Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Commit
·
b5c8d5a
1
Parent(s):
7cf16e2
chroma and models
Browse files
main.py
CHANGED
@@ -10,26 +10,14 @@ from fastapi.middleware.cors import CORSMiddleware
|
|
10 |
from pydantic import BaseModel
|
11 |
from contextlib import asynccontextmanager
|
12 |
import polars as pl
|
13 |
-
from huggingface_hub import
|
14 |
-
from datetime import datetime, timedelta
|
15 |
-
from typing import Generator
|
16 |
-
from huggingface_hub import ModelInfo, DatasetInfo
|
17 |
-
import stamina
|
18 |
-
import logging
|
19 |
-
import polars as pl
|
20 |
-
from huggingface_hub import dataset_info
|
21 |
-
from huggingface_hub import InferenceClient
|
22 |
from transformers import AutoTokenizer
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
from huggingface_hub import get_inference_endpoint
|
30 |
-
from huggingface_hub import AsyncInferenceClient
|
31 |
-
import asyncio
|
32 |
-
from typing import List
|
33 |
|
34 |
hf_api = HfApi()
|
35 |
|
@@ -74,7 +62,7 @@ app.add_middleware(
|
|
74 |
allow_origins=[
|
75 |
"https://*.hf.space", # Allow all Hugging Face Spaces
|
76 |
"https://*.huggingface.co", # Allow all Hugging Face domains
|
77 |
-
|
78 |
],
|
79 |
allow_credentials=True,
|
80 |
allow_methods=["*"],
|
@@ -93,12 +81,20 @@ def setup_database():
|
|
93 |
try:
|
94 |
embedding_function = get_embedding_function()
|
95 |
|
96 |
-
# Create collection
|
97 |
dataset_collection = client.get_or_create_collection(
|
98 |
embedding_function=embedding_function,
|
99 |
name="dataset_cards",
|
100 |
metadata={"hnsw:space": "cosine"},
|
101 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
102 |
# TODO incremental updates
|
103 |
df = pl.scan_parquet(
|
104 |
"hf://datasets/davanstrien/datasets_with_metadata_and_summaries/data/train-*.parquet"
|
@@ -139,42 +135,48 @@ def setup_database():
|
|
139 |
logger.info(f"Processed {i + len(batch_df):,} / {total_rows:,} rows")
|
140 |
|
141 |
logger.info(f"Database initialized with {dataset_collection.count():,} rows")
|
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 |
except Exception as e:
|
180 |
logger.error(f"Setup error: {e}")
|
@@ -196,6 +198,18 @@ class QueryResponse(BaseModel):
|
|
196 |
results: List[QueryResult]
|
197 |
|
198 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
199 |
@app.get("/")
|
200 |
async def redirect_to_docs():
|
201 |
from fastapi.responses import RedirectResponse
|
@@ -204,7 +218,7 @@ async def redirect_to_docs():
|
|
204 |
|
205 |
|
206 |
@app.get("/search/datasets", response_model=QueryResponse)
|
207 |
-
@cache(ttl=
|
208 |
async def search_datasets(
|
209 |
query: str,
|
210 |
k: int = Query(default=5, ge=1, le=100),
|
@@ -235,22 +249,7 @@ async def search_datasets(
|
|
235 |
)
|
236 |
|
237 |
# Process results
|
238 |
-
query_results =
|
239 |
-
for i in range(len(results["ids"][0])):
|
240 |
-
query_results.append(
|
241 |
-
QueryResult(
|
242 |
-
dataset_id=results["ids"][0][i],
|
243 |
-
similarity=float(results["distances"][0][i]),
|
244 |
-
summary=results["documents"][0][i],
|
245 |
-
likes=results["metadatas"][0][i]["likes"],
|
246 |
-
downloads=results["metadatas"][0][i]["downloads"],
|
247 |
-
)
|
248 |
-
)
|
249 |
-
|
250 |
-
# Sort results if needed
|
251 |
-
if sort_by != "similarity":
|
252 |
-
query_results.sort(key=lambda x: getattr(x, sort_by), reverse=True)
|
253 |
-
query_results = query_results[:k]
|
254 |
|
255 |
return QueryResponse(results=query_results)
|
256 |
|
@@ -260,7 +259,7 @@ async def search_datasets(
|
|
