模型中文名：小小Agent，英文名：SmallSmallAgent
模型简介
该模型“小小Agent”是由西安交通大学”大模型实训队“在2024年1月份于阿里与魔搭社区举行的“大模型进校园”实战训练营·西交大站中微调开发得到的Agent模型。他主要有如下几个功能：
1⃣️调用多种API，包括语⾳⽣成, 图像⽣成, python解析器(数学计算/代码运⾏)等多种API?
2⃣️具有自我认知能力?
3⃣️具有多轮对话能力?
实验环境
我们在Modelscope提供的单卡A10环境上进行微调
训练方法
1⃣️数据集：采用了damo-agent-mini-zh数据集，并将自我认知能力的数据集采样数设置为500
2⃣️微调的模型：我们基于Qwen-7b-chat模型进行微调
3⃣️ 训练过程
The following bitsandbytes quantization config was used during training:
- quant_method: bitsandbytes
- loadin8bit: False
- loadin4bit: True
- llmint8threshold: 6.0
- llmint8skip_modules: None
- llmint8enablefp32cpu_offload: False
- llmint8hasfp16weight: False
- bnb4bitquant_type: nf4
- bnb4bitusedoublequant: True
- bnb4bitcompute_dtype: bfloat16
4⃣️ 微调框架版本
- SWIFT 1.5.1
推理效果
自我认知能力
图片生成能力
语音生成能力
python解释器API调用能力+多轮对话能力
示例代码
1⃣️sft.sh
PYTHONPATH=../../.. \ CUDAVISIBLEDEVICES=0 \ python ../../../llmsft.py \ --modelidorpath qwen/Qwen-7B-Chat \ --modelrevision master \ --sfttype lora \ --tunerbackend swift \ --templatetype qwen \ --system '你是名为小小agent的大语言模型，是由大模型实训队于2024年1月开发的。你是一个helpful assistant。' \ --dtype AUTO \ --outputdir output \ --dataset damo-agent-mini-zh \ --traindatasetsample 20000 \ --numtrainepochs 1 \ --maxlength 4096 \ --checkdatasetstrategy warning \ --quantizationbit 4 \ --bnb4bitcompdtype AUTO \ --lorarank 8 \ --loraalpha 32 \ --loradropoutp 0.05 \ --loratargetmodules ALL \ --gradientcheckpointing true \ --batchsize 1 \ --weightdecay 0.01 \ --learningrate 1e-4 \ --gradientaccumulationsteps 8 \ --maxgradnorm 0.5 \ --warmupratio 0.03 \ --evalsteps 100 \ --savesteps 100 \ --savetotallimit 2 \ --loggingsteps 10 \ --useflashattn true \ --pushtohub false \ --hubmodelid pushtest \ --hubprivaterepo true \ --hubtoken 'dc2af58d-7c8a-4f45-8bcb-b0dc9be32f2e' \
2⃣️sft.py
import inspect import os from functools import partial from typing import Any, Dict, Union
import json import numpy as np import torch from modelscope import BitsAndBytesConfig, GenerationConfig
from swift.trainers import Seq2SeqTrainer, Seq2SeqTrainingArguments from swift.utils import (checkjsonformat, computeaccmetrics, computenlgmetrics, getdistsetting, getlogger, getmain, getmodelinfo, isddpplusmp, isdist, ismaster, plotimages, preprocesslogitsformetrics, seedeverything, showlayers) from .tuner import preparemodel from .utils import (TEMPLATEMAPPING, LazyLLMDataset, SftArguments, Template, addselfcognitiondataset, datasetmap, getadditionalsavedfiles, getdataset, getmodeltokenizer, gettemplate, gettimeinfo, printexample, setgenerationconfig, sortbymaxlength, stat_dataset)
logger = get_logger()
def llmsft(args: SftArguments) -> Dict[str, Union[str, Any]]: logger.info(f'args: {args}') print(f'devicecount: {torch.cuda.devicecount()}') rank, localrank, worldsize, localworldsize = getdistsetting() print(f'rank: {rank}, localrank: {localrank}, ' f'worldsize: {worldsize}, localworldsize: {localworldsize}') seedeverything(args.seed)
# Loading Model and Tokenizer
model_kwargs = {'low_cpu_mem_usage': True}
if is_dist() and not is_ddp_plus_mp():
model_kwargs['device_map'] = {'': local_rank}
else:
model_kwargs['device_map'] = 'auto'
if args.load_in_8bit or args.load_in_4bit:
quantization_config = BitsAndBytesConfig(
args.load_in_8bit,
args.load_in_4bit,
bnb_4bit_compute_dtype=args.bnb_4bit_compute_dtype,
bnb_4bit_quant_type=args.bnb_4bit_quant_type,
bnb_4bit_use_double_quant=args.bnb_4bit_use_double_quant)
logger.info(f'quantization_config: {quantization_config.__dict__}')
model_kwargs['quantization_config'] = quantization_config
kwargs = {}
if args.use_flash_attn is not None:
kwargs['use_flash_attn'] = args.use_flash_attn
if args.model_cache_dir is not None:
kwargs['model_dir'] = args.model_cache_dir
model, tokenizer = get_model_tokenizer(args.model_type, args.torch_dtype,
model_kwargs, **kwargs)
logger.debug(f'device_map: {dict(model.hf_device_map)}')
logger.info(f'model_config: {model.config}')
generation_config = GenerationConfig(
max_new_tokens=args.max_new_tokens,
temperature=args.temperature,
top_k=args.top_k,
top_p=args.top_p,
do_sample=args.do_sample,
repetition_penalty=args.repetition_penalty,
num_beams=args.num_beams,
pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id)
logger.info(f'generation_config: {generation_config}')
set_generation_config(model, generation_config)
# Preparing LoRA
model, callbacks = prepare_model(model, args)
show_layers(model)
model_info = get_model_info(model)
logger.info(model_info)
logger.info(model)
# Loading Dataset
random_state = np.random.RandomState(args.dataset_seed)
train_dataset, val_dataset = get_dataset(
args.dataset,
args.dataset_test_ratio,
random_state,
check_dataset_strategy=args.check_dataset_strategy)
val_dataset_sample = args.val_dataset_sample
if train_dataset is not None and args.train_dataset_sample >= 0:
train_dataset_sample = min(args.train_dataset_sample,
train_dataset.shape[0])
if train_dataset.shape[0] > train_dataset_sample:
logger.info(f'train_dataset_sample: {train_dataset_sample}')
train_idxs = random_state.permutation(train_dataset_sample)
train_dataset = train_dataset.select(train_idxs)
if val_dataset_sample is None:
val_dataset_sample = max(
int(train_dataset_sample * args.dataset_test_ratio), 1)
if val_dataset is not None and val_dataset_sample is not None and val_dataset_sample >= 0:
if val_dataset.shape[0] > val_dataset_sample:
logger.info(f'val_dataset_sample: {val_dataset_sample}')
val_dataset = val_dataset.select(range(val_dataset_sample))
# add self-cognition dataset
if args.self_cognition_sample > 0:
train_dataset = add_self_cognition_dataset(train_dataset,
args.self_cognition_sample,
args.model_name,
args.model_author)
logger.info(f'train_dataset: {train_dataset}')
logger.info(f'val_dataset: {val_dataset}')
template_kwargs = {}
template_info = TEMPLATE_MAPPING[args.template_type]
use_model = template_info.get('use_model', False)
if use_model:
template_kwargs['model'] = model
template: Template = get_template(args.template_type, tokenizer,
args.system, args.max_length,
args.truncation_strategy,
**template_kwargs)
args.system = template.default_system
logger.info(f'system: {args.system}')
logger.info(f'args.lazy_tokenize: {args.lazy_tokenize}')
if not args.lazy_tokenize:
dataset_info = {}
logger.info(f'Using num_proc: {args.preprocess_num_proc}')
train_dataset = dataset_map(train_dataset, template.encode,
args.preprocess_num_proc)
if val_dataset is not None:
val_dataset = dataset_map(val_dataset, template.encode,
args.preprocess_num_proc)
if args.test_oom_error:
train_dataset = sort_by_max_length(train_dataset, 20000)
# Data analysis
td0, tkwargs0 = train_dataset.data[0]
print_example(td0, tokenizer, tkwargs0)
dataset_info['train_dataset'] = stat_dataset(train_dataset)
if val_dataset is not None:
dataset_info['val_dataset'] = stat_dataset(val_dataset)
else:
dataset_info = None
td0, tkwargs0 = template.encode(train_dataset[0])
print_example(td0, tokenizer, tkwargs0)
train_dataset = LazyLLMDataset(train_dataset, template)
val_dataset = LazyLLMDataset(val_dataset, template)
padding_to = args.max_length if args.sft_type == 'longlora' else None
data_collator = partial(template.data_collator, padding_to=padding_to)
# Setting training_args
evaluation_strategy = args.evaluation_strategy
load_best_model_at_end = True
if val_dataset is None:
evaluation_strategy = 'no'
if evaluation_strategy == 'no':
load_best_model_at_end = False
additional_saved_files = []
if args.sft_type == 'full':
additional_saved_files = get_additional_saved_files(args.model_type)
kwargs = {}
parameters = inspect.signature(
Seq2SeqTrainingArguments.__init__).parameters
for key in ['neftune_noise_alpha']:
if key in parameters:
kwargs[key] = getattr(args, key)
training_args = Seq2SeqTrainingArguments(
output_dir=args.output_dir,
evaluation_strategy=evaluation_strategy,
logging_dir=args.logging_dir,
per_device_train_batch_size=args.batch_size,
per_device_eval_batch_size=args.eval_batch_size,
gradient_accumulation_steps=args.gradient_accumulation_steps,
learning_rate=args.learning_rate,
weight_decay=args.weight_decay,
max_grad_norm=args.max_grad_norm,
num_train_epochs=args.num_train_epochs,
max_steps=args.max_steps,
lr_scheduler_type=args.lr_scheduler_type,
warmup_ratio=args.warmup_ratio,
logging_steps=args.logging_steps,
