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
|
import logging
import math
import os
from copy import copy
from typing import Any
import litellm
import requests
from aiohttp import ClientError, ClientSession
from litellm import AuthenticationError, aembedding, embedding
from core.base import (
ChunkSearchResult,
EmbeddingConfig,
EmbeddingProvider,
EmbeddingPurpose,
R2RException,
)
logger = logging.getLogger()
class LiteLLMEmbeddingProvider(EmbeddingProvider):
def __init__(
self,
config: EmbeddingConfig,
*args,
**kwargs,
) -> None:
super().__init__(config)
self.litellm_embedding = embedding
self.litellm_aembedding = aembedding
provider = config.provider
if not provider:
raise ValueError(
"Must set provider in order to initialize `LiteLLMEmbeddingProvider`."
)
if provider != "litellm":
raise ValueError(
"LiteLLMEmbeddingProvider must be initialized with provider `litellm`."
)
self.rerank_url = None
if config.rerank_model:
if "huggingface" not in config.rerank_model:
raise ValueError(
"LiteLLMEmbeddingProvider only supports re-ranking via the HuggingFace text-embeddings-inference API"
)
url = os.getenv("HUGGINGFACE_API_BASE") or config.rerank_url
if not url:
raise ValueError(
"LiteLLMEmbeddingProvider requires a valid reranking API url to be set via `embedding.rerank_url` in the r2r.toml, or via the environment variable `HUGGINGFACE_API_BASE`."
)
self.rerank_url = url
self.base_model = config.base_model
if "amazon" in self.base_model:
logger.warn("Amazon embedding model detected, dropping params")
litellm.drop_params = True
self.base_dimension = config.base_dimension
def _get_embedding_kwargs(self, **kwargs):
embedding_kwargs = {
"model": self.base_model,
"dimensions": self.base_dimension,
}
embedding_kwargs.update(kwargs)
return embedding_kwargs
async def _execute_task(self, task: dict[str, Any]) -> list[list[float]]:
texts = task["texts"]
kwargs = self._get_embedding_kwargs(**task.get("kwargs", {}))
if "dimensions" in kwargs and math.isnan(kwargs["dimensions"]):
kwargs.pop("dimensions")
logger.warning("Dropping nan dimensions from kwargs")
try:
response = await self.litellm_aembedding(
input=texts,
**kwargs,
)
return [data["embedding"] for data in response.data]
except AuthenticationError:
logger.error(
"Authentication error: Invalid API key or credentials."
)
raise
except Exception as e:
error_msg = f"Error getting embeddings: {str(e)}"
logger.error(error_msg)
raise R2RException(error_msg, 400) from e
def _execute_task_sync(self, task: dict[str, Any]) -> list[list[float]]:
texts = task["texts"]
kwargs = self._get_embedding_kwargs(**task.get("kwargs", {}))
try:
response = self.litellm_embedding(
input=texts,
**kwargs,
)
return [data["embedding"] for data in response.data]
except AuthenticationError:
logger.error(
"Authentication error: Invalid API key or credentials."
)
raise
except Exception as e:
error_msg = f"Error getting embeddings: {str(e)}"
logger.error(error_msg)
raise R2RException(error_msg, 400) from e
async def async_get_embedding(
self,
text: str,
stage: EmbeddingProvider.Step = EmbeddingProvider.Step.BASE,
purpose: EmbeddingPurpose = EmbeddingPurpose.INDEX,
**kwargs,
) -> list[float]:
if stage != EmbeddingProvider.Step.BASE:
raise ValueError(
"LiteLLMEmbeddingProvider only supports search stage."
)
task = {
"texts": [text],
"stage": stage,
"purpose": purpose,
"kwargs": kwargs,
}
return (await self._execute_with_backoff_async(task))[0]
def get_embedding(
self,
text: str,
stage: EmbeddingProvider.Step = EmbeddingProvider.Step.BASE,
purpose: EmbeddingPurpose = EmbeddingPurpose.INDEX,
**kwargs,
) -> list[float]:
if stage != EmbeddingProvider.Step.BASE:
raise ValueError(
"Error getting embeddings: LiteLLMEmbeddingProvider only supports search stage."