260 |
|
261 |
|
262 |
@app.get("/similarity/datasets", response_model=QueryResponse)
|
263 |
-
@cache(ttl=
|
264 |
async def find_similar_datasets(
|
265 |
dataset_id: str,
|
266 |
k: int = Query(default=5, ge=1, le=100),
|
@@ -298,25 +297,9 @@ async def find_similar_datasets(
|
|
298 |
)
|
299 |
|
300 |
# Process results (excluding the query dataset itself)
|
301 |
-
query_results =
|
302 |
-
|
303 |
-
|
304 |
-
query_results.append(
|
305 |
-
QueryResult(
|
306 |
-
dataset_id=results["ids"][0][i],
|
307 |
-
similarity=float(results["distances"][0][i]),
|
308 |
-
summary=results["documents"][0][i],
|
309 |
-
likes=results["metadatas"][0][i]["likes"],
|
310 |
-
downloads=results["metadatas"][0][i]["downloads"],
|
311 |
-
)
|
312 |
-
)
|
313 |
-
|
314 |
-
# Sort results if needed
|
315 |
-
if sort_by != "similarity":
|
316 |
-
query_results.sort(key=lambda x: getattr(x, sort_by), reverse=True)
|
317 |
-
query_results = query_results[:k]
|
318 |
-
else:
|
319 |
-
query_results = query_results[:k]
|
320 |
|
321 |
return QueryResponse(results=query_results)
|
322 |
|
@@ -327,6 +310,119 @@ async def find_similar_datasets(
|
|
327 |
raise HTTPException(status_code=500, detail="Similarity search failed")
|
328 |
|
329 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
330 |
if __name__ == "__main__":
|
331 |
import uvicorn
|
332 |
|
|
|
10 |
from pydantic import BaseModel
|
11 |
from contextlib import asynccontextmanager
|
12 |
import polars as pl
|
13 |
+
from huggingface_hub import HfApi
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
from transformers import AutoTokenizer
|
15 |
+
|
16 |
+
# Configuration constants
|
17 |
+
MODEL_NAME = "davanstrien/SmolLM2-360M-tldr-sft-2025-02-12_15-13"
|
18 |
+
EMBEDDING_MODEL = "nomic-ai/modernbert-embed-base"
|
19 |
+
BATCH_SIZE = 1000
|
20 |
+
CACHE_TTL = "30"
|
|
|
|
|
|
|
|
|
21 |
|
22 |
hf_api = HfApi()
|
23 |
|
|
|
62 |
allow_origins=[
|
63 |
"https://*.hf.space", # Allow all Hugging Face Spaces
|
64 |
"https://*.huggingface.co", # Allow all Hugging Face domains
|
65 |
+
"http://localhost:5500", # Allow localhost:5500 # TODO remove before prod
|
66 |
],
|
67 |
allow_credentials=True,
|
68 |
allow_methods=["*"],
|
|
|
81 |
try:
|
82 |
embedding_function = get_embedding_function()
|
83 |
|
84 |
+
# Create dataset collection
|
85 |
dataset_collection = client.get_or_create_collection(
|
86 |
embedding_function=embedding_function,
|
87 |
name="dataset_cards",
|
88 |
metadata={"hnsw:space": "cosine"},
|
89 |
)
|
90 |
+
|
91 |
+
# Create model collection
|
92 |
+
model_collection = client.get_or_create_collection(
|
93 |
+
embedding_function=embedding_function,
|
94 |
+
name="model_cards",
|
95 |
+
metadata={"hnsw:space": "cosine"},
|
96 |
+
)
|
97 |
+
|
98 |
# TODO incremental updates
|
99 |
df = pl.scan_parquet(
|
100 |
"hf://datasets/davanstrien/datasets_with_metadata_and_summaries/data/train-*.parquet"
|
|
|
135 |
logger.info(f"Processed {i + len(batch_df):,} / {total_rows:,} rows")
|
136 |
|
137 |
logger.info(f"Database initialized with {dataset_collection.count():,} rows")
|
138 |
+
|
139 |
+
# Load model data
|
140 |
+
model_df = pl.scan_parquet(
|
141 |
+
"hf://datasets/davanstrien/models_with_metadata_and_summaries/data/train-*.parquet"
|
142 |
+
)
|
143 |
+
model_row_count = model_df.select(pl.len()).collect().item()
|
144 |
+
logger.info(f"Row count of new model data: {model_row_count}")
|
145 |
+
|
146 |
+
if model_collection.count() < model_row_count:
|
147 |
+
model_df = model_df.select(
|
148 |
+
