File size: 9,798 Bytes
9d5b280 |
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 |
import os
from functools import cached_property
from operator import itemgetter
from typing import Any, Dict, List, Optional, Tuple, Union
from lm_eval.api.registry import register_model
from lm_eval.models.api_models import TemplateAPI
from lm_eval.models.utils import handle_stop_sequences
from lm_eval.utils import eval_logger
@register_model("local-completions")
class LocalCompletionsAPI(TemplateAPI):
def __init__(
self,
base_url=None,
tokenizer_backend="huggingface",
**kwargs,
):
super().__init__(
base_url=base_url, tokenizer_backend=tokenizer_backend, **kwargs
)
def _create_payload(
self,
messages: Union[List[List[int]], List[dict], List[str], str],
generate=False,
gen_kwargs: Optional[dict] = None,
seed: int = 1234,
eos=None,
**kwargs,
) -> dict:
if generate:
gen_kwargs.pop("do_sample", False)
if "max_tokens" in gen_kwargs:
max_tokens = gen_kwargs.pop("max_tokens")
else:
max_tokens = gen_kwargs.pop("max_gen_toks", self._max_gen_toks)
temperature = gen_kwargs.pop("temperature", 0)
stop = handle_stop_sequences(gen_kwargs.pop("until", None), eos)
return {
"prompt": messages,
"model": self.model,
"max_tokens": max_tokens,
"temperature": temperature,
"stop": stop,
"seed": seed,
**gen_kwargs,
}
else:
return {
"model": self.model,
"prompt": messages,
"temperature": 0,
"max_tokens": 1,
"logprobs": 1,
"seed": seed,
"echo": True,
}
@staticmethod
def parse_logprobs(
outputs: Union[Dict, List[Dict]],
tokens: List[List[int]] = None,
ctxlens: List[int] = None,
**kwargs,
) -> List[Tuple[float, bool]]:
res = []
if not isinstance(outputs, list):
outputs = [outputs]
for out in outputs:
for choice, ctxlen in zip(
sorted(out["choices"], key=itemgetter("index")), ctxlens
):
assert ctxlen > 0, "Context length must be greater than 0"
logprobs = sum(choice["logprobs"]["token_logprobs"][ctxlen:-1])
tokens_logprobs = choice["logprobs"]["token_logprobs"][ctxlen:-1]
top_logprobs = choice["logprobs"]["top_logprobs"][ctxlen:-1]
is_greedy = True
for tok, top in zip(tokens_logprobs, top_logprobs):
if tok != max(top.values()):
is_greedy = False
break
res.append((logprobs, is_greedy))
return res
@staticmethod
def parse_generations(outputs: Union[Dict, List[Dict]], **kwargs) -> List[str]:
res = []
if not isinstance(outputs, list):
outputs = [outputs]
for out in outputs:
tmp = [None] * len(out["choices"])
for choices in out["choices"]:
tmp[choices["index"]] = choices["text"]
res = res + tmp
return res
@property
def api_key(self):
return os.environ.get("OPENAI_API_KEY", "")
@register_model("local-chat-completions")
class LocalChatCompletion(LocalCompletionsAPI):
def __init__(
self,
base_url=None,
tokenizer_backend=None,
tokenized_requests=False,
**kwargs,
):
eval_logger.warning(
"chat-completions endpoint requires the `--apply_chat_template` flag."
)
super().__init__(
base_url=base_url,
tokenizer_backend=tokenizer_backend,
tokenized_requests=tokenized_requests,
**kwargs,
)
if self._batch_size > 1:
eval_logger.warning(
"Chat completions does not support batching. Defaulting to batch size 1."
)
self._batch_size = 1
def _create_payload(
self,
messages: List[Dict],
generate=False,
gen_kwargs: dict = None,
seed=1234,
eos=None,
**kwargs,
) -> dict:
assert type(messages) is not str, (
"chat-completions require the --apply_chat_template flag."
