PraisonAI 一个快速ai 开发集成工具

PraisonAI 一个快速ai 开发集成工具，在新版本中PraisonAI 基于unsloth 提供了快速大模型微调训练能力

训练参考使用

核心是包装的unsloth

• 命令

praisonai train --model unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit --dataset yahma/alpaca-cleaned -hf  xxx  --ollama xxxxx

• 效果

启动

训练

 

说明

PraisonAI 这个工具的哥们经常在youtube 上分享一些关于ai 大模型周边的东西，值得看看,praisonai train 核心是对于unsloth的包装，同时这个工具还抽象一个一个配置文件，方便使用

• 参考配置文件模版

ollama_save: "true"

huggingface_save: "true"

train: "true"

 

model_name: "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit"

hf_model_name: "mervinpraison/llama-3.1-tamilan-8B-test"

ollama_model: "mervinpraison/llama3.1-tamilan-test"

model_parameters: "8b"

 

max_seq_length: 2048

load_in_4bit: true

lora_r: 16

lora_target_modules:

  - "q_proj"

  - "k_proj"

  - "v_proj"

  - "o_proj"

  - "gate_proj"

  - "up_proj"

  - "down_proj"

lora_alpha: 16

lora_dropout: 0

lora_bias: "none"

use_gradient_checkpointing: "unsloth"

random_state: 3407

use_rslora: false

loftq_config: null

 

dataset:

  - name: "yahma/alpaca-cleaned"

    split_type: "train"

    processing_func: "format_prompts"

    rename:

      input: "input"

      output: "output"

      instruction: "instruction"

    filter_data: false

    filter_column_value: "id"

    filter_value: "alpaca"

    num_samples: 20000

 

dataset_text_field: "text"

dataset_num_proc: 2

packing: false

 

per_device_train_batch_size: 2

gradient_accumulation_steps: 2

warmup_steps: 5

num_train_epochs: 1

max_steps: 10

learning_rate: 2.0e-4

logging_steps: 1

optim: "adamw_8bit"

weight_decay: 0.01

lr_scheduler_type: "linear"

seed: 3407

output_dir: "outputs"

 

quantization_method: 

  - "q4_k_m"

参考资料

praisonai/train.py
https://github.com/unslothai/unsloth
https://unsloth.ai/
https://github.com/MervinPraison/PraisonAI
https://docs.praison.ai/