情感分类模型作业

Traiig procedure

Framework versios

• SWIFT 1.5.3

Base model iformatio

• BaseModel Class QWeLMHeadModel

情感分类，输⼊⼀段⽂本, 模型可以给出对应的情感分类

环境准备

git cloe https://github.com/modelscope/swift.git
cd swift
pip istall -e .[llm]

代码

import os
os.eviro['CUDA_VISIBLE_DEVICES'] = '0'
import torch
from swift.llm import (
DatasetName, IferArgumets, ModelType, SftArgumets,
ifer_mai, sft_mai, app_ui_mai, merge_lora_mai
)
model_type = ModelType.qwe_1_8b

sft_args = SftArgumets(
model_type=model_type,
sft_type='lora',
trai_dataset_sample=2000,
dataset=[DatasetName.jd_setimet_zh],

output_dir='output')
result = sft_mai(sft_args)
# adapter的⼤⼩: 6.1MiB
best_model_checkpoit = result['best_model_checkpoit']
prit(f'best_model_checkpoit: {best_model_checkpoit}')
torch.cuda.empty_cache()
ifer_args = IferArgumets(
ckpt_dir=best_model_checkpoit,
load_dataset_cofig=True,
do_sample=False)
result = ifer_mai(ifer_args)

结果展示

[PROMPT]Task: Setimet Classificatio
Setece: 店家太不负责任了，衣服质量太差劲了，和图片上的不一样
Category: egative, positive
Output: [OUTPUT]egative<|edoftext|>

[LABELS]egative
--------------------------------------------------
[PROMPT]Task: Setimet Classificatio
Setece: 送国际友人挺好的，不错不错！
Category: egative, positive
Output: [OUTPUT]positive<|edoftext|>

[LABELS]positive
--------------------------------------------------
[PROMPT]Task: Setimet Classificatio
Setece: 很好,装好一定很漂亮
Category: egative, positive
Output: [OUTPUT]positive<|edoftext|>

[LABELS]positive
--------------------------------------------------
[PROMPT]Task: Setimet Classificatio
Setece: 东西给你退回去了，你要黑我钱！！！
Category: egative, positive
Output: [OUTPUT]egative<|edoftext|>

[LABELS]egative
--------------------------------------------------
[PROMPT]Task: Setimet Classificatio
Setece: 送货很快，书是正品，买书一直京东是首选！
Category: egative, positive
Output: [OUTPUT]positive<|edoftext|>

[LABELS]positive
--------------------------------------------------
[PROMPT]Task: Setimet Classificatio
Setece: 口感相当的好 都想买第二次了
Category: egative, positive
Output: [OUTPUT]positive<|edoftext|>

[LABELS]positive
--------------------------------------------------
[PROMPT]Task: Setimet Classificatio
Setece: 硅胶味道太重，样子与图片差距太大
Category: egative, positive
Output: [OUTPUT]egative<|edoftext|>

[LABELS]egative
--------------------------------------------------
[PROMPT]Task: Setimet Classificatio
Setece: 很伤心，买了放到三星4尽然不能用，客服各种推
Category: egative, positive
Output: [OUTPUT]egative<|edoftext|>

[LABELS]egative
--------------------------------------------------
[PROMPT]Task: Setimet Classificatio
Setece: 质量不错，大小合适，应当是正品！但是我买的是黑灰，发来的却是纯黑，懒得换了，给个差评，希望以后改进！
Category: egative, positive
Output: [OUTPUT]positive<|edoftext|>
[INFO:swift] save_result_path: /mt/workspace/output/qwe-1_8b/v4-20240128-154925/checkpoit-100/ifer_result_20240128-155452.jsol

[LABELS]egative

Training procedure Framework versions SWIFT 1.5.3 Base model information BaseModel Class QWenLMHe