Spaces:
Running
Running
File size: 22,204 Bytes
80f110c ad80405 80f110c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 |
# -*- coding: utf-8 -*-
"""
HF Space - main.py de substitution pour tests Qdrant / indexation minimale
Fonctions clés :
- POST /wipe?project_id=XXX : supprime la collection Qdrant
- POST /index : lance un job d'indexation (JSON files=[{path,text},...])
- GET /status/{job_id} : état du job + logs
- GET /collections/{proj}/count : retourne le nombre de points dans Qdrant
- POST /query : recherche sémantique (top_k, text, project_id)
Une UI Gradio minimale est montée sur "/" pour déclencher les tests sans console.
ENV attendues :
- QDRANT_URL : ex. https://xxxxx.eu-central-1-0.aws.cloud.qdrant.io:6333
- QDRANT_API_KEY : clé Qdrant Cloud
- COLLECTION_PREFIX : défaut "proj_"
- EMB_PROVIDER : "hf" (défaut) ou "dummy"
- HF_EMBED_MODEL : défaut "BAAI/bge-m3"
- HUGGINGFACEHUB_API_TOKEN : token HF Inference (si EMB_PROVIDER=hf)
- LOG_LEVEL : DEBUG (défaut), INFO...
Dépendances (requirements) suggérées :
fastapi>=0.111
uvicorn>=0.30
httpx>=0.27
pydantic>=2.7
gradio>=4.43
numpy>=2.0
"""
from __future__ import annotations
import os
import time
import uuid
import math
import json
import hashlib
import logging
import asyncio
from typing import List, Dict, Any, Optional, Tuple
import numpy as np
import httpx
from pydantic import BaseModel, Field, ValidationError
from fastapi import FastAPI, HTTPException, Query
from fastapi.middleware.cors import CORSMiddleware
import gradio as gr
# ------------------------------------------------------------------------------
# Configuration & logs
# ------------------------------------------------------------------------------
LOG_LEVEL = os.getenv("LOG_LEVEL", "DEBUG").upper()
logging.basicConfig(
level=getattr(logging, LOG_LEVEL, logging.DEBUG),
format="%(asctime)s - %(levelname)s - %(message)s",
)
LOG = logging.getLogger("remote_indexer_min")
QDRANT_URL = os.getenv("QDRANT_URL", "").rstrip("/")
QDRANT_API_KEY = os.getenv("QDRANT_API_KEY", "")
COLLECTION_PREFIX = os.getenv("COLLECTION_PREFIX", "proj_").strip() or "proj_"
EMB_PROVIDER = os.getenv("EMB_PROVIDER", "hf").lower() # "hf" | "dummy"
HF_EMBED_MODEL = os.getenv("HF_EMBED_MODEL", "BAAI/bge-m3")
HF_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN", "")
if not QDRANT_URL or not QDRANT_API_KEY:
LOG.warning("QDRANT_URL / QDRANT_API_KEY non fournis : l'upsert échouera. Fournis-les dans les Secrets du Space.")
if EMB_PROVIDER == "hf" and not HF_TOKEN:
LOG.warning("EMB_PROVIDER=hf mais HUGGINGFACEHUB_API_TOKEN absent. Tu peux basculer EMB_PROVIDER=dummy pour tester sans token.")
# ------------------------------------------------------------------------------
# Schémas Pydantic
# ------------------------------------------------------------------------------
class FileItem(BaseModel):
path: str
text: str
class IndexRequest(BaseModel):
project_id: str = Field(..., min_length=1)
files: List[FileItem] = Field(default_factory=list)
chunk_size: int = Field(200, ge=64, le=4096)
overlap: int = Field(20, ge=0, le=512)
batch_size: int = Field(32, ge=1, le=1024)
store_text: bool = True
class QueryRequest(BaseModel):
project_id: str
text: str
top_k: int = Field(5, ge=1, le=100)
# ------------------------------------------------------------------------------
# Job store (en mémoire)
# ------------------------------------------------------------------------------
class JobState(BaseModel):
job_id: str
project_id: str
stage: str = "pending" # pending -> embedding -> upserting -> done/failed
total_files: int = 0
total_chunks: int = 0
embedded: int = 0
upserted: int = 0
errors: List[str] = Field(default_factory=list)
messages: List[str] = Field(default_factory=list)
started_at: float = Field(default_factory=time.time)
finished_at: Optional[float] = None
def log(self, msg: str) -> None:
stamp = time.strftime("%H:%M:%S")
line = f"[{stamp}] {msg}"
self.messages.append(line)
LOG.debug(f"[{self.job_id}] {msg}")
JOBS: Dict[str, JobState] = {}
# ------------------------------------------------------------------------------
# Utilitaires
# ------------------------------------------------------------------------------
def hash8(s: str) -> str:
return hashlib.sha256(s.encode("utf-8")).hexdigest()[:16]
def l2_normalize(vec: List[float]) -> List[float]:
arr = np.array(vec, dtype=np.float32)
n = float(np.linalg.norm(arr))
if n > 0:
arr = arr / n
return arr.astype(np.float32).tolist()
def flatten_any(x: Any) -> List[float]:
"""
Certaines APIs renvoient [[...]] ou [[[...]]]; on aplanit en 1D.