)
task = {
"texts": [text],
"stage": stage,
"purpose": purpose,
"kwargs": kwargs,
}
return self._execute_with_backoff_sync(task)[0]
async def async_get_embeddings(
self,
texts: list[str],
stage: EmbeddingProvider.Step = EmbeddingProvider.Step.BASE,
purpose: EmbeddingPurpose = EmbeddingPurpose.INDEX,
**kwargs,
) -> list[list[float]]:
if stage != EmbeddingProvider.Step.BASE:
raise ValueError(
"LiteLLMEmbeddingProvider only supports search stage."
)
task = {
"texts": texts,
"stage": stage,
"purpose": purpose,
"kwargs": kwargs,
}
return await self._execute_with_backoff_async(task)
def get_embeddings(
self,
texts: list[str],
stage: EmbeddingProvider.Step = EmbeddingProvider.Step.BASE,
purpose: EmbeddingPurpose = EmbeddingPurpose.INDEX,
**kwargs,
) -> list[list[float]]:
if stage != EmbeddingProvider.Step.BASE:
raise ValueError(
"LiteLLMEmbeddingProvider only supports search stage."
)
task = {
"texts": texts,
"stage": stage,
"purpose": purpose,
"kwargs": kwargs,
}
return self._execute_with_backoff_sync(task)
def rerank(
self,
query: str,
results: list[ChunkSearchResult],
stage: EmbeddingProvider.Step = EmbeddingProvider.Step.RERANK,
limit: int = 10,
):
if self.config.rerank_model is not None:
if not self.rerank_url:
raise ValueError(
"Error, `rerank_url` was expected to be set inside LiteLLMEmbeddingProvider"
)
texts = [result.text for result in results]
payload = {
"query": query,
"texts": texts,
"model-id": self.config.rerank_model.split("huggingface/")[1],
}
headers = {"Content-Type": "application/json"}
try:
response = requests.post(
self.rerank_url, json=payload, headers=headers
)
response.raise_for_status()
reranked_results = response.json()
# Copy reranked results into new array
scored_results = []
for rank_info in reranked_results:
original_result = results[rank_info["index"]]
copied_result = copy(original_result)
# Inject the reranking score into the result object
copied_result.score = rank_info["score"]
scored_results.append(copied_result)
# Return only the ChunkSearchResult objects, limited to specified count
return scored_results[:limit]
except requests.RequestException as e:
logger.error(f"Error during reranking: {str(e)}")
# Fall back to returning the original results if reranking fails
return results[:limit]
else:
return results[:limit]
async def arerank(
self,
query: str,
results: list[ChunkSearchResult],
stage: EmbeddingProvider.Step = EmbeddingProvider.Step.RERANK,
limit: int = 10,
) -> list[ChunkSearchResult]:
"""Asynchronously rerank search results using the configured rerank
model.
Args:
query: The search query string
results: List of ChunkSearchResult objects to rerank
limit: Maximum number of results to return
Returns:
List of reranked ChunkSearchResult objects, limited to specified count
"""
if self.config.rerank_model is not None:
if not self.rerank_url:
raise ValueError(
"Error, `rerank_url` was expected to be set inside LiteLLMEmbeddingProvider"
)
texts = [result.text for result in results]
payload = {
"query": query,
"texts": texts,
"model-id": self.config.rerank_model.split("huggingface/")[1],
}
headers = {"Content-Type": "application/json"}
try:
async with ClientSession() as session:
async with session.post(
self.rerank_url, json=payload, headers=headers
) as response:
response.raise_for_status()
reranked_results = await response.json()
# Copy reranked results into new array
scored_results = []
for rank_info in reranked_results:
original_result = results[rank_info["index"]]
copied_result = copy(original_result)
# Inject the reranking score into the result object
copied_result.score = rank_info["score"]
scored_results.append(copied_result)
# Return only the ChunkSearchResult objects, limited to specified count
return scored_results[:limit]
except (ClientError, Exception) as e:
logger.error(f"Error during async reranking: {str(e)}")
# Fall back to returning the original results if reranking fails
return results[:limit]
else:
return results[:limit]
|
