Spaces:
Running
Running
File size: 9,678 Bytes
0231daa |
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 |
"""
Single/Batch embedding generation endpoints.
This module provides routes for generating embeddings for
multiple texts in a single request.
"""
import time
from typing import Union
from fastapi import APIRouter, Depends, HTTPException, status
from loguru import logger
from src.models.schemas import (
EmbedRequest,
DenseEmbedResponse,
SparseEmbedResponse,
SparseEmbedding,
)
from src.core.manager import ModelManager
from src.core.cache import EmbeddingCache
from src.core.exceptions import (
ModelNotFoundError,
ModelNotLoadedError,
EmbeddingGenerationError,
ValidationError,
)
from src.api.dependencies import get_model_manager, get_cache_if_enabled
from src.utils.validators import extract_embedding_kwargs, validate_texts
from src.config.settings import get_settings
router = APIRouter(prefix="/embeddings", tags=["embeddings"])
@router.post(
"/embed",
response_model=Union[DenseEmbedResponse, SparseEmbedResponse],
summary="Generate single/batch embeddings spesialization document",
description="Generate embeddings for multiple texts in a single request",
)
async def create_embeddings_document(
request: EmbedRequest,
manager: ModelManager = Depends(get_model_manager),
cache: EmbeddingCache = Depends(get_cache_if_enabled),
settings=Depends(get_settings),
):
"""
Generate embeddings for multiple texts.
This endpoint efficiently processes multiple texts in a single batch,
reducing overhead compared to multiple single requests.
Args:
request: BatchEmbedRequest with texts, model_id, and optional parameters
manager: Model manager dependency
cache: Cache dependency (if enabled)
settings: Application settings
Returns:
DenseEmbedResponse or SparseEmbedResponse depending on model type
Raises:
HTTPException: On validation or generation errors
"""
try:
# Validate input
validate_texts(
request.texts,
max_length=settings.MAX_TEXT_LENGTH,
max_batch_size=settings.MAX_BATCH_SIZE,
)
# Extract kwargs
kwargs = extract_embedding_kwargs(request)
# Check cache first (batch requests typically not cached due to size)
# But we can cache if batch is small
if cache is not None and len(request.texts) <= 10:
cache_key = str(sorted(request.texts)) # Simple key for small batches
cached_result = cache.get(
texts=cache_key,
model_id=request.model_id,
prompt=request.prompt,
**kwargs,
)
if cached_result is not None:
logger.debug(f"Cache hit for batch (size={len(request.texts)})")
return cached_result
# Get model
model = manager.get_model(request.model_id)
config = manager.model_configs[request.model_id]
start_time = time.time()
# Generate embeddings based on model type
if config.type == "sparse-embeddings":
# Sparse batch embeddings
sparse_results = model.embed_documents(
texts=request.texts, prompt=request.prompt, **kwargs
)
processing_time = time.time() - start_time
# Convert to SparseEmbedding objects
sparse_embeddings = []
for idx, sparse_result in enumerate(sparse_results):
sparse_embeddings.append(
SparseEmbedding(
text=request.texts[idx],
indices=sparse_result["indices"],
values=sparse_result["values"],
)
)
response = SparseEmbedResponse(
embeddings=sparse_embeddings,
count=len(sparse_embeddings),
model_id=request.model_id,
processing_time=processing_time,
)
else:
# Dense batch embeddings
embeddings = model.embed_documents(
texts=request.texts, prompt=request.prompt, **kwargs
)
processing_time = time.time() - start_time
response = DenseEmbedResponse(
embeddings=embeddings,
dimension=len(embeddings[0]) if embeddings else 0,
count=len(embeddings),
model_id=request.model_id,
processing_time=processing_time,
)
# Cache small batches
if cache is not None and len(request.texts) <= 10:
cache_key = str(sorted(request.texts))
cache.set(
texts=cache_key,
model_id=request.model_id,
result=response,
prompt=request.prompt,
**kwargs,
)
logger.info(
f"Generated {len(request.texts)} embeddings "
f"in {processing_time:.3f}s ({len(request.texts) / processing_time:.1f} texts/s)"
)
return response
except (ValidationError, ModelNotFoundError) as e:
raise HTTPException(status_code=e.status_code, detail=e.message)
except ModelNotLoadedError as e:
raise HTTPException(status_code=e.status_code, detail=e.message)
except EmbeddingGenerationError as e:
raise HTTPException(status_code=e.status_code, detail=e.message)
except Exception as e:
logger.exception("Unexpected error in create_embeddings_document")
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
detail=f"Failed to create batch embeddings: {str(e)}",
)
@router.post(
"/query",
response_model=Union[DenseEmbedResponse, SparseEmbedResponse],
summary="Generate single/batch embeddings spesialization document",
description="Generate embedding for a multiple query text",
)
async def create_query_embedding(
request: EmbedRequest,
manager: ModelManager = Depends(get_model_manager),
cache: EmbeddingCache = Depends(get_cache_if_enabled),
settings=Depends(get_settings),
):
"""
Generate a single/batch query embedding.
This endpoint creates embeddings optimized for search queries.
Some models differentiate between query and document embeddings.
Args:
request: EmbedRequest with text, model_id, and optional parameters
manager: Model manager dependency
cache: Cache dependency (if enabled)
settings: Application settings
Returns:
DenseEmbedResponse or SparseEmbedResponse depending on model type
Raises:
HTTPException: On validation or generation errors
"""
try:
# Validate input
validate_texts(request.texts)
# Extract kwargs
kwargs = extract_embedding_kwargs(request)
# Check cache (with 'query' prefix in key)
cache_key_kwargs = {"endpoint": "query", **kwargs}
if cache is not None:
cached_result = cache.get(
texts=request.text,
model_id=request.model_id,
prompt=request.prompt,
**cache_key_kwargs,
)
if cached_result is not None:
logger.debug(f"Cache hit for query model {request.model_id}")
return cached_result
# Get model
model = manager.get_model(request.model_id)
config = manager.model_configs[request.model_id]
start_time = time.time()
# Generate embedding based on model type
if config.type == "sparse-embeddings":
# Sparse embedding
sparse_results = model.embed_query(
texts=request.texts, prompt=request.prompt, **kwargs
)
processing_time = time.time() - start_time
sparse_result = sparse_results[0]
sparse_embedding = SparseEmbedding(
text=request.texts,
indices=sparse_result["indices"],
values=sparse_result["values"],
)
response = SparseEmbedResponse(
sparse_embedding=sparse_embedding,
model_id=request.model_id,
processing_time=processing_time,
)
else:
# Dense embedding
embeddings = model.embed_query(
texts=request.texts, prompt=request.prompt, **kwargs
)
processing_time = time.time() - start_time
response = DenseEmbedResponse(
embedding=embeddings[0],
dimension=len(embeddings[0]),
model_id=request.model_id,
processing_time=processing_time,
)
# Cache the result
if cache is not None:
cache.set(
texts=request.texts,
model_id=request.model_id,
result=response,
prompt=request.prompt,
**cache_key_kwargs,
)
return response
except (ValidationError, ModelNotFoundError) as e:
raise HTTPException(status_code=e.status_code, detail=e.message)
except ModelNotLoadedError as e:
raise HTTPException(status_code=e.status_code, detail=e.message)
except EmbeddingGenerationError as e:
raise HTTPException(status_code=e.status_code, detail=e.message)
except Exception as e:
logger.exception("Unexpected error in create_query_embedding")
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
detail=f"Failed to create query embedding: {str(e)}",
)
|