File size: 13,043 Bytes
974817f |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 |
from contextlib import contextmanager
from codetiming import Timer
@contextmanager
def _timer(name: str, timing_raw):
with Timer(name=name, logger=None) as timer:
yield
timing_raw[name] = timer.last
from buffer import SurveyManager
from buffer import BufferManager_V2 as BufferManager
from vllm import LLM, SamplingParams
from transformers import AutoTokenizer
import re
from fastapi import FastAPI, WebSocket, WebSocketDisconnect
from fastapi.middleware.cors import CORSMiddleware
import asyncio
import argparse
from pydantic import BaseModel
import json
import aiohttp
app = FastAPI()
# 允许跨域(如果前端和后端端口不同需要加上)
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
active_connections = set()
@app.websocket("/ws")
async def websocket_endpoint(websocket: WebSocket):
await websocket.accept()
active_connections.add(websocket)
try:
while True:
await websocket.receive_text()
except WebSocketDisconnect:
active_connections.remove(websocket)
async def post_to_frontend(payload):
print(f"Sending payload to frontend: {payload}") # Log the payload being sent
for ws in list(active_connections):
try:
await ws.send_text(payload)
except Exception as e:
print(f"Error sending to WebSocket: {e}")
active_connections.remove(ws)
def write_to_json(data, path):
with open(path, 'w', encoding='utf8') as f:
f.write(json.dumps(data, ensure_ascii=False, indent=4))
class OriginalvLLMRollout:
def __init__(self, model_name_or_path):
# init vLLM
self.rollout_model = LLM(
model=model_name_or_path,
tokenizer=model_name_or_path,
gpu_memory_utilization=0.95,
trust_remote_code=True,
)
self.sampling_params = SamplingParams(
temperature=0.7,
top_p=0.8,
repetition_penalty=1.05,
top_k=20,
max_tokens=2748,
)
def generate(self, input_texts):
generated_texts = []
completions = self.rollout_model.generate(input_texts, self.sampling_params, use_tqdm=False)
for output in completions:
generated_text = output.outputs[0].text
generated_texts.append(generated_text)
return generated_texts
def chat(self, input_messages):
generated_texts = []
completions = self.rollout_model.chat(input_messages, self.sampling_params, use_tqdm=False)
for output in completions:
generated_text = output.outputs[0].text
generated_texts.append(generated_text)
return generated_texts
async def rollout_with_env(querys, batch_size, max_turns, model_path, url,
deploy_port=None):
"""
Args:
querys: [string]
"""
###############################
#### splited by batch size ####
###############################
n = len(querys) // batch_size
batch_querys = []
for i in range(n+1):
temp_data = querys[i*batch_size: (i+1)*batch_size]
if len(temp_data) > 0:
batch_querys.append(temp_data)
print("QUERY NUMBER with BATCH: ", [len(x) for x in batch_querys])
###################
#### init vllm ####
###################
vllm_manager = OriginalvLLMRollout(model_path)
############################
#### init Format Reward ####
############################
tokenizer = AutoTokenizer.from_pretrained(model_path)
total_rollout_data = []
for querys in batch_querys:
###########################################
#### acquire env configs and init envs ####
###########################################
buffer_manager = BufferManager(querys)
while True:
# Break at max-turns
if buffer_manager.step >= max_turns:
break
###############################
#### prepare input prompts ####
###############################
messagess_todo = buffer_manager.build_prompt_for_generator()
# breakpoint()
# Break when no tasks
if len(messagess_todo) == 0:
break
##########################
#### generate by vLLM ####
##########################
timing_raw = {}
with _timer('vllm sampling', timing_raw):
# response_texts = vllm_manager.chat(messagess_todo)
response_texts = await asyncio.to_thread(vllm_manager.chat, messagess_todo)
##################################
#### preprocess the responses ####
##################################
# 对response的详细处理可以集成到环境类中,因环境而异, 先对Response进行预处理
extracted_results = []
for response_text in response_texts:
result = BufferManager.parse_generator_response(response_text)
extracted_results.append(result)
#################################################
#### execute in environment and get feedback ####
#################################################
payload = {
"tool_calls": [x["tool_call"] for x in extracted_results]
}
if buffer_manager.step <=2:
payload["topk"] = 20
with _timer('get env feedback', timing_raw):
# env_response_batched = requests.post(url, json=payload).json()
async with aiohttp.ClientSession() as session:
async with session.post(url, json=payload) as resp:
env_response_batched = await resp.json()
###################################
#### postprocess the feedbacks ####
###################################
with _timer('postprocessing', timing_raw):
buffer_manager.update_trajectory(extracted_results, env_response_batched)
buffer_manager.step += 1
print(timing_raw)
if deploy_port is not None:
now_text = json_to_markdown(buffer_manager.batch_rollout_data[-1])
now_search_keywords= buffer_manager.batch_rollout_data[-1]["trajectory"][-1]["search_keywords"]
now_update = buffer_manager.batch_rollout_data[-1]["trajectory"][-1]["answer_thought"]
next_update = buffer_manager.batch_rollout_data[-1]["trajectory"][-1]["tool_call_thought"]
now_query = buffer_manager.batch_rollout_data[-1]["query"]
trajs = buffer_manager.batch_rollout_data[-1]["trajectory"]
updated_success = buffer_manager.batch_rollout_data[-1]["trajectory"][-1]["update_success"]
if updated_success:
for traj in reversed(trajs):
if len(traj["summarys"]) > 0:
break
summary_num = len(traj["summarys"])
if summary_num == 0:
summary_text = "No summaries yet."