"""
if isinstance(x, (list, tuple)):
if len(x) > 0 and isinstance(x[0], (list, tuple)):
# Aplanit récursif
return flatten_any(x[0])
return list(map(float, x))
raise ValueError("Embedding vector mal formé")
def chunk_text(text: str, chunk_size: int, overlap: int) -> List[Tuple[int, int, str]]:
"""
Retourne une liste de (start, end, chunk_text)
Ignore les petits fragments (< 30 chars) pour éviter le bruit.
"""
text = text or ""
if not text.strip():
return []
res = []
n = len(text)
i = 0
while i < n:
j = min(i + chunk_size, n)
chunk = text[i:j]
if len(chunk.strip()) >= 30:
res.append((i, j, chunk))
i = j - overlap
if i <= 0:
i = j
return res
async def ensure_collection(client: httpx.AsyncClient, coll: str, vector_size: int) -> None:
"""
Crée ou ajuste la collection Qdrant (distance = Cosine).
"""
url = f"{QDRANT_URL}/collections/{coll}"
# Vérifie l'existence
r = await client.get(url, headers={"api-key": QDRANT_API_KEY}, timeout=20)
if r.status_code == 200:
# Optionnel: vérifier la taille du vecteur ; si mismatch, on peut supprimer/recréer
data = r.json()
existing_size = data.get("result", {}).get("vectors", {}).get("size")
if existing_size and int(existing_size) != int(vector_size):
LOG.warning(f"Collection {coll} dim={existing_size} ≠ attendu {vector_size} → recréation")
await client.delete(url, headers={"api-key": QDRANT_API_KEY}, timeout=20)
else:
LOG.debug(f"Collection {coll} déjà prête (dim={existing_size})")
# (Re)création
body = {
"vectors": {"size": vector_size, "distance": "Cosine"}
}
r2 = await client.put(url, headers={"api-key": QDRANT_API_KEY}, json=body, timeout=30)
if r2.status_code not in (200, 201):
raise HTTPException(status_code=500, detail=f"Qdrant PUT collection a échoué: {r2.text}")
async def qdrant_upsert(
client: httpx.AsyncClient,
coll: str,
points: List[Dict[str, Any]],
) -> int:
if not points:
return 0
url = f"{QDRANT_URL}/collections/{coll}/points?wait=true"
body = {"points": points}
r = await client.put(url, headers={"api-key": QDRANT_API_KEY}, json=body, timeout=60)
if r.status_code not in (200, 202):
raise HTTPException(status_code=500, detail=f"Qdrant upsert échoué: {r.text}")
return len(points)
async def qdrant_count(client: httpx.AsyncClient, coll: str) -> int:
url = f"{QDRANT_URL}/collections/{coll}/points/count"
r = await client.post(
url,
headers={"api-key": QDRANT_API_KEY},
json={"exact": True},
timeout=20,
)
if r.status_code != 200:
raise HTTPException(status_code=500, detail=f"Qdrant count échoué: {r.text}")
return int(r.json().get("result", {}).get("count", 0))
async def qdrant_search(
client: httpx.AsyncClient,
coll: str,
vector: List[float],
limit: int = 5,
) -> Dict[str, Any]:
url = f"{QDRANT_URL}/collections/{coll}/points/search"
r = await client.post(
url,
headers={"api-key": QDRANT_API_KEY},
json={"vector": vector, "limit": limit, "with_payload": True},
timeout=30,
)
if r.status_code != 200:
raise HTTPException(status_code=500, detail=f"Qdrant search échoué: {r.text}")
return r.json()
# ------------------------------------------------------------------------------
# Embeddings (HF Inference ou dummy)
# ------------------------------------------------------------------------------
async def embed_hf(
client: httpx.AsyncClient,
texts: List[str],
model: str = HF_EMBED_MODEL,
token: str = HF_TOKEN,
) -> List[List[float]]:
"""
Appel HuggingFace Inference (feature extraction) - batch.