["modelId", "summary", "likes", "downloads", "last_modified"]
|
149 |
+
)
|
150 |
+
model_df = model_df.collect()
|
151 |
+
BATCH_SIZE = 1000
|
152 |
+
total_rows = len(model_df)
|
153 |
+
|
154 |
+
for i in range(0, total_rows, BATCH_SIZE):
|
155 |
+
batch_df = model_df.slice(i, min(BATCH_SIZE, total_rows - i))
|
156 |
+
|
157 |
+
model_collection.upsert(
|
158 |
+
ids=batch_df.select(["modelId"]).to_series().to_list(),
|
159 |
+
documents=batch_df.select(["summary"]).to_series().to_list(),
|
160 |
+
metadatas=[
|
161 |
+
{
|
162 |
+
"likes": int(likes),
|
163 |
+
"downloads": int(downloads),
|
164 |
+
"last_modified": str(last_modified),
|
165 |
+
}
|
166 |
+
for likes, downloads, last_modified in zip(
|
167 |
+
batch_df.select(["likes"]).to_series().to_list(),
|
168 |
+
batch_df.select(["downloads"]).to_series().to_list(),
|
169 |
+
batch_df.select(["last_modified"]).to_series().to_list(),
|
170 |
+
)
|
171 |
+
],
|
172 |
+
)
|
173 |
+
logger.info(
|
174 |
+
f"Processed {i + len(batch_df):,} / {total_rows:,} model rows"
|
175 |
+
)
|
176 |
+
|
177 |
+
logger.info(
|
178 |
+
f"Model database initialized with {model_collection.count():,} rows"
|
179 |
+
)
|
180 |
|
181 |
except Exception as e:
|
182 |
logger.error(f"Setup error: {e}")
|
|
|
198 |
results: List[QueryResult]
|
199 |
|
200 |
|
201 |
+
class ModelQueryResult(BaseModel):
|
202 |
+
model_id: str
|
203 |
+
similarity: float
|
204 |
+
summary: str
|
205 |
+
likes: int
|
206 |
+
downloads: int
|
207 |
+
|
208 |
+
|
209 |
+
class ModelQueryResponse(BaseModel):
|
210 |
+
results: List[ModelQueryResult]
|
211 |
+
|
212 |
+
|
213 |
@app.get("/")
|
214 |
async def redirect_to_docs():
|
215 |
from fastapi.responses import RedirectResponse
|
|
|
218 |
|
219 |
|
220 |
@app.get("/search/datasets", response_model=QueryResponse)
|
221 |
+
@cache(ttl=CACHE_TTL)
|
222 |
async def search_datasets(
|
223 |
query: str,
|
224 |
k: int = Query(default=5, ge=1, le=100),
|
|
|
249 |
)
|
250 |
|
251 |
# Process results
|
252 |
+
query_results = process_search_results(results, "dataset", k, sort_by)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
253 |
|
254 |
return QueryResponse(results=query_results)
|
255 |
|
|
|
259 |
|
260 |
|
261 |
@app.get("/similarity/datasets", response_model=QueryResponse)
|
262 |
+
@cache(ttl=CACHE_TTL)
|
263 |
async def find_similar_datasets(
|
264 |
dataset_id: str,
|
265 |
k: int = Query(default=5, ge=1, le=100),
|
|
|
297 |
)
|
298 |
|
299 |
# Process results (excluding the query dataset itself)
|
300 |
+
query_results = process_search_results(
|
301 |
+
results, "dataset", k, sort_by, dataset_id
|
302 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
303 |
|
304 |
return QueryResponse(results=query_results)
|
305 |
|
|
|
310 |
raise HTTPException(status_code=500, detail="Similarity search failed")
|
311 |
|
312 |
|
313 |
+
@app.get("/search/models", response_model=ModelQueryResponse)
|
314 |
+
@cache(ttl=CACHE_TTL)
|
315 |
+
async def search_models(
|
316 |
+
query: str,
|
317 |
+
k: int = Query(default=5, ge=1, le=100),
|
318 |
+
sort_by: str = Query(
|
319 |
+
default="similarity", enum=["similarity", "likes", "downloads"]
|
320 |
+
),
|
321 |
+
min_likes: int = Query(default=0, ge=0),
|
322 |
+
min_downloads: int = Query(default=0, ge=0),
|
323 |
+
):
|
324 |
+
try:
|
325 |
+
collection = client.get_collection(
|
326 |
+
name="model_cards", embedding_function=get_embedding_function()
|
327 |
+
)
|
328 |
+
|
329 |
+
results = collection.query(
|
330 |
+
query_texts=[f"search_query: {query}"],
