app.py

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel, Field
from typing import List, Optional
import numpy as np

from sentence_transformers import SentenceTransformer

app = FastAPI(title="FRIDA Embedding API", version="1.0")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

MODEL_NAME = "ai-forever/FRIDA"
model = SentenceTransformer(MODEL_NAME)
EMBED_DIM = model.get_sentence_embedding_dimension()

SUPPORTED_PROMPTS = [
    "search_query",
    "search_document",
    "paraphrase",
    "categorize",
    "categorize_sentiment",
    "categorize_topic",
    "categorize_entailment",
]

class EmbedRequest(BaseModel):
    texts: List[str] = Field(..., description="Список текстов")
    prompt_name: Optional[str] = Field("search_document", description="FRIDA prompt_name")

class EmbedResponse(BaseModel):
    embeddings: List[List[float]]
    dim: int

@app.get("/health")
def health():
    return {"status": "ok"}

@app.get("/metadata")
def metadata():
    return {
        "model": MODEL_NAME,
        "embedding_dim": EMBED_DIM,
        "pooling": "cls",
        "prompts_supported": SUPPORTED_PROMPTS,
    }

@app.post("/embed", response_model=EmbedResponse)
def embed(req: EmbedRequest):
    if not req.texts:
        raise HTTPException(status_code=400, detail="texts must be non-empty")
    prompt = req.prompt_name or "search_document"
    if prompt not in SUPPORTED_PROMPTS:
        raise HTTPException(status_code=400, detail=f"Unsupported prompt_name: {prompt}")

    vectors = model.encode(
        req.texts,
        convert_to_numpy=True,
        prompt_name=prompt,
        normalize_embeddings=True,
        batch_size=min(16, max(1, len(req.texts))),
        show_progress_bar=False,
    ).astype(np.float32)

    return {"embeddings": vectors.tolist(), "dim": int(vectors.shape[1])}