Normalise L2 les vecteurs.
"""
if not token:
raise HTTPException(status_code=400, detail="HUGGINGFACEHUB_API_TOKEN manquant pour EMB_PROVIDER=hf")
url = f"https://api-inference.huggingface.co/models/{model}"
headers = {"Authorization": f"Bearer {token}"}
# HF accepte une liste de textes directement
payload = {"inputs": texts, "options": {"wait_for_model": True}}
r = await client.post(url, headers=headers, json=payload, timeout=120)
if r.status_code != 200:
raise HTTPException(status_code=502, detail=f"HF Inference error: {r.text}")
data = r.json()
# data peut être une liste de listes (ou de listes de listes...)
embeddings: List[List[float]] = []
if isinstance(data, list):
for row in data:
vec = flatten_any(row)
embeddings.append(l2_normalize(vec))
else:
vec = flatten_any(data)
embeddings.append(l2_normalize(vec))
return embeddings
def embed_dummy(texts: List[str], dim: int = 128) -> List[List[float]]:
"""
Embedding déterministe basé sur un hash -> vecteur pseudo-aléatoire stable.
Suffisant pour tester le pipeline Qdrant (dimensions cohérentes, upsert, count, search).
"""
out: List[List[float]] = []
for t in texts:
h = hashlib.sha256(t.encode("utf-8")).digest()
# Étale sur dim floats
arr = np.frombuffer((h * ((dim // len(h)) + 1))[:dim], dtype=np.uint8).astype(np.float32)
# Centrage et normalisation
arr = (arr - 127.5) / 127.5
arr = arr / (np.linalg.norm(arr) + 1e-9)
out.append(arr.astype(np.float32).tolist())
return out
async def embed_texts(client: httpx.AsyncClient, texts: List[str]) -> List[List[float]]:
if EMB_PROVIDER == "hf":
return await embed_hf(client, texts)
return embed_dummy(texts, dim=128)
# ------------------------------------------------------------------------------
# Pipeline d'indexation
# ------------------------------------------------------------------------------
async def run_index_job(job: JobState, req: IndexRequest) -> None:
job.stage = "embedding"
job.total_files = len(req.files)
job.log(f"Index start project={req.project_id} files={len(req.files)} chunk_size={req.chunk_size} overlap={req.overlap} batch_size={req.batch_size} store_text={req.store_text}")
# Dédup global par hash du texte brut des fichiers
file_hashes = [hash8(f.text) for f in req.files]
uniq = len(set(file_hashes))
if uniq != len(file_hashes):
job.log(f"Attention: {len(file_hashes)-uniq} fichiers ont un texte identique (hash dupliqué).")
# Chunking
records: List[Dict[str, Any]] = []
for f in req.files:
chunks = chunk_text(f.text, req.chunk_size, req.overlap)
if not chunks:
job.log(f"{f.path}: 0 chunk (trop court ou vide)")
for idx, (start, end, ch) in enumerate(chunks):
payload = {
"path": f.path,
"chunk": idx,
"start": start,
"end": end,
}
if req.store_text:
payload["text"] = ch
records.append({"payload": payload, "raw": ch})
job.total_chunks = len(records)
job.log(f"Total chunks = {job.total_chunks}")
if job.total_chunks == 0:
job.stage = "failed"
job.errors.append("Aucun chunk à indexer.")
job.finished_at = time.time()
return
# Embedding + Upsert (en batches)
async with httpx.AsyncClient(timeout=120) as client:
# Dimension à partir du 1er embedding (warmup)
warmup_vec = (await embed_texts(client, [records[0]["raw"]]))[0]
vec_dim = len(warmup_vec)
job.log(f"Warmup embeddings dim={vec_dim} provider={EMB_PROVIDER}")
# Qdrant collection
coll = f"{COLLECTION_PREFIX}{req.project_id}"
await ensure_collection(client, coll, vector_size=vec_dim)
job.stage = "upserting"
batch_vectors: List[List[float]] = []
batch_points: List[Dict[str, Any]] = []
async def flush_batch():
nonlocal batch_vectors, batch_points
if not batch_points:
return 0
added = await qdrant_upsert(client, coll, batch_points)
job.upserted += added
job.log(f"+{added} points upsert (total={job.upserted})")
batch_vectors = []
batch_points = []
return added
# Traite par lot d'embeddings (embedding_batch_size indépendant de l'upsert batch_size)
EMB_BATCH = max(8, min(64, req.batch_size * 2))
i = 0
while i < len(records):
sub = records[i : i + EMB_BATCH]
texts = [r["raw"] for r in sub]
vecs = await embed_texts(client, texts)
if len(vecs) != len(sub):
raise HTTPException(status_code=500, detail="Embedding batch size mismatch")
job.embedded += len(vecs)
for r, v in zip(sub, vecs):
payload = r["payload"]
point = {
"id": str(uuid.uuid4()),
"vector": v,
"payload": payload,
}
batch_points.append(point)
if len(batch_points) >= req.batch_size:
await flush_batch()
i += EMB_BATCH
# Flush final
await flush_batch()
job.stage = "done"
job.finished_at = time.time()
job.log("Index job terminé.")
