File size: 10,660 Bytes
45cf443 92c7321 45cf443 c6bc767 45cf443 2c87b68 45cf443 c6bc767 45cf443 | 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 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 | import argparse
import json
import re
import os
import time
import torch
import warnings
import uvicorn
from threading import Thread
from queue import Queue
from fastapi import FastAPI, HTTPException
from fastapi.responses import StreamingResponse
from pydantic import BaseModel, Field
from transformers import AutoTokenizer, AutoModelForCausalLM, TextStreamer
from models import LMConfig, LMForCausalLM
from models.lm.lora import apply_lora, load_lora
warnings.filterwarnings('ignore')
app = FastAPI()
def init_model(args):
tokenizer = AutoTokenizer.from_pretrained(args.load_from)
if 'model' in args.load_from:
moe_suffix = '_moe' if args.use_moe else ''
ckp = f'../{args.save_dir}/{args.weight}_{args.hidden_size}{moe_suffix}.pth'
model = LMForCausalLM(LMConfig(
hidden_size=args.hidden_size,
num_hidden_layers=args.num_hidden_layers,
max_seq_len=args.max_seq_len,
use_moe=bool(args.use_moe),
inference_rope_scaling=args.inference_rope_scaling
))
model.load_state_dict(torch.load(ckp, map_location=device), strict=True)
if args.lora_weight != 'None':
apply_lora(model)
load_lora(model, f'../{args.save_dir}/lora/{args.lora_weight}_{args.hidden_size}.pth')
else:
model = AutoModelForCausalLM.from_pretrained(args.load_from, trust_remote_code=True)
print(f'MiniMind模型参数量: {sum(p.numel() for p in model.parameters()) / 1e6:.2f} M(illion)')
return model.half().eval().to(device), tokenizer
class ChatRequest(BaseModel):
model: str
messages: list
temperature: float = 0.7
top_p: float = 0.92
max_tokens: int = 8192
stream: bool = True
tools: list = Field(default_factory=list)
open_thinking: bool = False
chat_template_kwargs: dict = None
def get_open_thinking(self) -> bool:
"""兼容多种方式开启 thinking"""
if self.open_thinking:
return True
if self.chat_template_kwargs:
return self.chat_template_kwargs.get('open_thinking', False) or \
self.chat_template_kwargs.get('enable_thinking', False)
return False
class CustomStreamer(TextStreamer):
def __init__(self, tokenizer, queue):
super().__init__(tokenizer, skip_prompt=True, skip_special_tokens=True)
self.queue = queue
self.tokenizer = tokenizer
def on_finalized_text(self, text: str, stream_end: bool = False):
self.queue.put(text)
if stream_end:
self.queue.put(None)
def parse_response(text):
reasoning_content = None
think_match = re.search(r'<think>(.*?)</think>', text, re.DOTALL)
if think_match:
reasoning_content = think_match.group(1).strip()
text = re.sub(r'<think>.*?</think>\s*', '', text, flags=re.DOTALL)
elif '</think>' in text:
parts = text.split('</think>', 1)
reasoning_content = parts[0].strip()
text = parts[1].strip() if len(parts) > 1 else ''
tool_calls = []
for i, m in enumerate(re.findall(r'<tool_call>(.*?)</tool_call>', text, re.DOTALL)):
try:
call = json.loads(m.strip())
tool_calls.append({"id": f"call_{int(time.time())}_{i}", "type": "function", "function": {"name": call.get("name", ""), "arguments": json.dumps(call.get("arguments", {}), ensure_ascii=False)}})
except Exception:
pass
if tool_calls:
text = re.sub(r'<tool_call>.*?</tool_call>', '', text, flags=re.DOTALL)
return text.strip(), reasoning_content, tool_calls or None
def generate_stream_response(messages, temperature, top_p, max_tokens, tools=None, open_thinking=False):
try:
new_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, tools=tools or None, open_thinking=open_thinking)
inputs = tokenizer(new_prompt, return_tensors="pt", truncation=True).to(device)
queue = Queue()
streamer = CustomStreamer(tokenizer, queue)
def _generate():
try:
model.generate(
inputs.input_ids,
max_new_tokens=max_tokens,
do_sample=True,
temperature=temperature,
top_p=top_p,
attention_mask=inputs.attention_mask,
pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id,
streamer=streamer
)
except Exception as e:
queue.put({"error": str(e)})
queue.put(None)
Thread(target=_generate).start()
full_text = ""
emitted = 0
thinking_ended = not bool(open_thinking)
