File size: 9,488 Bytes
3079197 484e5ab 3079197 a86164e c9a1362 79ada0b 8f39e7a 3079197 8f9784a 3079197 8f39e7a da67205 c87ddd7 ba51460 8f39e7a da67205 ba51460 8f39e7a c9a1362 da67205 c9a1362 ba51460 3079197 79ada0b 3079197 e32ef75 3079197 2d09c38 0c30cc9 3079197 484e5ab 3079197 c87ddd7 3079197 79ada0b 3079197 9bf75d4 3079197 ba51460 e06e08c 3079197 c87ddd7 3079197 c87ddd7 e32ef75 c87ddd7 e32ef75 3079197 e06e08c 3079197 e32ef75 3079197 c87ddd7 e32ef75 3079197 79ada0b e32ef75 a8294f2 e32ef75 a8294f2 e32ef75 79ada0b 5e0a689 e06e08c 5e0a689 adb65d7 ba51460 adb65d7 5e0a689 adb65d7 8f9784a ba51460 8f9784a ba51460 8f9784a 63df91a a86164e ba51460 a86164e ba51460 a86164e ba51460 a86164e 63df91a a86164e ba51460 3069c36 ba51460 3069c36 ba51460 3069c36 22390c0 ba51460 3069c36 ba51460 3069c36 ba51460 3069c36 ba51460 |
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 |
#
# Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
from typing import Optional
from huggingface_hub import snapshot_download
from zhipuai import ZhipuAI
import os
from abc import ABC
from ollama import Client
import dashscope
from openai import OpenAI
from FlagEmbedding import FlagModel
import torch
import numpy as np
from api.utils.file_utils import get_project_base_directory, get_home_cache_dir
from rag.utils import num_tokens_from_string, truncate
try:
flag_model = FlagModel(os.path.join(get_home_cache_dir(), "bge-large-zh-v1.5"),
query_instruction_for_retrieval="为这个句子生成表示以用于检索相关文章：",
use_fp16=torch.cuda.is_available())
except Exception as e:
model_dir = snapshot_download(repo_id="BAAI/bge-large-zh-v1.5",
local_dir=os.path.join(get_home_cache_dir(), "bge-large-zh-v1.5"),
local_dir_use_symlinks=False)
flag_model = FlagModel(model_dir,
query_instruction_for_retrieval="为这个句子生成表示以用于检索相关文章：",
use_fp16=torch.cuda.is_available())
class Base(ABC):
def __init__(self, key, model_name):
pass
def encode(self, texts: list, batch_size=32):
raise NotImplementedError("Please implement encode method!")
def encode_queries(self, text: str):
raise NotImplementedError("Please implement encode method!")
class DefaultEmbedding(Base):
def __init__(self, *args, **kwargs):
"""
If you have trouble downloading HuggingFace models, -_^ this might help!!
For Linux:
export HF_ENDPOINT=https://hf-mirror.com
For Windows:
Good luck
^_-
"""
self.model = flag_model
def encode(self, texts: list, batch_size=32):
texts = [truncate(t, 2048) for t in texts]
token_count = 0
for t in texts:
token_count += num_tokens_from_string(t)
res = []
for i in range(0, len(texts), batch_size):
res.extend(self.model.encode(texts[i:i + batch_size]).tolist())
return np.array(res), token_count
def encode_queries(self, text: str):
token_count = num_tokens_from_string(text)
return self.model.encode_queries([text]).tolist()[0], token_count
class OpenAIEmbed(Base):
def __init__(self, key, model_name="text-embedding-ada-002",
base_url="https://api.openai.com/v1"):
if not base_url:
base_url = "https://api.openai.com/v1"
self.client = OpenAI(api_key=key, base_url=base_url)
self.model_name = model_name
def encode(self, texts: list, batch_size=32):
texts = [truncate(t, 8196) for t in texts]
res = self.client.embeddings.create(input=texts,
model=self.model_name)
return np.array([d.embedding for d in res.data]
), res.usage.total_tokens
def encode_queries(self, text):
res = self.client.embeddings.create(input=[truncate(text, 8196)],
model=self.model_name)
return np.array(res.data[0].embedding), res.usage.total_tokens
class QWenEmbed(Base):
def __init__(self, key, model_name="text_embedding_v2", **kwargs):
dashscope.api_key = key
self.model_name = model_name
def encode(self, texts: list, batch_size=10):
import dashscope
res = []
token_count = 0
texts = [truncate(t, 2048) for t in texts]
for i in range(0, len(texts), batch_size):
resp = dashscope.TextEmbedding.call(
model=self.model_name,
input=texts[i:i + batch_size],
text_type="document"
