File size: 11,684 Bytes
3d3d712 |
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 |
import hashlib
import json
import os
import random
import time
from dataclasses import dataclass
from typing import Any, Dict, Generator, List, Literal, Optional
import yaml
from injector import inject
from taskweaver.llm.util import ChatMessageRoleType, ChatMessageType, format_chat_message
from .base import CompletionService, EmbeddingService, LLMServiceConfig
MockServiceModeType = Literal[
"fixed",
"record_only",
"playback_only",
"playback_or_record",
]
class MockApiServiceConfig(LLMServiceConfig):
def _configure(self) -> None:
self._set_name("mock")
mock_mode = self._get_enum(
"mode",
options=["fixed", "record_only", "playback_only", "playback_or_record"],
default="playback_or_record",
)
assert mock_mode in [
"fixed",
"record_only",
"playback_only",
"playback_or_record",
]
self.mode: MockServiceModeType = mock_mode # type: ignore
self.fixed_chat_responses: str = self._get_str(
"fixed_chat_responses",
json.dumps(format_chat_message("assistant", "Hello!")),
)
self.fixed_embedding_responses: str = self._get_str(
"fixed_embedding_responses",
json.dumps([[0.0] * 64]),
)
self.cache_path: str = self._get_path(
"cache_path",
os.path.join(self.src.app_base_path, "cache", "mock.yaml"),
)
# split the chat completion response into chunks and delay each chunk by this amount
# if negative, return the whole response at once
self.playback_delay: float = self._get_float(
"playback_delay",
0.05,
)
@dataclass
class MockCacheEntry:
value: str
query: str
created_at: float
last_accessed_at: float
class MockCacheStore:
def __init__(self, path: str):
self.path = path
self.completion_store: Dict[str, MockCacheEntry] = {}
self.embedding_store: Dict[str, MockCacheEntry] = {}
if os.path.exists(self.path):
self._init_from_disk()
def get_completion(self, query: List[ChatMessageType]) -> Optional[ChatMessageType]:
serialized_query = self._serialize_completion_query(query)
serialized_value = self._get_from_store(self.completion_store, serialized_query)
if serialized_value is None:
return None
return self._deserialize_completion_response(serialized_value)
def get_embedding(self, query: str) -> Optional[List[float]]:
serialized_query = self._serialize_embedding_query(query)
serialized_value = self._get_from_store(self.embedding_store, serialized_query)
if serialized_value is None:
return None
return self._deserialize_embedding_response(serialized_value)
def _get_from_store(
self,
store: Dict[str, MockCacheEntry],
query: str,
) -> Optional[str]:
key = self._query_to_key(query)
if key in store:
entry = store[key]
entry.last_accessed_at = time.time()
return entry.value
return None
def set_completion(
self,
query: List[ChatMessageType],
value: ChatMessageType,
) -> None:
serialized_query = self._serialize_completion_query(query)
serialized_value = self._serialize_completion_response(value)
self._set_to_store(self.completion_store, serialized_query, serialized_value)
def set_embedding(self, query: str, value: List[float]) -> None:
serialized_query = self._serialize_embedding_query(query)
serialized_value = self._serialize_embedding_response(value)
self._set_to_store(self.embedding_store, serialized_query, serialized_value)
def _set_to_store(
self,
store: Dict[str, MockCacheEntry],
query: str,
value: str,
) -> None:
key = self._query_to_key(query)
store[key] = MockCacheEntry(
value=value,
query=query,
created_at=time.time(),
last_accessed_at=time.time(),
)
self._save_to_disk()
def _serialize_completion_query(self, query: List[ChatMessageType]) -> str:
return "\n".join([self._serialize_completion_response(x) for x in query])
def _serialize_completion_response(self, response: ChatMessageType) -> str:
return f"{response['role']}:{response['content']}"
def _deserialize_completion_response(self, response: str) -> ChatMessageType:
segment = response.split(":", 1)
role = segment[0] if len(segment) > 0 else "assistant"
if role not in ["assistant", "user", "system"]:
raise ValueError(f"Invalid role {role}")
content = segment[1] if len(segment) > 1 else ""
return format_chat_message(role, content) # type: ignore
def _serialize_embedding_query(self, query: str) -> str:
return query
def _serialize_embedding_response(self, response: List[float]) -> str:
return ",".join([str(x) for x in response])
def _deserialize_embedding_response(self, response: str) -> List[float]:
return [float(x) for x in response.split(",")]
def _query_to_key(self, query: str) -> str:
return hashlib.md5(query.encode("utf-8")).hexdigest()
def _init_from_disk(self):
try:
cache = yaml.safe_load(open(self.path, "r"))
except Exception as e:
