File size: 17,449 Bytes
42da440 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 | import asyncio
import uuid
from openai import AsyncOpenAI
import cohere
from httpx import Timeout
import traceback
import json
def make_id():
return str(uuid.uuid4())
# How it'll work:
# Things that we want to attach to the engine wrapper will execute either before or after a generation
# it's like input or output observers
# they take the input or output as their input respectively; they're functions
# if we want to customize their behavior (i.e., log outputs to a specific directory) then we make them return a callback that's passed in instead
# inputs = all the messages
# outputs = the completion
# oh and correction, they also take a flag for completion mode, since the input will vary based on that
class EngineWrapper:
def __init__(
self,
model,
api_key=None,
base_url=None,
mode="api", # can be one of api, aphrodite, llama.cpp, cohere
input_observers=[],
output_observers=[],
timeout_total=500.0, # Total operation timeout
timeout_read=240.0, # Read timeout between chunks
timeout_api_call=600, # Timeout for individual API calls
**kwargs,
):
self.mode = mode
self.model = model
self.input_observers = input_observers
self.output_observers = output_observers
self.timeout_api_call = timeout_api_call # Store for use in API calls
if mode == "cohere":
self.client = cohere.AsyncClient(api_key=api_key)
elif mode == "api":
self.client = AsyncOpenAI(
timeout=Timeout(
timeout=timeout_total, # Total operation timeout
connect=10.0, # Connection timeout
read=timeout_read, # Read timeout between chunks (increased for streaming)
write=30.0, # Write timeout
pool=10.0, # Pool timeout
),
api_key=api_key,
base_url=base_url,
)
async def submit_completion(
self, prompt, sampling_params, return_completion_only=False
): # Submit request and wait for it to stream back fully
print(prompt)
if "temperature" not in sampling_params:
sampling_params["temperature"] = 1
if "top_p" not in sampling_params:
sampling_params["top_p"] = 1
if "max_tokens" not in sampling_params:
sampling_params["max_tokens"] = 3000
if "stop" not in sampling_params:
sampling_params["stop"] = [
"<|im_end|>"
] # NOTE we hardcode the im end stop token due to the custom dataset generation model always needing it. I really need to get better at configuring my tokenizers...
else:
if "<|im_end|>" not in sampling_params["stop"]:
sampling_params["stop"].append("<|im_end|>")
if "n_predict" not in sampling_params:
sampling_params["n_predict"] = sampling_params["max_tokens"]
use_min_p = False
if "min_p" in sampling_params:
use_min_p = True
for input_observer in self.input_observers:
input_observer(prompt, completion_mode=True)
if self.mode == "api":
timed_out = False
completion = ""
if use_min_p:
stream = await self.client.completions.create(
model=self.model,
prompt=prompt,
temperature=sampling_params["temperature"],
top_p=sampling_params["top_p"],
stop=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
extra_body={"min_p": sampling_params["min_p"]},
stream=True,
timeout=self.timeout_api_call, # Use configurable timeout
)
else:
stream = await self.client.completions.create(
model=self.model,
prompt=prompt,
temperature=sampling_params["temperature"],
top_p=sampling_params["top_p"],
stop=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
stream=True,
timeout=self.timeout_api_call, # Use configurable timeout
)
async for chunk in stream:
try:
completion = completion + chunk.choices[0].text
except Exception as e:
timed_out = True
for output_observer in self.output_observers:
output_observer(
prompt, completion, True
) # input, output, completion_mode (this is the input format for output observers)
if not return_completion_only:
return prompt + completion, timed_out
else:
return completion, timed_out
if self.mode == "cohere":
raise Exception("Cohere not compatible with completion mode!")
