qwerrwe / README.md
Nanobit's picture
Update 4bit notes
a9e502e
|
raw
history blame
10.4 kB

Axolotl

axolotl

One repo to finetune them all!

Go ahead and axolotl questions!!

Axolotl supports

fp16/fp32 fp16/fp32 w/ lora 4bit-quant 4bit-quant w/flash attention flash attention xformers attention
llama βœ… βœ… βœ… βœ… βœ… βœ…
Pythia βœ… βœ… ❌ ❌ ❌ ❓
cerebras βœ… βœ… ❌ ❌ ❌ ❓
mpt βœ… ❌ ❌ ❌ ❌ ❓

Quickstart ⚑

Requirements: Python 3.9.

git clone https://github.com/OpenAccess-AI-Collective/axolotl

pip3 install -e .[int4]

accelerate config

# finetune
accelerate launch scripts/finetune.py examples/4bit-lora-7b/config.yml

# inference
accelerate launch scripts/finetune.py examples/4bit-lora-7b/config.yml \
    --inference --lora_model_dir="./llama-7b-lora-int4"

Installation

Environment

  • Docker

    docker run --gpus '"all"' --rm -it winglian/axolotl:main
    
    • winglian/axolotl:dev: dev branch
    • winglian/axolotl-runpod:main: for runpod
  • Conda/Pip venv

    1. Install python 3.9

    2. Install python dependencies with ONE of the following:

      • pip3 install -e .[int4] (recommended)
      • pip3 install -e .[int4_triton]
      • pip3 install -e .

Dataset

Have dataset(s) in one of the following format (JSONL recommended):

  • alpaca: instruction; input(optional)
    {"instruction": "...", "input": "...", "output": "..."}
    
  • sharegpt: conversations
    {"conversations": [{"from": "...", "value": "..."}]}
    
  • completion: raw corpus
    {"text": "..."}
    
See other formats
  • jeopardy: question and answer
    {"question": "...", "category": "...", "answer": "..."}
    
  • oasst: instruction
    {"INSTRUCTION": "...", "RESPONSE": "..."}
    
  • gpteacher: instruction; input(optional)
    {"instruction": "...", "input": "...", "response": "..."}
    
  • reflection: instruction with reflect; input(optional)
    {"instruction": "...", "input": "...", "output": "...", "reflection": "...", "corrected": "..."}
    
  • explainchoice: question, choices, (solution OR explanation)
    {"question": "...", "choices": ["..."], "solution": "...", "explanation": "..."}
    
  • concisechoice: question, choices, (solution OR explanation)
    {"question": "...", "choices": ["..."], "solution": "...", "explanation": "..."}
    
  • summarizetldr: article and summary
    {"article": "...", "summary": "..."}
    

Have some new format to propose? Check if it's already defined in data.py in dev branch!

Optionally, download some datasets, see data/README.md

Config

See sample configs in configs folder or examples for quick start. It is recommended to duplicate and modify to your needs. The most important options are:

  • model

    base_model: ./llama-7b-hf # local or huggingface repo
    

    Note: The code will load the right architecture.

  • dataset

    datasets:
      - path: vicgalle/alpaca-gpt4 # local or huggingface repo
        type: alpaca # format from earlier
    sequence_len: 2048 # max token length / prompt
    
  • loading

    load_in_4bit: true
    load_in_8bit: true
    bf16: true # require >=ampere
    fp16: true
    tf32: true # require >=ampere
    

    Note: Repo does not do 4-bit quantization.

  • lora

    adapter: lora # qlora or leave blank for full finetune
    lora_r: 8
    lora_alpha: 16
    lora_dropout: 0.05
    lora_target_modules:
      - q_proj
      - v_proj
    
All yaml options
# this is the huggingface model that contains *.pt, *.safetensors, or *.bin files
# this can also be a relative path to a model on disk
base_model: ./llama-7b-hf
# you can specify an ignore pattern if the model repo contains more than 1 model type (*.pt, etc)
base_model_ignore_patterns:
# if the base_model repo on hf hub doesn't include configuration .json files,
# you can set that here, or leave this empty to default to base_model
base_model_config: ./llama-7b-hf
# If you want to specify the type of model to load, AutoModelForCausalLM is a good choice too
model_type: AutoModelForCausalLM
# Corresponding tokenizer for the model AutoTokenizer is a good choice
tokenizer_type: AutoTokenizer
# Trust remote code for untrusted source
trust_remote_code:

# whether you are training a 4-bit GPTQ quantized model
load_4bit: true
gptq_groupsize: 128 # group size
gptq_model_v1: false # v1 or v2

# this will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer
load_in_8bit: true
# use bitsandbytes 4 bit
load_in_4bit:

# Use CUDA bf16
bf16: true # bool or 'full' for `bf16_full_eval`. require >=ampere
# Use CUDA fp16
fp16: true
# Use CUDA tf32
tf32: true # require >=ampere

# a list of one or more datasets to finetune the model with
datasets:
  # this can be either a hf dataset, or relative path
  - path: vicgalle/alpaca-gpt4
  # The type of prompt to use for training. [alpaca, sharegpt, gpteacher, oasst, reflection]
    type: alpaca # format OR format:prompt_style (chat/instruct)
    data_files: # path to source data files
    shards: # true if use subset data. make sure to set `shards` param also
shards: # number of shards to split dataset into

# 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
# How much of the dataset to set aside as evaluation. 1 = 100%, 0.50 = 50%, etc
val_set_size: 0.04

# 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
# max sequence length to concatenate training samples together up to
# inspired by StackLLaMA. see https://huggingface.co/blog/stackllama#supervised-fine-tuning
max_packed_sequence_len: 1024

# 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
# lora hyperparameters
lora_model_dir:
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_modules_to_save:
#  - embed_tokens
#  - lm_head
lora_out_dir:
lora_fan_in_fan_out: false

# wandb configuration if you're using it
wandb_mode:
wandb_project:
wandb_watch:
wandb_run_id:
wandb_log_model: # 'checkpoint'

# where to save the finished model to
output_dir: ./completed-model

# training hyperparameters
batch_size: 8
micro_batch_size: 2
eval_batch_size: 2
num_epochs: 3
warmup_steps: 100
learning_rate: 0.00003
logging_steps:

# whether to mask out or include the human's prompt from the training labels
train_on_inputs: false
# don't use this, leads to wonky training (according to someone on the internet)
group_by_length: false

# does not work with current implementation of 4-bit LoRA
gradient_checkpointing: 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:

# for one_cycle optim
lr_div_factor: # learning rate div factor

# for log_sweep optim
log_sweep_min_lr:
log_sweep_max_lr:

# specify optimizer
optimizer:
# specify weight decay
weight_decay:

# whether to use xformers attention patch https://github.com/facebookresearch/xformers:
xformers_attention:
# whether to use flash attention patch https://github.com/HazyResearch/flash-attention:
flash_attention:  # require a100 for llama

# 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
special_tokens:
  # bos_token: "<s>"
  # eos_token: "</s>"
  # unk_token: "<unk>"
# add extra tokens
tokens:

# FSDP
fsdp:
fsdp_config:

# Deepspeed
deepspeed:

# Path to torch distx for optim 'adamw_anyprecision'
torchdistx_path:

# Set padding for data collator to 'longest'
collator_pad_to_longest:

# Debug mode
debug:

# Seed
seed:

Accelerate

Configure accelerate

accelerate config

# Edit manually
# nano ~/.cache/huggingface/accelerate/default_config.yaml

Train

Run

accelerate launch scripts/finetune.py configs/your_config.yml

Inference

Pass the appropriate flag to the train command:

  • Pretrained LORA:
    --inference --lora_model_dir ./completed-model
    
  • Full weights finetune:
    --inference --base_model ./completed-model
    

Merge LORA to base

Add below flag to train command above

--merge_lora --lora_model_dir="./completed-model"

Common Errors 🧰

Cuda out of memory

Please reduce any below

  • micro_batch_size
  • eval_batch_size
  • sequence_len

RuntimeError: expected scalar type Float but found Half

Try set fp16: true

Contributing 🀝

Bugs? Please check for open issue else create a new Issue.

PRs are greatly welcome!