)
gen_kwargs.pop("do_sample", False)
if "max_tokens" in gen_kwargs:
max_tokens = gen_kwargs.pop("max_tokens")
else:
max_tokens = gen_kwargs.pop("max_gen_toks", self._max_gen_toks)
temperature = gen_kwargs.pop("temperature", 0)
stop = handle_stop_sequences(gen_kwargs.pop("until", None), eos)
if not isinstance(stop, (list, tuple)):
stop = [stop]
return {
"messages": messages,
"model": self.model,
"max_tokens": max_tokens,
"temperature": temperature,
"stop": stop[:4],
"seed": seed,
**gen_kwargs,
}
@staticmethod
def parse_generations(outputs: Union[Dict, List[Dict]], **kwargs) -> List[str]:
res = []
if not isinstance(outputs, list):
outputs = [outputs]
for out in outputs:
tmp = [None] * len(out["choices"])
for choices in out["choices"]:
tmp[choices["index"]] = choices["message"]["content"]
res = res + tmp
return res
def tok_encode(
self,
string: Union[str, Any],
left_truncate_len=None,
add_special_tokens=None,
**kwargs,
) -> Union[List[str], List[int], Any]:
return string
def loglikelihood(self, requests, **kwargs):
raise NotImplementedError(
"Loglikelihood is not supported for chat completions. Consider using the completions API instead."
)
@register_model(
"openai-completions",
)
class OpenAICompletionsAPI(LocalCompletionsAPI):
def __init__(
self,
base_url="https://api.openai.com/v1/completions",
tokenizer_backend="tiktoken",
**kwargs,
):
super().__init__(
base_url=base_url, tokenizer_backend=tokenizer_backend, **kwargs
)
@cached_property
def api_key(self):
"""Override this property to return the API key for the API request."""
key = os.environ.get("OPENAI_API_KEY", None)
if key is None:
raise ValueError(
"API key not found. Please set the `OPENAI_API_KEY` environment variable."
)
return key
def loglikelihood(self, requests, **kwargs):
assert self.model in [
"babbage-002",
"davinci-002",
], (
f"Prompt loglikelihoods are only supported by OpenAI's API for {['babbage-002', 'davinci-002']}."
)
return super().loglikelihood(requests, **kwargs)
def chat_template(self, chat_template: Union[bool, str] = False) -> Optional[str]:
return ""
@register_model("openai-chat-completions")
class OpenAIChatCompletion(LocalChatCompletion):
def __init__(
self,
base_url="https://api.openai.com/v1/chat/completions",
tokenizer_backend=None,
tokenized_requests=False,
**kwargs,
):
if "o1" in kwargs.get("model", ""):
eval_logger.warning(
"o1 models do not support `stop` and only support temperature=1"
)
super().__init__(
base_url=base_url,
tokenizer_backend=tokenizer_backend,
tokenized_requests=tokenized_requests,
**kwargs,
)
@cached_property
def api_key(self):
"""Override this property to return the API key for the API request."""
key = os.environ.get("OPENAI_API_KEY", None)
if key is None:
raise ValueError(
"API key not found. Please set the `OPENAI_API_KEY` environment variable."
)
return key
def loglikelihood(self, requests, **kwargs):
raise NotImplementedError(
"Loglikelihood (and therefore `multiple_choice`-type tasks) is not supported for chat completions as OpenAI does not provide prompt logprobs. See https://github.com/EleutherAI/lm-evaluation-harness/issues/942#issuecomment-1777836312 or https://github.com/EleutherAI/lm-evaluation-harness/issues/1196 for more background on this limitation."
)
def _create_payload(
self,
messages: List[Dict],
generate=False,
gen_kwargs: dict = None,
seed=1234,
eos="<|endoftext|>",
**kwargs,
) -> dict:
assert type(messages) is not str, (
"chat-completions require the --apply_chat_template flag."
)
gen_kwargs.pop("do_sample", False)
if "max_tokens" in gen_kwargs:
max_tokens = gen_kwargs.pop("max_tokens")
else:
max_tokens = gen_kwargs.pop("max_gen_toks", self._max_gen_toks)
temperature = gen_kwargs.pop("temperature", 0)
stop = handle_stop_sequences(gen_kwargs.pop("until", ["<|endoftext|>"]), eos)
if not isinstance(stop, (list, tuple)):
stop = [stop]
output = {
"messages": messages,
"model": self.model,
"max_completion_tokens": max_tokens,
"temperature": temperature,
"stop": stop[:4],
"seed": seed,
**gen_kwargs,
}
if "o1" in self.model:
output.pop("stop")
output["temperature"] = 1
return output
|