async def submit_chat(
self, messages, sampling_params
): # Submit request and wait for it to stream back fully
if "temperature" not in sampling_params:
sampling_params["temperature"] = 1
if "top_p" not in sampling_params:
sampling_params["top_p"] = 1
if "max_tokens" not in sampling_params:
sampling_params["max_tokens"] = 3000
if "stop" not in sampling_params:
sampling_params["stop"] = ["<|im_end|>"]
else:
if "<|im_end|>" not in sampling_params["stop"]:
sampling_params["stop"].append("<|im_end|>")
use_min_p = False
if "min_p" in sampling_params:
use_min_p = True
for input_observer in self.input_observers:
input_observer(messages, False)
if self.mode == "api":
completion = ""
timed_out = False
if use_min_p:
stream = await self.client.chat.completions.create(
model=self.model,
messages=messages,
temperature=sampling_params["temperature"],
top_p=sampling_params["top_p"],
stop=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
extra_body={"min_p": sampling_params["min_p"]},
stream=True,
timeout=self.timeout_api_call, # Use configurable timeout
)
else:
stream = await self.client.chat.completions.create(
model=self.model,
messages=messages,
temperature=sampling_params["temperature"],
top_p=sampling_params["top_p"],
stop=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
stream=True,
timeout=self.timeout_api_call, # Use configurable timeout
)
async for chunk in stream:
try:
# print(chunk.choices)
try:
if chunk.choices[0].delta.content:
completion = completion + chunk.choices[0].delta.content
except Exception as e:
# print("Really strange exception!")
# print(chunk)
# traceback.print_exc()
pass
except Exception as e:
print("\n\n------------CAUGHT EXCEPTION DURING GENERATION")
print(e)
traceback.print_exc()
timed_out = True
print("\n\n-----/\------")
for output_observer in self.output_observers:
output_observer(messages, completion, False)
return completion, timed_out
elif self.mode == "cohere":
timed_out = False
completion = ""
messages_cohereified = [
{
"role": "USER" if message["role"] == "user" else "CHATBOT",
"message": message["content"],
}
for message in messages
]
stream = self.client.chat_stream(
model=self.model,
chat_history=messages_cohereified[1:-1],
message=messages_cohereified[-1]["message"],
preamble=messages_cohereified[0]["message"],
temperature=sampling_params["temperature"],
p=sampling_params["top_p"],
stop_sequences=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
)
async for chunk in stream:
try:
if chunk.event_type == "text-generation":
completion = completion + chunk.text
except Exception as e:
print("THIS RESPONSE TIMED OUT PARTWAY THROUGH GENERATION!")
print(e)
timed_out = True
for output_observer in self.output_observers:
output_observer(messages, completion, False)
return completion, timed_out
else:
raise Exception("Aphrodite not compatible with chat mode!")
async def submit_completion_streaming(
self, prompt, sampling_params, return_completion_only=False
):
"""Submit request and yield chunks as they arrive for streaming"""
print(prompt)
if "temperature" not in sampling_params:
sampling_params["temperature"] = 1
if "top_p" not in sampling_params:
sampling_params["top_p"] = 1
if "max_tokens" not in sampling_params:
sampling_params["max_tokens"] = 3000
if "stop" not in sampling_params:
sampling_params["stop"] = ["<|im_end|>"]
else:
if "<|im_end|>" not in sampling_params["stop"]:
sampling_params["stop"].append("<|im_end|>")
if "n_predict" not in sampling_params:
sampling_params["n_predict"] = sampling_params["max_tokens"]
use_min_p = False
if "min_p" in sampling_params:
use_min_p = True
for input_observer in self.input_observers:
input_observer(prompt, completion_mode=True)
if self.mode == "api":
timed_out = False
completion = ""
if use_min_p:
stream = await self.client.completions.create(
model=self.model,
prompt=prompt,
temperature=sampling_params["temperature"],
top_p=sampling_params["top_p"],
stop=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
extra_body={"min_p": sampling_params["min_p"]},
stream=True,
timeout=self.timeout_api_call,
)
else:
stream = await self.client.completions.create(
model=self.model,
prompt=prompt,
temperature=sampling_params["temperature"],
top_p=sampling_params["top_p"],
stop=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
stream=True,
timeout=self.timeout_api_call,
)
async for chunk in stream:
try:
text_chunk = chunk.choices[0].text
completion += text_chunk
# Yield chunk in SSE format
yield f"data: {json.dumps({'text': text_chunk, 'done': False})}\n\n"
except Exception as e:
timed_out = True
yield f"data: {json.dumps({'error': str(e), 'done': True})}\n\n"
break
# Send completion signal
if not timed_out:
yield f"data: {json.dumps({'text': '', 'done': True})}\n\n"
for output_observer in self.output_observers:
output_observer(prompt, completion, True)
if self.mode == "cohere":
raise Exception("Cohere not compatible with completion mode!")