)
embds = [[] for _ in range(len(resp["output"]["embeddings"]))]
for e in resp["output"]["embeddings"]:
embds[e["text_index"]] = e["embedding"]
res.extend(embds)
token_count += resp["usage"]["total_tokens"]
return np.array(res), token_count
def encode_queries(self, text):
resp = dashscope.TextEmbedding.call(
model=self.model_name,
input=text[:2048],
text_type="query"
)
return np.array(resp["output"]["embeddings"][0]
["embedding"]), resp["usage"]["total_tokens"]
class ZhipuEmbed(Base):
def __init__(self, key, model_name="embedding-2", **kwargs):
self.client = ZhipuAI(api_key=key)
self.model_name = model_name
def encode(self, texts: list, batch_size=32):
arr = []
tks_num = 0
for txt in texts:
res = self.client.embeddings.create(input=txt,
model=self.model_name)
arr.append(res.data[0].embedding)
tks_num += res.usage.total_tokens
return np.array(arr), tks_num
def encode_queries(self, text):
res = self.client.embeddings.create(input=text,
model=self.model_name)
return np.array(res.data[0].embedding), res.usage.total_tokens
class OllamaEmbed(Base):
def __init__(self, key, model_name, **kwargs):
self.client = Client(host=kwargs["base_url"])
self.model_name = model_name
def encode(self, texts: list, batch_size=32):
arr = []
tks_num = 0
for txt in texts:
res = self.client.embeddings(prompt=txt,
model=self.model_name)
arr.append(res["embedding"])
tks_num += 128
return np.array(arr), tks_num
def encode_queries(self, text):
res = self.client.embeddings(prompt=text,
model=self.model_name)
return np.array(res["embedding"]), 128
class FastEmbed(Base):
def __init__(
self,
key: Optional[str] = None,
model_name: str = "BAAI/bge-small-en-v1.5",
cache_dir: Optional[str] = None,
threads: Optional[int] = None,
**kwargs,
):
from fastembed import TextEmbedding
self._model = TextEmbedding(model_name, cache_dir, threads, **kwargs)
def encode(self, texts: list, batch_size=32):
# Using the internal tokenizer to encode the texts and get the total
# number of tokens
encodings = self._model.model.tokenizer.encode_batch(texts)
total_tokens = sum(len(e) for e in encodings)
embeddings = [e.tolist() for e in self._model.embed(texts, batch_size)]
return np.array(embeddings), total_tokens
def encode_queries(self, text: str):
# Using the internal tokenizer to encode the texts and get the total
# number of tokens
encoding = self._model.model.tokenizer.encode(text)
embedding = next(self._model.query_embed(text)).tolist()
return np.array(embedding), len(encoding.ids)
class XinferenceEmbed(Base):
def __init__(self, key, model_name="", base_url=""):
self.client = OpenAI(api_key="xxx", base_url=base_url)
self.model_name = model_name
def encode(self, texts: list, batch_size=32):
res = self.client.embeddings.create(input=texts,
model=self.model_name)
return np.array([d.embedding for d in res.data]
), res.usage.total_tokens
def encode_queries(self, text):
res = self.client.embeddings.create(input=[text],
model=self.model_name)
return np.array(res.data[0].embedding), res.usage.total_tokens
class YoudaoEmbed(Base):
_client = None
def __init__(self, key=None, model_name="maidalun1020/bce-embedding-base_v1", **kwargs):
from BCEmbedding import EmbeddingModel as qanthing
if not YoudaoEmbed._client:
try:
print("LOADING BCE...")
YoudaoEmbed._client = qanthing(model_name_or_path=os.path.join(
get_home_cache_dir(),
"bce-embedding-base_v1"))
except Exception as e:
YoudaoEmbed._client = qanthing(
model_name_or_path=model_name.replace(
"maidalun1020", "InfiniFlow"))
def encode(self, texts: list, batch_size=10):
res = []
token_count = 0
for t in texts:
token_count += num_tokens_from_string(t)
for i in range(0, len(texts), batch_size):
embds = YoudaoEmbed._client.encode(texts[i:i + batch_size])
res.extend(embds)
return np.array(res), token_count
def encode_queries(self, text):
embds = YoudaoEmbed._client.encode([text])
return np.array(embds[0]), num_tokens_from_string(text)
|