Spaces:
Running
Running
Download app/bge_m3_api.py from SmartDataAlex/Jupyter: direct link, hf CLI and curl.
- Browser
- Download file 3.27 kB
-
https://huggingface.co/spaces/SmartDataAlex/Jupyter/resolve/main/app/bge_m3_api.py
- Command line
-
hf download hf://spaces/SmartDataAlex/Jupyter/app/bge_m3_api.py
-
curl -L -o bge_m3_api.py https://huggingface.co/spaces/SmartDataAlex/Jupyter/resolve/main/app/bge_m3_api.py
3.27 kB
| """Authenticated BGE-M3 triple-vector API, separate from dense-only APIs.""" | |
| from __future__ import annotations | |
| from typing import Literal | |
| from fastapi import APIRouter, Depends, HTTPException | |
| from pydantic import BaseModel, ConfigDict, Field, field_validator | |
| from app.bge_m3 import BGE_M3_ID | |
| from app.embedding import TooManyTokens | |
| from app.models import ModelRegistry, ServiceBusy | |
| from app.voyage_api import authenticate, resident_models | |
| router = APIRouter(prefix="/bge-m3", tags=["BGE-M3 triple embeddings"]) | |
| class BgeM3Request(BaseModel): | |
| model_config = ConfigDict(extra="forbid") | |
| input: str | list[str] = Field(description="One to eight texts (without prefixes)") | |
| input_type: Literal["query", "document"] = "document" | |
| return_colbert: bool = Field( | |
| default=False, | |
| description="Compute and include ColBERT multivectors (large response). " | |
| "Default false returns only dense and sparse.", | |
| ) | |
| def check_texts(cls, value): | |
| texts = [value] if isinstance(value, str) else value | |
| if not texts or len(texts) > 8: | |
| raise ValueError("BGE-M3 requires between 1 and 8 texts") | |
| if any(not text.strip() for text in texts): | |
| raise ValueError("BGE-M3 inputs cannot be blank") | |
| return value | |
| def encode_bge_m3(payload: BgeM3Request, | |
| registry: ModelRegistry = Depends(resident_models)): | |
| if registry.bge_m3 is None: | |
| raise HTTPException(status_code=503, detail="BGE-M3 model is disabled") | |
| texts = [payload.input] if isinstance(payload.input, str) else payload.input | |
| # BGE-M3 is symmetric: query and document use the same tokenizer, | |
| # no artificial 'query:' or 'passage:' prefix is added. | |
| try: | |
| results = registry.bge_m3_encode( | |
| texts, return_colbert=payload.return_colbert, | |
| ) | |
| except TooManyTokens as exc: | |
| raise HTTPException(status_code=422, detail=str(exc)) from exc | |
| except ValueError as exc: | |
| raise HTTPException(status_code=422, detail=str(exc)) from exc | |
| except ServiceBusy as exc: | |
| raise HTTPException(status_code=429, detail=str(exc), | |
| headers={"Retry-After": "2"}) from exc | |
| except Exception as exc: | |
| import logging | |
| logging.getLogger(__name__).exception("BGE-M3 inference failed") | |
| raise HTTPException(status_code=500, detail="BGE-M3 inference failed") from exc | |
| return { | |
| "model": BGE_M3_ID, | |
| "object": "list", | |
| "dense_dimensions": registry.bge_m3.dimensions, | |
| "colbert_dimensions": registry.bge_m3.colbert_dimensions, | |
| "data": [ | |
| { | |
| "index": index, | |
| "dense": item.dense, | |
| "sparse": { | |
| "indices": item.sparse_indices, | |
| "values": item.sparse_values, | |
| }, | |
| **({"colbert": item.colbert} if payload.return_colbert else {}), | |
| "tokens_count": item.tokens_count, | |
| } | |
| for index, item in enumerate(results) | |
| ], | |
| "usage": {"total_tokens": sum(item.tokens_count for item in results)}, | |
| } | |