# ------------------------------------------------------------------------------
# FastAPI app + endpoints
# ------------------------------------------------------------------------------
fastapi_app = FastAPI(title="Remote Indexer - Minimal Test Space")
fastapi_app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_methods=["*"],
allow_headers=["*"],
)
@fastapi_app.get("/")
async def root():
return {"ok": True, "service": "remote-indexer-min", "qdrant": bool(QDRANT_URL), "emb_provider": EMB_PROVIDER}
@fastapi_app.post("/wipe")
async def wipe(project_id: str = Query(..., min_length=1)):
if not QDRANT_URL or not QDRANT_API_KEY:
raise HTTPException(status_code=400, detail="QDRANT_URL / QDRANT_API_KEY requis")
coll = f"{COLLECTION_PREFIX}{project_id}"
async with httpx.AsyncClient() as client:
r = await client.delete(f"{QDRANT_URL}/collections/{coll}", headers={"api-key": QDRANT_API_KEY}, timeout=30)
if r.status_code not in (200, 202, 404):
raise HTTPException(status_code=500, detail=f"Echec wipe: {r.text}")
return {"ok": True, "collection": coll, "wiped": True}
@fastapi_app.post("/index")
async def index(req: IndexRequest):
if not QDRANT_URL or not QDRANT_API_KEY:
raise HTTPException(status_code=400, detail="QDRANT_URL / QDRANT_API_KEY requis")
job_id = uuid.uuid4().hex[:12]
job = JobState(job_id=job_id, project_id=req.project_id)
JOBS[job_id] = job
# Lance en tâche de fond
asyncio.create_task(run_index_job(job, req))
job.log(f"Job {job_id} créé pour project {req.project_id}")
return {"job_id": job_id, "project_id": req.project_id}
@fastapi_app.get("/status/{job_id}")
async def status(job_id: str):
job = JOBS.get(job_id)
if not job:
raise HTTPException(status_code=404, detail="job_id inconnu")
return job.model_dump()
@fastapi_app.get("/collections/{project_id}/count")
async def coll_count(project_id: str):
if not QDRANT_URL or not QDRANT_API_KEY:
raise HTTPException(status_code=400, detail="QDRANT_URL / QDRANT_API_KEY requis")
coll = f"{COLLECTION_PREFIX}{project_id}"
async with httpx.AsyncClient() as client:
cnt = await qdrant_count(client, coll)
return {"project_id": project_id, "collection": coll, "count": cnt}
@fastapi_app.post("/query")
async def query(req: QueryRequest):
if not QDRANT_URL or not QDRANT_API_KEY:
raise HTTPException(status_code=400, detail="QDRANT_URL / QDRANT_API_KEY requis")
coll = f"{COLLECTION_PREFIX}{req.project_id}"
async with httpx.AsyncClient() as client:
vec = (await embed_texts(client, [req.text]))[0]
data = await qdrant_search(client, coll, vec, limit=req.top_k)
return data
# ------------------------------------------------------------------------------
# Gradio UI (montée sur "/")
# ------------------------------------------------------------------------------
def _default_two_docs() -> List[Dict[str, str]]:
a = "Alpha bravo charlie delta echo foxtrot golf hotel india. " * 3
b = "Lorem ipsum dolor sit amet, consectetuer adipiscing elit, sed diam nonummy." * 3
return [
{"path": "a.txt", "text": a},
{"path": "b.txt", "text": b},
]
async def ui_wipe(project: str):
try:
resp = await wipe(project) # appelle la route interne
return f"✅ Wipe ok — collection {resp['collection']} supprimée."
