File size: 2,963 Bytes
dced7e4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 |
---
license: apache-2.0
---
Have to use with basemodel "princeton-nlp/Llama-3-Instruct-8B-SimPO".
Here's a example Demo code with Gradio:
```
import gradio as gr
from llamafactory.chat import ChatModel
from llamafactory.extras.misc import torch_gc
import re
def split_into_sentences(text):
sentence_endings = re.compile(r'(?<!\w\.\w.)(?<![A-Z][a-z]\.)(?<=\.|\?|\!)\s')
sentences = sentence_endings.split(text)
return [sentence.strip() for sentence in sentences if sentence]
def process_paragraph(paragraph, progress=gr.Progress()):
sentences = split_into_sentences(paragraph)
results = []
total_sentences = len(sentences)
for i, sentence in enumerate(sentences):
progress((i + 1) / total_sentences)
messages.append({"role": "user", "content": sentence})
sentence_response = ""
for new_text in chat_model.stream_chat(messages, temperature=0.7, top_p=0.9, top_k=50, max_new_tokens=300):
sentence_response += new_text.strip()
category = sentence_response.strip().lower().replace(' ', '_')
if category != "fair":
results.append((sentence, category))
else:
results.append((sentence, "fair"))
messages.append({"role": "assistant", "content": sentence_response})
torch_gc()
return results
args = dict(
model_name_or_path="princeton-nlp/Llama-3-Instruct-8B-SimPO", # 使用量化的 Llama-3-8B-Instruct 模型
adapter_name_or_path="StevenChen16/llama3-8b-compliance-review-adapter", # 加载保存的 LoRA 适配器
template="llama3", # 与训练时使用的模板相同
finetuning_type="lora", # 与训练时使用的微调类型相同
quantization_bit=8, # 加载 4-bit 量化模型
use_unsloth=True, # 使用 UnslothAI 的 LoRA 优化以加速生成
)
chat_model = ChatModel(args)
messages = []
# 定义类型到颜色的映射
label_to_color = {
"fair": "green",
"limitation_of_liability": "red",
"unilateral_termination": "orange",
"unilateral_change": "yellow",
"content_removal": "purple",
"contract_by_using": "blue",
"choice_of_law": "cyan",
"jurisdiction": "magenta",
"arbitration": "brown",
}
with gr.Blocks() as demo:
with gr.Row(equal_height=True):
with gr.Column():
input_text = gr.Textbox(label="Input Paragraph", lines=10, placeholder="Enter the paragraph here...")
btn = gr.Button("Process")
with gr.Column():
output = gr.HighlightedText(label="Processed Paragraph", color_map=label_to_color)
progress = gr.Progress()
def on_click(paragraph):
results = process_paragraph(paragraph, progress=progress)
return results
btn.click(on_click, inputs=input_text, outputs=[output])
demo.launch(share=True)
``` |