async def submit_chat_streaming(self, messages, sampling_params):
"""Submit chat request and yield chunks as they arrive for streaming"""
if "temperature" not in sampling_params:
sampling_params["temperature"] = 1
if "top_p" not in sampling_params:
sampling_params["top_p"] = 1
if "max_tokens" not in sampling_params:
sampling_params["max_tokens"] = 3000
if "stop" not in sampling_params:
sampling_params["stop"] = ["<|im_end|>"]
else:
if "<|im_end|>" not in sampling_params["stop"]:
sampling_params["stop"].append("<|im_end|>")
use_min_p = False
if "min_p" in sampling_params:
use_min_p = True
for input_observer in self.input_observers:
input_observer(messages, False)
if self.mode == "api":
completion = ""
timed_out = False
if use_min_p:
stream = await self.client.chat.completions.create(
model=self.model,
messages=messages,
temperature=sampling_params["temperature"],
top_p=sampling_params["top_p"],
stop=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
extra_body={"min_p": sampling_params["min_p"]},
stream=True,
timeout=self.timeout_api_call,
)
else:
stream = await self.client.chat.completions.create(
model=self.model,
messages=messages,
temperature=sampling_params["temperature"],
top_p=sampling_params["top_p"],
stop=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
stream=True,
timeout=self.timeout_api_call,
)
async for chunk in stream:
try:
try:
if chunk.choices[0].delta.content:
text_chunk = chunk.choices[0].delta.content
completion += text_chunk
# Yield chunk in SSE format
yield f"data: {json.dumps({'text': text_chunk, 'done': False})}\n\n"
except Exception as e:
pass
except Exception as e:
print("\n\n------------CAUGHT EXCEPTION DURING GENERATION")
print(e)
traceback.print_exc()
timed_out = True
yield f"data: {json.dumps({'error': str(e), 'done': True})}\n\n"
break
# Send completion signal
if not timed_out:
yield f"data: {json.dumps({'text': '', 'done': True})}\n\n"
for output_observer in self.output_observers:
output_observer(messages, completion, False)
elif self.mode == "cohere":
timed_out = False
completion = ""
messages_cohereified = [
{
"role": "USER" if message["role"] == "user" else "CHATBOT",
"message": message["content"],
}
for message in messages
]
stream = self.client.chat_stream(
model=self.model,
chat_history=messages_cohereified[1:-1],
message=messages_cohereified[-1]["message"],
preamble=messages_cohereified[0]["message"],
temperature=sampling_params["temperature"],
p=sampling_params["top_p"],
stop_sequences=sampling_params["stop"],
max_tokens=sampling_params["max_tokens"],
)
async for chunk in stream:
try:
if chunk.event_type == "text-generation":
text_chunk = chunk.text
completion += text_chunk
yield f"data: {json.dumps({'text': text_chunk, 'done': False})}\n\n"
except Exception as e:
print("THIS RESPONSE TIMED OUT PARTWAY THROUGH GENERATION!")
print(e)
timed_out = True
yield f"data: {json.dumps({'error': str(e), 'done': True})}\n\n"
break
if not timed_out:
yield f"data: {json.dumps({'text': '', 'done': True})}\n\n"
for output_observer in self.output_observers:
output_observer(messages, completion, False)
else:
raise Exception("Aphrodite not compatible with chat mode!")
|