    • granite-20b-code-base-v2/step_280000_ckpt/
    • granite-34b-code-base/
    • granite-8b-code-base/
    • granite-8b-japanese-base-v1-llama/
    • mistralai-mistral-7b-v0.1/
    • mistral-large/fp16_240620
  • The PVC ray-disorch-storage mounted under /data with the preprocessed artificial-dataset files (https://github.ibm.com/ai-foundation/watson-fm-stack-tracker/issues/550) under /data/fms-hf-tuning/artificial-dataset

Entity space

Required:

  • model_name: Supported models: ["granite-3b-1.5", "hf-tiny-model-private/tiny-random-BloomForCausalLM", "llama-7b", "granite-13b-v2", "llama-13b", "granite-20b-v2", "granite-7b-base", "granite-8b-japanese", "granite-8b-code-base", "granite-34b-code-base", "mistral-7b-v0.1", "llama3-8b", "llama3-70b", "mixtral-8x7b-instruct-v0.1", "llama2-70b", "llama3.1-8b", "llama3.1-70b", "llama3.1-405b", "granite-3b-code-base-128k", "granite-8b-code-base-128k", "allam-1-13b", "granite-3-8b", "granite-3.1-2b", "granite-3.1-8b-instruct", "mistral-123b-v2", "granite-3.1-3b-a800m-instruct", "granite-vision-3.2-2b", "smollm2-135m", "llava-v1.6-mistral-7b", "granite-4.0-micro", "granite-4.0-h-1b", "granite-4.0-350m", "granite-4.0-h-small", "granite-4.0-h-micro", "granite-4.0-h-tiny"]
  • model_max_length: Maximum sequence length. Sequences will be right padded (and possibly truncated)
  • number_gpus: The effective number of GPUs (to be evenly distributed to number_nodes machines)
  • batch_size: the effective batch_size (will be evenly distributed to max(1, number_gpus) devices)
  • gpu_model: The value of the kubernetes node label nvidia.com/gpu.prod for example
    • NVIDIA-A100-80GB-PCIe
    • NVIDIA-A100-SXM4-80GB
    • NVIDIA-H100-PCIe

Optional:

  • dataset_id: Default is news-tokens-16384plus-entries-4096. Available options are:
    • news-chars-512-entries-4096: 4096 entries with samples of 512 + 127 (prompt) + 512 characters
    • news-chars-1024-entries-4096: 4096 entries with samples of 1024 + 127 (prompt) + 1024 characters
    • news-chars-2048-entries-4096: 4096 entries with samples of 2048 + 127 (prompt) + 2048 characters
    • news-tokens-16384plus-entries-4096: 4096 entries, each entry has least 16384 tokens when tokenized with any of the granite-13b-v2, llama-13b-v2, llama-7b, or granite-20b-v2 tokenizers
    • vision-384x384-16384plus-entries-4096: A vision dataset containing 4096 entries. Each entry includes at least 16384 tokens when tokenized with granite-vision-3.2-2b, and consists of repeated copies of a single image with dimensions 384×384.
    • vision-384x768-16384plus-entries-4096: Similar to the above, this dataset also contains 4096 entries with a minimum of 16384 tokens per entry (tokenized using granite-vision-3.2-2b). Each entry uses repeated copies of a single image sized 384×768.
  • gradient_checkpointing: Default is True. If True, use gradient checkpointing to save memory (i.e. higher batchsizes) at the expense of slower backward pass
  • gradient_accumulation_steps: Default is 4. Number of update steps to accumulate before performing a backward/update pass. Only takes effect when gradient_checkpointing is True
  • torch_dtype: Default is bfloat16. One of bfloat16, float32, float16
  • max_steps: Default is -1. The number of optimization steps to perform. Set to -1 to respect num_train_epochs instead.
  • num_train_epochs: Default is 1.0. How many epochs to run. Ignored if max_steps is greater than 0.
  • stop_after_seconds: Default is -1.0. If set, the optimizer will be asked to stop after the specified time elapses. The check is performed after the end of each training step.
  • auto_stop_method: The default value is None. This parameter defines the method used to automatically stop the fine-tuning job. Supported values are WARMUP_60S_STABLE_120S_OR_10_STEPS and None. If set to WARMUP_60S_STABLE_120S_OR_10_STEPS, the job stops after spending at least 60 seconds in the warmup phase plus the longer of 120 seconds or the duration of 10 optimization steps. This method excludes the first 60 seconds of training when calculating throughput and system metrics.
  • distributed_backend: Default is FSDP for multi-gpu measurements, None (i.e. Data Parallel (DP)) for single-gpu measurements. Which pytorch backend to use when training with multiple GPU devices.
  • number_nodes: Default is 1. If set, actuator distributes tasks on multiple nodes. Each Node will use number_gpus/number_nodes GPUs. Each Node will use 1 process for each GPU it uses
  • fms_hf_tuning_version: Default is 2.1.2. Which version of fms-hf-tuning to use. Available options are: 3.0.0, 2.8.2, 2.7.1, 2.6.0, 2.5.0, 2.4.0, 2.3.1, 2.2.1, 2.1.2, 2.1.0, 2.0.1
  • enable_roce: Default is False. This setting is only in effect for multi-node runs. It controls whether RDMA over Converged Ethernet (RoCE) is switched on or not.
  • fast_moe: Default is 0. Configures the amount of expert parallel sharding. number_gpus must be divisible by it
  • fast_kernels: Default is None. Switches on fast kernels, the value is a list with strings of boolean values for [fast_loss, fast_rms_layernorm, fast_rope_embeddings]
  • optim: Default is adamw_torch. The optimizer to use. Available options are adamw_hf, adamw_torch, adamw_torch_fused, adamw_torch_xla, adamw_torch_npu_fused, adamw_apex_fused, adafactor, adamw_anyprecision, adamw_torch_4bit, ademamix, sgd, adagrad, adamw_bnb_8bit, adamw_8bit, ademamix_8bit, lion_8bit, lion_32bit, paged_adamw_32bit, paged_adamw_8bit, paged_ademamix_32bit, paged_ademamix_8bit, paged_lion_32bit, paged_lion_8bit, rmsprop, rmsprop_bnb, rmsprop_bnb_8bit, rmsprop_bnb_32bit, galore_adamw, galore_adamw_8bit, galore_adafactor, galore_adamw_layerwise, galore_adamw_8bit_layerwise, galore_adafactor_layerwise, lomo, adalomo, grokadamw, schedule_free_adamw, schedule_free_sgd
  • bf16: Default is False. Whether to use bf16 (mixed) precision instead of 32-bit. Requires Ampere or higher NVIDIA add bf16 mixed precision support for NPU architecture or using CPU (use_cpu) or Ascend NPU. This is an experimental API and it may change. Can be True, False.
  • gradient_checkpointing_use_reentrant: Default is False Specify whether to use the activation checkpoint variant that requires reentrant autograd. This parameter should be passed explicitly. Torch version 2.5 will raise an exception if use_reentrant is not passed. If use_reentrant=False, checkpoint will use an implementation that does not require reentrant autograd. This allows checkpoint to support additional functionality, such as working as expected with torch.autograd.grad and support for keyword arguments input into the checkpointed function. Can be True, False.
  • fsdp_sharding_strategy: Default is FULL_SHARD. [1] FULL_SHARD (shards optimizer states, gradients and parameters), " [2] SHARD_GRAD_OP (shards optimizer states and gradients), [3] NO_SHARD (DDP), [4] HYBRID_SHARD (shards optimizer states, gradients and parameters within each node while each node has full copy), [5] HYBRID_SHARD_ZERO2 (shards optimizer states and gradients within each node while each node has full copy). For more information, please refer the official PyTorch docs.
  • fsdp_state_dict_type: Default is FULL_STATE_DICT. [1] FULL_STATE_DICT, [2] LOCAL_STATE_DICT, [3] SHARDED_STATE_DICT
  • fsdp_use_orig_params: Default is True. If True, allows non-uniform requires_grad during init, which means support for interspersed frozen and trainable parameters. (useful only when use_fsdp flag is passed).
  • accelerate_config_mixed_precision: Default is no. Whether to use mixed precision training or not. Choose from no,fp16,bf16 or fp8. fp8 requires the installation of transformers-engine.
  • accelerate_config_fsdp_transformer_layer_cls_to_wrap: Default is None. List of transformer layer class names (case-sensitive) to wrap, e.g, GraniteDecoderLayer, LlamaDecoderLayer, MistralDecoderLayer, BertLayer, GPTJBlock, T5Block ... (useful only when using FSDP)
  • dataset_text_field: Default is None. Training dataset text field containing single sequence. Either the dataset_text_field or data_formatter_template need to be supplied. For running vision language model tuning pass the column name for text data.
  • dataset_image_field: Default is None. For running vision language model tuning pass the column name of the image data in the dataset.
  • remove_unused_columns: Default is True. Remove columns not required by the model when using an nlp.Dataset.
  • dataset_kwargs_skip_prepare_dataset: Default is False. When True, configures trl to skip preparing the dataset

NOTE: Because running accelerate with a single gpu is unsupported, when setting number_gpus to 1 this experiment actually runs the tuning.sft_trainer script directly (i.e. a DataParallel (DP) run).

Measured properties

We use AIM to collect profiling metadata. Then we convert the timeseries that AIM collects into the metrics below.