save_strategy=args.save_strategy,
save_steps=args.save_steps,
save_total_limit=args.save_total_limit,
remove_unused_columns=False,
bf16=args.bf16,
fp16=args.fp16,
eval_steps=args.eval_steps,
dataloader_num_workers=args.dataloader_num_workers,
load_best_model_at_end=load_best_model_at_end,
metric_for_best_model='rouge-l'
if args.predict_with_generate else 'loss',
greater_is_better=args.predict_with_generate,
sortish_sampler=True,
optim=args.optim,
adam_beta1=args.adam_beta1,
adam_beta2=args.adam_beta2,
hub_model_id=args.hub_model_id,
hub_private_repo=args.hub_private_repo,
push_hub_strategy=args.push_hub_strategy,
hub_token=args.hub_token,
push_to_hub=args.push_to_hub,
resume_from_checkpoint=args.resume_from_checkpoint,
ddp_backend=args.ddp_backend,
gradient_checkpointing=args.gradient_checkpointing,
predict_with_generate=args.predict_with_generate,
generation_config=generation_config,
local_rank=local_rank,
save_only_model=args.save_only_model,
train_sampler_random=args.train_sampler_random,
report_to=args.report_to,
deepspeed=args.deepspeed,
additional_saved_files=additional_saved_files,
disable_tqdm=args.disable_tqdm,
save_on_each_node=args.save_on_each_node,
acc_strategy=args.acc_strategy,
save_safetensors=args.save_safetensors,
logging_first_step=True,
**kwargs)
if args.gradient_checkpointing:
model.config.use_cache = False # fix transformers==4.36
logger.info('Setting model.config.use_cache: False')
model.enable_input_require_grads()
if is_dist():
# Compatible with https://github.com/huggingface/transformers/pull/25903
training_args._frozen = False
if args.gradient_checkpointing:
training_args.ddp_find_unused_parameters = False
training_args.ddp_broadcast_buffers = False
else:
training_args.ddp_find_unused_parameters = True
training_args.ddp_broadcast_buffers = True
logger.info(f'training_args: {training_args}')
trainer_kwargs = {}
if args.predict_with_generate:
trainer_kwargs['compute_metrics'] = partial(
compute_nlg_metrics, tokenizer=tokenizer)
else:
compute_metrics = partial(
compute_acc_metrics, acc_strategy=args.acc_strategy)
trainer_kwargs['compute_metrics'] = compute_metrics
trainer_kwargs[
'preprocess_logits_for_metrics'] = preprocess_logits_for_metrics
if args.check_model_is_latest is False:
trainer_kwargs['check_model'] = False
trainer = Seq2SeqTrainer(
model=model,
args=training_args,
data_collator=data_collator,
train_dataset=train_dataset,
eval_dataset=val_dataset,
tokenizer=tokenizer,
callbacks=callbacks,
**trainer_kwargs)
trainer.sft_args = args
if is_master():
for args_obj, fname in zip([args, training_args],
['sft_args.json', 'training_args.json']):
fpath = os.path.join(args.output_dir, fname)
logger.info(
f'The {args_obj.__class__.__name__} will be saved in: {fpath}')
with open(fpath, 'w', encoding='utf-8') as f:
json.dump(
check_json_format(args_obj.__dict__),
f,
ensure_ascii=False,
indent=2)
logging_path = os.path.join(args.output_dir, 'logging.jsonl')
logger.info(f'The logging file will be saved in: {logging_path}')
trainer.train(training_args.resume_from_checkpoint)
last_model_checkpoint = getattr(trainer.state, 'last_model_checkpoint',
None)
logger.info(f'last_model_checkpoint: {last_model_checkpoint}')
logger.info(
f'best_model_checkpoint: {trainer.state.best_model_checkpoint}')
train_time = get_time_info(trainer.state.log_history, len(train_dataset))
# Visualization
if is_master():
images_dir = os.path.join(args.output_dir, 'images')
logger.info(f'images_dir: {images_dir}')
tb_dir = os.path.join(args.output_dir, 'runs')
plot_images(images_dir, tb_dir, ['train/loss'], 0.9)
if args.push_to_hub:
trainer._add_patterns_to_gitignore(['images/'])
trainer.push_to_hub()
return {
'memory': trainer.perf['memory'],
'train_time': train_time,
'last_model_checkpoint': last_model_checkpoint,
'best_model_checkpoint': trainer.state.best_model_checkpoint,
'best_metric': trainer.state.best_metric,
'global_step': trainer.state.global_step,
'log_history': trainer.state.log_history,
'model_info': model_info,
'dataset_info': dataset_info,
}
sftmain = getmain(SftArguments, llm_sft)
- BaseModel Class QWenLMHeadModel
评论