|
331 |
+
n_results=k * 4 if sort_by != "similarity" else k,
|
332 |
+
where={
|
333 |
+
"$and": [
|
334 |
+
{"likes": {"$gte": min_likes}},
|
335 |
+
{"downloads": {"$gte": min_downloads}},
|
336 |
+
]
|
337 |
+
}
|
338 |
+
if min_likes > 0 or min_downloads > 0
|
339 |
+
else None,
|
340 |
+
)
|
341 |
+
|
342 |
+
query_results = process_search_results(results, "model", k, sort_by)
|
343 |
+
|
344 |
+
return ModelQueryResponse(results=query_results)
|
345 |
+
|
346 |
+
except Exception as e:
|
347 |
+
logger.error(f"Model search error: {str(e)}")
|
348 |
+
raise HTTPException(status_code=500, detail="Model search failed")
|
349 |
+
|
350 |
+
|
351 |
+
@app.get("/similarity/models", response_model=ModelQueryResponse)
|
352 |
+
@cache(ttl=CACHE_TTL)
|
353 |
+
async def find_similar_models(
|
354 |
+
model_id: str,
|
355 |
+
k: int = Query(default=5, ge=1, le=100),
|
356 |
+
sort_by: str = Query(
|
357 |
+
default="similarity", enum=["similarity", "likes", "downloads"]
|
358 |
+
),
|
359 |
+
min_likes: int = Query(default=0, ge=0),
|
360 |
+
min_downloads: int = Query(default=0, ge=0),
|
361 |
+
):
|
362 |
+
try:
|
363 |
+
collection = client.get_collection("model_cards")
|
364 |
+
|
365 |
+
results = collection.get(ids=[model_id], include=["embeddings"])
|
366 |
+
|
367 |
+
if not results["ids"]:
|
368 |
+
raise HTTPException(
|
369 |
+
status_code=404, detail=f"Model ID '{model_id}' not found"
|
370 |
+
)
|
371 |
+
|
372 |
+
results = collection.query(
|
373 |
+
query_embeddings=[results["embeddings"][0]],
|
374 |
+
n_results=k * 4 if sort_by != "similarity" else k + 1,
|
375 |
+
where={
|
376 |
+
"$and": [
|
377 |
+
{"likes": {"$gte": min_likes}},
|
378 |
+
{"downloads": {"$gte": min_downloads}},
|
379 |
+
]
|
380 |
+
}
|
381 |
+
if min_likes > 0 or min_downloads > 0
|
382 |
+
else None,
|
383 |
+
)
|
384 |
+
|
385 |
+
query_results = process_search_results(results, "model", k, sort_by, model_id)
|
386 |
+
|
387 |
+
return ModelQueryResponse(results=query_results)
|
388 |
+
|
389 |
+
except HTTPException:
|
390 |
+
raise
|
391 |
+
except Exception as e:
|
392 |
+
logger.error(f"Model similarity search error: {str(e)}")
|
393 |
+
raise HTTPException(status_code=500, detail="Model similarity search failed")
|
394 |
+
|
395 |
+
|
396 |
+
def process_search_results(results, id_field, k, sort_by, exclude_id=None):
|
397 |
+
"""Process search results into a standardized format."""
|
398 |
+
query_results = []
|
399 |
+
for i in range(len(results["ids"][0])):
|
400 |
+
current_id = results["ids"][0][i]
|
401 |
+
if exclude_id and current_id == exclude_id:
|
402 |
+
continue
|
403 |
+
|
404 |
+
result = {
|
405 |
+
f"{id_field}_id": current_id,
|
406 |
+
"similarity": float(results["distances"][0][i]),
|
407 |
+
"summary": results["documents"][0][i],
|
408 |
+
"likes": results["metadatas"][0][i]["likes"],
|
409 |
+
"downloads": results["metadatas"][0][i]["downloads"],
|
410 |
+
}
|
411 |
+
|
412 |
+
if id_field == "dataset":
|
413 |
+
query_results.append(QueryResult(**result))
|
414 |
+
else:
|
415 |
+
query_results.append(ModelQueryResult(**result))
|
416 |
+
|
417 |
+
if sort_by != "similarity":
|
418 |
+
query_results.sort(key=lambda x: getattr(x, sort_by), reverse=True)
|
419 |
+
query_results = query_results[:k]
|
420 |
+
elif exclude_id: # We fetched extra for similarity + exclude_id case
|
421 |
+
query_results = query_results[:k]
|
422 |
+
|
423 |
+
return query_results
|
424 |
+
|
425 |
+
|
426 |
if __name__ == "__main__":
|
427 |
import uvicorn
|
428 |
|