except Exception as e:
LOG.exception("wipe UI error")
return f"❌ Wipe erreur: {e}"
async def ui_index_sample(project: str, chunk_size: int, overlap: int, batch_size: int, store_text: bool):
files = _default_two_docs()
req = IndexRequest(
project_id=project,
files=[FileItem(**f) for f in files],
chunk_size=chunk_size,
overlap=overlap,
batch_size=batch_size,
store_text=store_text,
)
try:
data = await index(req)
job_id = data["job_id"]
return f"🚀 Job lancé: {job_id}"
except ValidationError as ve:
return f"❌ Payload invalide: {ve}"
except Exception as e:
LOG.exception("index UI error")
return f"❌ Index erreur: {e}"
async def ui_status(job_id: str):
if not job_id.strip():
return "⚠️ Renseigne un job_id"
try:
st = await status(job_id)
# Formatage
lines = [f"Job {st['job_id']} — stage={st['stage']} files={st['total_files']} chunks={st['total_chunks']} embedded={st['embedded']} upserted={st['upserted']}"]
lines += st.get("messages", [])[-50:] # dernières lignes
if st.get("errors"):
lines.append("Erreurs:")
lines += [f" - {e}" for e in st["errors"]]
return "\n".join(lines)
except Exception as e:
return f"❌ Status erreur: {e}"
async def ui_count(project: str):
try:
resp = await coll_count(project)
return f"📊 Count — collection={resp['collection']} → {resp['count']} points"
except Exception as e:
LOG.exception("count UI error")
return f"❌ Count erreur: {e}"
async def ui_query(project: str, text: str, topk: int):
try:
data = await query(QueryRequest(project_id=project, text=text, top_k=topk))
hits = data.get("result", [])
if not hits:
return "Aucun résultat."
out = []
for h in hits:
score = h.get("score")
payload = h.get("payload", {})
path = payload.get("path")
chunk = payload.get("chunk")
preview = (payload.get("text") or "")[:120].replace("\n", " ")
out.append(f"{score:.4f} — {path} [chunk {chunk}] — {preview}…")
return "\n".join(out)
except Exception as e:
LOG.exception("query UI error")
return f"❌ Query erreur: {e}"
with gr.Blocks(title="Remote Indexer - Minimal Test", analytics_enabled=False) as ui:
gr.Markdown("## 🔬 Remote Indexer — Tests sans console\n"
"Wipe → Index 2 docs → Status → Count → Query\n"
f"- **Embeddings**: `{EMB_PROVIDER}` (model: `{HF_EMBED_MODEL}`)\n"
f"- **Qdrant**: `{'OK' if QDRANT_URL else 'ABSENT'}`\n"
"Conseil: si tu n'as pas de token HF, mets `EMB_PROVIDER=dummy` dans les Secrets du Space.")
with gr.Row():
project_tb = gr.Textbox(label="Project ID", value="DEEPWEB")
jobid_tb = gr.Textbox(label="Job ID (pour Status)", value="", interactive=True)
with gr.Row():
wipe_btn = gr.Button("🧨 Wipe collection", variant="stop")
index_btn = gr.Button("🚀 Indexer 2 documents", variant="primary")
count_btn = gr.Button("📊 Count points", variant="secondary")
with gr.Row():
chunk_size = gr.Slider(64, 1024, value=200, step=8, label="chunk_size")
overlap = gr.Slider(0, 256, value=20, step=2, label="overlap")
batch_size = gr.Slider(1, 128, value=32, step=1, label="batch_size")
store_text = gr.Checkbox(value=True, label="store_text (payload)")
out_log = gr.Textbox(lines=18, label="Logs / Résultats", interactive=False)
with gr.Row():
query_tb = gr.Textbox(label="Query text", value="alpha bravo")
topk = gr.Slider(1, 20, value=5, step=1, label="top_k")
query_btn = gr.Button("🔎 Query")
query_out = gr.Textbox(lines=10, label="Résultats Query", interactive=False)
wipe_btn.click(ui_wipe, inputs=[project_tb], outputs=[out_log])
index_btn.click(ui_index_sample, inputs=[project_tb, chunk_size, overlap, batch_size, store_text], outputs=[out_log])
# Petit auto-poll status: on relance ui_status à la main en collant le job_id
count_btn.click(ui_count, inputs=[project_tb], outputs=[out_log])
query_btn.click(ui_query, inputs=[project_tb, query_tb, topk], outputs=[query_out])
# Monte l'UI Gradio sur la FastAPI
app = gr.mount_gradio_app(fastapi_app, ui, path="/")
|