else:
summary_text = "\n".join(traj["summarys"])
frontend_payload = {
"markdown": now_text,
"searchKeywords": now_search_keywords,
"nowUpdate": now_update,
"nextUpdate": next_update,
"query": now_query,
"papers": summary_text
}
frontend_payload = json.dumps(frontend_payload, ensure_ascii=False)
try:
await post_to_frontend(frontend_payload)
except Exception as e:
print(f"Error posting to frontend: {e}")
for item in buffer_manager.batch_rollout_data:
item["survey_text"] = SurveyManager.convert_survey_dict_to_str(item["state"]["current_survey"])
total_rollout_data.extend(buffer_manager.batch_rollout_data)
#####################################
#### clear all envs and shutdown ####
#####################################
del buffer_manager
return total_rollout_data
def json_to_markdown(json_data):
text = SurveyManager.convert_survey_dict_to_str(json_data["state"]["current_survey"])
all_summarys = {}
for traj in json_data["trajectory"]:
for item in traj["summarys"]:
split_text = item.split("\n")
bibkey = split_text[0].split(":")[1].strip()
title_begin_index = item.find("Title:") + len("Title:")
title_end_index = item.find("Abstract:")
title = item[title_begin_index:title_end_index].strip()
arxivid = bibkey.split("arxivid")[-1].strip()
html = f"arxiv.org/abs/{arxivid}"
all_summarys[bibkey] = f"[{title}](https://{html})"
reg = r"\\cite\{(.+?)\}"
placeholder_reg = re.compile(r"^#\d+$")
reg_bibkeys = re.findall(reg, text)
bibkeys = []
for bibkey in reg_bibkeys:
single_bib = bibkey.split(",")
for bib in single_bib:
if not placeholder_reg.match(bib):
bib = bib.strip()
if bib and bib != "*" and bib not in bibkeys:
bibkeys.append(bib)
bibkeys_index = {bibkey: i+1 for i, bibkey in enumerate(bibkeys)}
def replace_bibkey(bibkey):
bibkey = bibkey.group(1)
single_bib = bibkey.split(",")
new_bibs = []
for bib in single_bib:
if not placeholder_reg.match(bib):
bib = bib.strip()
if bib and bib != "*":
if bib in bibkeys_index:
new_bibs.append(f"{bibkeys_index[bib]}")
else:
print(f"Warning: {bib} not found in bibkeys")
if len(new_bibs) > 0:
return "[" + ",".join(new_bibs) + "]"
else:
return ""
text = re.sub(reg, replace_bibkey, text)
reference_text = "\n\n".join([f"[{i}] {all_summarys[bibkey]}" for bibkey, i in bibkeys_index.items()])
text += "\n## References\n" + reference_text
return text
async def test_surveyGen(model_path, out_path,querys, url, deploy_port=None):
total_rollout_data = await rollout_with_env(querys, 1, 1000, model_path, url, deploy_port)
all_md_texts = []
for json_data in total_rollout_data:
md_text = json_to_markdown(json_data)
all_md_texts.append(md_text)
all_md_texts = "\n\n".join(all_md_texts)
with open(out_path, 'w', encoding='utf8') as f:
f.write(all_md_texts)
# with jsonlines.open(out_path, 'w') as writer:
# for item in total_rollout_data:
# writer.write(item)
class QueryRequest(BaseModel):
query: str
@app.post("/generate_survey")
async def generate_survey(request: QueryRequest):
global args # Ensure args is accessible
# 这里可以根据需要处理查询
model_path = args.model_path
out_path = args.output_file
query = request.query
querys = [query] # 将查询转换为列表
url = args.retriver_url
deploy_port = args.port if args.port is not None else None
try:
await test_surveyGen(model_path, out_path, querys, url, deploy_port)
return {"status": "success", "message": "Survey generated successfully."}
except Exception as e:
print(f"Error generating survey: {e}")
return {"status": "error", "message": str(e)}
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Run survey generation with vLLM.")
parser.add_argument("--model_path", type=str, required=True, help="Path to the model.")
parser.add_argument("--query", type=str, required=True, help="Query to generate survey.")
parser.add_argument("--output_file", type=str, required=True, help="Path to the output Markdown file.")
parser.add_argument("--retriver_url", type=str, default="http://localhost:8400", help="URL of the retriever service.")
parser.add_argument("--port", type=str, default=None, help="Deploy port, default is None, which means not deploy.")
args = parser.parse_args()
if args.port is not None:
import uvicorn
uvicorn.run(app, host="localhost", port=int(args.port))# log_level="debug")
# Run the survey generation
else:
asyncio.run(
test_surveyGen(
model_path=args.model_path,
out_path=args.output_file,
querys=[args.query],
url=args.retriver_url
)
)
|