print(f"Error loading cache file {self.path}: {e}")
return
try:
completion_store = cache["completion_store"]
for key, value in completion_store.items():
try:
self.completion_store[key] = MockCacheEntry(**value)
except Exception as e:
print(f"Error loading cache entry {key}: {e}")
except Exception as e:
print(f"Error loading completion store: {e}")
try:
embedding_store = cache["embedding_store"]
for key, value in embedding_store.items():
try:
self.embedding_store[key] = MockCacheEntry(**value)
except Exception as e:
print(f"Error loading cache entry {key}: {e}")
except Exception as e:
print(f"Error loading embedding store: {e}")
def _save_to_disk(self):
# TODO: postpone immediate update and periodically save to disk
try:
yaml.safe_dump(
{
"completion_store": {k: v.__dict__ for k, v in self.completion_store.items()},
"embedding_store": {k: v.__dict__ for k, v in self.embedding_store.items()},
},
open(self.path, "w"),
)
except Exception as e:
print(f"Error saving cache file {self.path}: {e}")
class MockApiService(CompletionService, EmbeddingService):
@inject
def __init__(self, config: MockApiServiceConfig):
self.config = config
self.base_completion_service: Optional[CompletionService] = None
self.base_embedding_service: Optional[EmbeddingService] = None
self.cache = MockCacheStore(self.config.cache_path)
def set_base_completion_service(
self,
base_completion_service: Optional[CompletionService],
):
self.base_completion_service = base_completion_service
def set_base_embedding_service(
self,
base_embedding_service: Optional[EmbeddingService],
):
self.base_embedding_service = base_embedding_service
def chat_completion(
self,
messages: List[ChatMessageType],
use_backup_engine: bool = False,
stream: bool = True,
temperature: Optional[float] = None,
max_tokens: Optional[int] = None,
top_p: Optional[float] = None,
stop: Optional[List[str]] = None,
**kwargs: Any,
) -> Generator[ChatMessageType, None, None]:
if self.config.mode == "fixed":
return self._get_from_fixed_completion()
cached_value = self.cache.get_completion(messages)
# playback
if cached_value is None:
if self.config.mode == "playback_only":
raise Exception("No cached value found")
else:
if self.config.mode != "record_only":
return self._get_from_playback_completion(cached_value)
# record
def get_from_base() -> Generator[ChatMessageType, None, None]:
if self.base_completion_service is None:
raise Exception("base_completion_service is not set")
new_value = format_chat_message("assistant", "")
for chunk in self.base_completion_service.chat_completion(
messages,
use_backup_engine,
stream,
temperature,
max_tokens,
top_p,
stop,
**kwargs,
):
new_value["role"] = chunk["role"]
new_value["content"] += chunk["content"]
yield chunk
self.cache.set_completion(messages, new_value)
return get_from_base()
def get_embeddings(self, strings: List[str]) -> List[List[float]]:
if self.config.mode == "fixed":
return [self._get_from_fixed_embedding() for _ in strings]
cached_values = [self.cache.get_embedding(x) for x in strings]
cache_missed_values = [strings[i] for i, v in enumerate(cached_values) if v is None]
if len(cache_missed_values) > 0:
if self.config.mode == "playback_only":
raise Exception("Not all cached values found")
if self.base_embedding_service is None:
raise Exception("base_embedding_service is not set")
new_values = self.base_embedding_service.get_embeddings(cache_missed_values)
cache_missed_values_index = 0
result_values: List[List[float]] = []
for i, v in enumerate(cached_values):
if v is None:
self.cache.set_embedding(
strings[i],
new_values[cache_missed_values_index],
)
result_values.append(new_values[cache_missed_values_index])
cache_missed_values_index += 1
else:
result_values.append(v)
return result_values
def _get_from_fixed_completion(
self,
) -> Generator[ChatMessageType, None, None]:
fixed_responses: ChatMessageType = json.loads(
self.config.fixed_chat_responses,
)
return self._get_from_playback_completion(fixed_responses)
def _get_from_fixed_embedding(
self,
) -> List[float]:
fixed_responses: List[float] = json.loads(self.config.fixed_embedding_responses)
return fixed_responses
def _get_from_playback_completion(
self,
cached_value: ChatMessageType,
) -> Generator[ChatMessageType, None, None]:
if self.config.playback_delay < 0:
yield cached_value
return
role: ChatMessageRoleType = cached_value["role"] # type: ignore
content = cached_value["content"]
cur_pos = 0
while cur_pos < len(content):
chunk_size = random.randint(3, 20)
next_pos = min(cur_pos + chunk_size, len(content))
yield format_chat_message(role, content[cur_pos:next_pos])
cur_pos = next_pos
time.sleep(self.config.playback_delay)
|