while True:
text = queue.get()
if text is None:
break
if isinstance(text, dict):
yield json.dumps(text, ensure_ascii=False)
continue
full_text += text
if not thinking_ended:
pos = full_text.find('</think>')
if pos >= 0:
thinking_ended = True
new_r = full_text[emitted:pos]
if new_r:
yield json.dumps({"choices": [{"delta": {"reasoning_content": new_r}}]}, ensure_ascii=False)
emitted = pos + len('</think>')
after = full_text[emitted:].lstrip('\n')
emitted = len(full_text) - len(after)
if after:
yield json.dumps({"choices": [{"delta": {"content": after}}]}, ensure_ascii=False)
emitted = len(full_text)
else:
new_r = full_text[emitted:]
if new_r:
yield json.dumps({"choices": [{"delta": {"reasoning_content": new_r}}]}, ensure_ascii=False)
emitted = len(full_text)
else:
new_c = full_text[emitted:]
if new_c:
yield json.dumps({"choices": [{"delta": {"content": new_c}}]}, ensure_ascii=False)
emitted = len(full_text)
_, _, tool_calls = parse_response(full_text)
if tool_calls:
yield json.dumps({"choices": [{"delta": {"tool_calls": tool_calls}}]}, ensure_ascii=False)
yield json.dumps({"choices": [{"delta": {}, "finish_reason": "tool_calls" if tool_calls else "stop"}]}, ensure_ascii=False)
except Exception as e:
yield json.dumps({"error": str(e)})
@app.post("/v1/chat/completions")
async def chat_completions(request: ChatRequest):
try:
if request.stream:
return StreamingResponse(
(f"data: {chunk}\n\n" for chunk in generate_stream_response(
messages=request.messages,
temperature=request.temperature,
top_p=request.top_p,
max_tokens=request.max_tokens,
tools=request.tools,
open_thinking=request.get_open_thinking()
)),
media_type="text/event-stream"
)
else:
new_prompt = tokenizer.apply_chat_template(
request.messages,
tokenize=False,
add_generation_prompt=True,
tools=request.tools or None,
open_thinking=request.get_open_thinking()
)
inputs = tokenizer(new_prompt, return_tensors="pt", truncation=True).to(device)
with torch.no_grad():
generated_ids = model.generate(
inputs["input_ids"],
max_length=inputs["input_ids"].shape[1] + request.max_tokens,
do_sample=True,
attention_mask=inputs["attention_mask"],
pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id,
top_p=request.top_p,
temperature=request.temperature
)
answer = tokenizer.decode(generated_ids[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
content, reasoning_content, tool_calls = parse_response(answer)
message = {"role": "assistant", "content": content}
if reasoning_content:
message["reasoning_content"] = reasoning_content
if tool_calls:
message["tool_calls"] = tool_calls
return {
"id": f"chatcmpl-{int(time.time())}",
"object": "chat.completion",
"created": int(time.time()),
"model": "omni",
"choices": [
{
"index": 0,
"message": message,
"finish_reason": "tool_calls" if tool_calls else "stop"
}
]
}
except Exception as e:
raise HTTPException(status_code=500, detail=str(e))
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Server for MiniMind")
parser.add_argument('--load_from', default='../model', type=str, help="模型加载路径（model=原生torch权重，其他路径=transformers格式）")
parser.add_argument('--save_dir', default='checkpoint', type=str, help="模型权重目录")
parser.add_argument('--weight', default='full_sft', type=str, help="权重名称前缀（pretrain, full_sft, dpo, reason, ppo_actor, grpo, spo）")
parser.add_argument('--lora_weight', default='None', type=str, help="LoRA权重名称（None表示不使用，可选：lora_identity, lora_medical）")
parser.add_argument('--hidden_size', default=768, type=int, help="隐藏层维度")
parser.add_argument('--num_hidden_layers', default=8, type=int, help="隐藏层数量")
parser.add_argument('--max_seq_len', default=8192, type=int, help="最大序列长度")
parser.add_argument('--use_moe', default=0, type=int, choices=[0, 1], help="是否使用MoE架构（0=否，1=是）")
parser.add_argument('--inference_rope_scaling', default=False, action='store_true', help="启用RoPE位置编码外推（4倍，仅解决位置编码问题）")
parser.add_argument('--device', default='cuda' if torch.cuda.is_available() else 'cpu', type=str, help="运行设备")
args = parser.parse_args()
device = args.device
model, tokenizer = init_model(args)
uvicorn.run(app, host="0.0.0.0", port=8998)
|