  • gpu_compute_utilization_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_compute_utilization_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_compute_utilization_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_memory_utilization_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_min (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_avg (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_percent_max (0.0 if not using any GPUs): Measured in Percentages (0 to 100) (see note 1)
  • gpu_power_watts_min (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_power_watts_avg (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_power_watts_max (0.0 if not using any GPUs): Measured in Watts (see Note 1)
  • gpu_memory_utilization_peak (0.0 if not using any GPUs): peak GPU memory utilization percentage across all devices
  • cpu_compute_utilization: Measured in Percentages (0 to 100 where 100 means 1 full core) (see note 2)
  • cpu_memory_utilization: Measured in Percentages (0 to 100) taken from AIM (see note 2)
  • train_runtime: Measured in seconds
  • train_samples_per_second
  • train_steps_per_second
  • train_tokens_per_second: How many tokens (including padding tokens) the run processed every second (for FSDP this is estimated from num_gpus * rank_0_train_tokens_per_second). Omitted when stop_after_seconds is greater than 0
  • train_tokens_per_gpu_per_second (will be equal to train_tokens_per_second when number_gpus <= 1, for FSDP this is reported using just rank 0). Omitted when stop_after_seconds is greater than 0
  • dataset_tokens_per_second: How many tokens from the dataset the run processed every second (see note 3)
  • dataset_tokens_per_second_per_gpu: How many tokens from the dataset the run processed every second per GPU (see note 3)
  • is_valid (see is_valid logic)

Notes:

  • (1) They are reported as the min/max/avg of the average in the timeseries metrics that AIM collects for the in-use GPUs of a run.
  • (2) CPU compute and memory utilization are percentages. They are reported as the min/max/avg of the metrics that AIM collects for the sft_trainer.py process. The total memory capacity of the nodes varies from 100 GB to 400 GB. Currently, we do not store this information in our database.
  • (3) dataset_tokens_per_second and dataset_tokens_per_second_per_gpu take into account the tokenizer.model_max_length and max_seq_length (i.e. for each entry, we report min(len(tokens(entry["output"])), tokenizer.model_max_length, max_seq_length)).

is_valid logic

A run for an entity is invalid if:

  1. batch_size cannot be evenly divided by number_gpus (i.e. batch_size % number_gpus != 0)
  2. number_gpus cannot be evenly divided by number_nodes (i.e. number_gpus % number_nodes != 0)
  3. number_nodes must be greater than 0
  4. batch_size must be greater than 0
  5. if number_gpus is greater than 0 then gpu_model must be a non-empty string
  6. if fast_moe is set and number_gpus is not divisible by it
  7. if fast_moe is set and the num_local_experts of the Mixture of Experts (MoE) model is not divisible by fast_moe (which is interpreted as ep_degrees by fms-hf-tuning)

Runs raising the following errors are considered invalid due to running out of GPU memory:

  • torch.cuda.OutOfMemoryError
  • RuntimeError: CUDA error: an illegal memory access was encountered

Measurements raising any other exception (including for example RuntimeError containing the string NCCL Error) are considered to have Failed. They will not contain the is_valid measured property, or any other property for that matter. Failed measurements do not record any properties and can be repeated.

Project details

Verified details

These details have been verified by PyPI
Maintainers
Avatar for alessandropomponio-ibm from gravatar.com alessandropomponio-ibm

Unverified details

These details have not been verified by PyPI

Release history Release notifications | RSS feed

1.7.0

1.6.0

1.5.0

1.4.1

1.4.0

1.3.3

1.3.2

1.3.1

1.3.0

1.2.4

1.2.3

1.2.2

This version

1.2.1

1.2.0

1.1.0

1.0.1

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ado_sfttrainer-1.2.1.tar.gz (109.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

ado_sfttrainer-1.2.1-py3-none-any.whl (144.3 kB view details)

Uploaded Python 3

File details

Details for the file ado_sfttrainer-1.2.1.tar.gz.

File metadata

  • Download URL: ado_sfttrainer-1.2.1.tar.gz
  • Upload date:
  • Size: 109.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.9.7

File hashes

Hashes for ado_sfttrainer-1.2.1.tar.gz
Algorithm Hash digest
SHA256 4f67f4776e7eb0a8b0f92502dc1702e981939b23df9a57282e46a8a17d7e5a70
MD5 7c2942a78a7029b0f065b4c4f1a281d3
BLAKE2b-256 e4ffd8efeb6d115e9aa17cd7e697ded46c9ac960ce40df68c019dc4bd8dcde69

See more details on using hashes here.

File details

Details for the file ado_sfttrainer-1.2.1-py3-none-any.whl.

File metadata

File hashes

Hashes for ado_sfttrainer-1.2.1-py3-none-any.whl
Algorithm Hash digest
SHA256 c3a42b9217acc0e51358f9ba2d8e2f41b33d36d16fcd722c7b367a0bc8cdc09d
MD5 bd30cfa5d4490a3dbf909cc1fb007ccb
BLAKE2b-256 3931400ffa490bd6b1e92d4b77086e215860cc10b8a6d4164da32e9bed9d4650

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page