File size: 6,993 Bytes
7db0ae4 |
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 |
import os, types
import json
from enum import Enum
import requests
import time
from typing import Callable, Optional
import litellm
from litellm.utils import ModelResponse, Usage
from .prompt_templates.factory import prompt_factory, custom_prompt
class PetalsError(Exception):
def __init__(self, status_code, message):
self.status_code = status_code
self.message = message
super().__init__(
self.message
) # Call the base class constructor with the parameters it needs
class PetalsConfig:
"""
Reference: https://github.com/petals-infra/chat.petals.dev#post-apiv1generate
The `PetalsConfig` class encapsulates the configuration for the Petals API. The properties of this class are described below:
- `max_length` (integer): This represents the maximum length of the generated text (including the prefix) in tokens.
- `max_new_tokens` (integer): This represents the maximum number of newly generated tokens (excluding the prefix).
The generation parameters are compatible with `.generate()` from Hugging Face's Transformers library:
- `do_sample` (boolean, optional): If set to 0 (default), the API runs greedy generation. If set to 1, the API performs sampling using the parameters below:
- `temperature` (float, optional): This value sets the temperature for sampling.
- `top_k` (integer, optional): This value sets the limit for top-k sampling.
- `top_p` (float, optional): This value sets the limit for top-p (nucleus) sampling.
- `repetition_penalty` (float, optional): This helps apply the repetition penalty during text generation, as discussed in this paper.
"""
max_length: Optional[int] = None
max_new_tokens: Optional[
int
] = litellm.max_tokens # petals requires max tokens to be set
do_sample: Optional[bool] = None
temperature: Optional[float] = None
top_k: Optional[int] = None
top_p: Optional[float] = None
repetition_penalty: Optional[float] = None
def __init__(
self,
max_length: Optional[int] = None,
max_new_tokens: Optional[
int
] = litellm.max_tokens, # petals requires max tokens to be set
do_sample: Optional[bool] = None,
temperature: Optional[float] = None,
top_k: Optional[int] = None,
top_p: Optional[float] = None,
repetition_penalty: Optional[float] = None,
) -> None:
locals_ = locals()
for key, value in locals_.items():
if key != "self" and value is not None:
setattr(self.__class__, key, value)
@classmethod
def get_config(cls):
return {
k: v
for k, v in cls.__dict__.items()
if not k.startswith("__")
and not isinstance(
v,
(
types.FunctionType,
types.BuiltinFunctionType,
classmethod,
staticmethod,
),
)
and v is not None
}
def completion(
model: str,
messages: list,
api_base: Optional[str],
model_response: ModelResponse,
print_verbose: Callable,
encoding,
logging_obj,
optional_params=None,
stream=False,
litellm_params=None,
logger_fn=None,
):
## Load Config
config = litellm.PetalsConfig.get_config()
for k, v in config.items():
if (
k not in optional_params
): # completion(top_k=3) > petals_config(top_k=3) <- allows for dynamic variables to be passed in
optional_params[k] = v
if model in litellm.custom_prompt_dict:
# check if the model has a registered custom prompt
model_prompt_details = litellm.custom_prompt_dict[model]
prompt = custom_prompt(
role_dict=model_prompt_details["roles"],
initial_prompt_value=model_prompt_details["initial_prompt_value"],
final_prompt_value=model_prompt_details["final_prompt_value"],
messages=messages,
)
else:
prompt = prompt_factory(model=model, messages=messages)
if api_base:
## LOGGING
logging_obj.pre_call(
input=prompt,
api_key="",
additional_args={
"complete_input_dict": optional_params,
"api_base": api_base,
},
)
data = {"model": model, "inputs": prompt, **optional_params}
## COMPLETION CALL
response = requests.post(api_base, data=data)
## LOGGING
logging_obj.post_call(
input=prompt,
api_key="",
original_response=response.text,
additional_args={"complete_input_dict": optional_params},
)
## RESPONSE OBJECT
try:
output_text = response.json()["outputs"]
except Exception as e:
PetalsError(status_code=response.status_code, message=str(e))
else:
try:
import torch
from transformers import AutoTokenizer
from petals import AutoDistributedModelForCausalLM # type: ignore
except:
raise Exception(
"Importing torch, transformers, petals failed\nTry pip installing petals \npip install git+https://github.com/bigscience-workshop/petals"
)
model = model
tokenizer = AutoTokenizer.from_pretrained(
model, use_fast=False, add_bos_token=False
)
model_obj = AutoDistributedModelForCausalLM.from_pretrained(model)
## LOGGING
logging_obj.pre_call(
input=prompt,
api_key="",
additional_args={"complete_input_dict": optional_params},
)
## COMPLETION CALL
inputs = tokenizer(prompt, return_tensors="pt")["input_ids"]
# optional params: max_new_tokens=1,temperature=0.9, top_p=0.6
outputs = model_obj.generate(inputs, **optional_params)
## LOGGING
logging_obj.post_call(
input=prompt,
api_key="",
original_response=outputs,
additional_args={"complete_input_dict": optional_params},
)
## RESPONSE OBJECT
output_text = tokenizer.decode(outputs[0])
if len(output_text) > 0:
model_response["choices"][0]["message"]["content"] = output_text
prompt_tokens = len(encoding.encode(prompt))
completion_tokens = len(
encoding.encode(model_response["choices"][0]["message"].get("content"))
)
model_response["created"] = int(time.time())
model_response["model"] = model
usage = Usage(
prompt_tokens=prompt_tokens,
completion_tokens=completion_tokens,
total_tokens=prompt_tokens + completion_tokens,
)
model_response.usage = usage
return model_response
def embedding():
# logic for parsing in - calling - parsing out model embedding calls
pass
|