Jordi Catafal
commited on
Commit
·
0610fdd
1
Parent(s):
dda5c3b
fixing api problem
Browse files- app.py +22 -25
- app_old.py +159 -0
- test_api.py +1 -1
app.py
CHANGED
@@ -1,6 +1,5 @@
|
|
1 |
from fastapi import FastAPI, HTTPException
|
2 |
from fastapi.middleware.cors import CORSMiddleware
|
3 |
-
from contextlib import asynccontextmanager
|
4 |
from typing import List
|
5 |
import torch
|
6 |
import uvicorn
|
@@ -8,31 +7,10 @@ import uvicorn
|
|
8 |
from models.schemas import EmbeddingRequest, EmbeddingResponse, ModelInfo
|
9 |
from utils.helpers import load_models, get_embeddings, cleanup_memory
|
10 |
|
11 |
-
# Global model cache
|
12 |
-
models_cache = {}
|
13 |
-
|
14 |
-
@asynccontextmanager
|
15 |
-
async def lifespan(app: FastAPI):
|
16 |
-
"""Application lifespan handler for startup and shutdown"""
|
17 |
-
# Startup
|
18 |
-
try:
|
19 |
-
global models_cache
|
20 |
-
print("Loading models...")
|
21 |
-
models_cache = load_models()
|
22 |
-
print("All models loaded successfully!")
|
23 |
-
yield
|
24 |
-
except Exception as e:
|
25 |
-
print(f"Failed to load models: {str(e)}")
|
26 |
-
raise
|
27 |
-
finally:
|
28 |
-
# Shutdown - cleanup resources
|
29 |
-
cleanup_memory()
|
30 |
-
|
31 |
app = FastAPI(
|
32 |
title="Multilingual & Legal Embedding API",
|
33 |
description="Multi-model embedding API for Spanish, Catalan, English and Legal texts",
|
34 |
-
version="3.0.0"
|
35 |
-
lifespan=lifespan
|
36 |
)
|
37 |
|
38 |
# Add CORS middleware to allow cross-origin requests
|
@@ -44,6 +22,21 @@ app.add_middleware(
|
|
44 |
allow_headers=["*"],
|
45 |
)
|
46 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
@app.get("/")
|
48 |
async def root():
|
49 |
return {
|
@@ -58,6 +51,9 @@ async def root():
|
|
58 |
async def create_embeddings(request: EmbeddingRequest):
|
59 |
"""Generate embeddings for input texts"""
|
60 |
try:
|
|
|
|
|
|
|
61 |
if not request.texts:
|
62 |
raise HTTPException(status_code=400, detail="No texts provided")
|
63 |
|
@@ -144,11 +140,12 @@ async def health_check():
|
|
144 |
"""Health check endpoint"""
|
145 |
models_loaded = len(models_cache) == 5
|
146 |
return {
|
147 |
-
"status": "healthy" if models_loaded else "
|
148 |
"models_loaded": models_loaded,
|
149 |
"available_models": list(models_cache.keys()),
|
150 |
"expected_models": ["jina", "robertalex", "jina-v3", "legal-bert", "roberta-ca"],
|
151 |
-
"models_count": len(models_cache)
|
|
|
152 |
}
|
153 |
|
154 |
if __name__ == "__main__":
|
|
|
1 |
from fastapi import FastAPI, HTTPException
|
2 |
from fastapi.middleware.cors import CORSMiddleware
|
|
|
3 |
from typing import List
|
4 |
import torch
|
5 |
import uvicorn
|
|
|
7 |
from models.schemas import EmbeddingRequest, EmbeddingResponse, ModelInfo
|
8 |
from utils.helpers import load_models, get_embeddings, cleanup_memory
|
9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
app = FastAPI(
|
11 |
title="Multilingual & Legal Embedding API",
|
12 |
description="Multi-model embedding API for Spanish, Catalan, English and Legal texts",
|
13 |
+
version="3.0.0"
|
|
|
14 |
)
|
15 |
|
16 |
# Add CORS middleware to allow cross-origin requests
|
|
|
22 |
allow_headers=["*"],
|
23 |
)
|
24 |
|
25 |
+
# Global model cache - loaded on demand
|
26 |
+
models_cache = {}
|
27 |
+
|
28 |
+
def ensure_models_loaded():
|
29 |
+
"""Load models on first request if not already loaded"""
|
30 |
+
global models_cache
|
31 |
+
if not models_cache:
|
32 |
+
try:
|
33 |
+
print("Loading models on demand...")
|
34 |
+
models_cache = load_models()
|
35 |
+
print("All models loaded successfully!")
|
36 |
+
except Exception as e:
|
37 |
+
print(f"Failed to load models: {str(e)}")
|
38 |
+
raise HTTPException(status_code=500, detail=f"Model loading failed: {str(e)}")
|
39 |
+
|
40 |
@app.get("/")
|
41 |
async def root():
|
42 |
return {
|
|
|
51 |
async def create_embeddings(request: EmbeddingRequest):
|
52 |
"""Generate embeddings for input texts"""
|
53 |
try:
|
54 |
+
# Load models on first request
|
55 |
+
ensure_models_loaded()
|
56 |
+
|
57 |
if not request.texts:
|
58 |
raise HTTPException(status_code=400, detail="No texts provided")
|
59 |
|
|
|
140 |
"""Health check endpoint"""
|
141 |
models_loaded = len(models_cache) == 5
|
142 |
return {
|
143 |
+
"status": "healthy" if models_loaded else "ready",
|
144 |
"models_loaded": models_loaded,
|
145 |
"available_models": list(models_cache.keys()),
|
146 |
"expected_models": ["jina", "robertalex", "jina-v3", "legal-bert", "roberta-ca"],
|
147 |
+
"models_count": len(models_cache),
|
148 |
+
"note": "Models load on first embedding request" if not models_loaded else "All models ready"
|
149 |
}
|
150 |
|
151 |
if __name__ == "__main__":
|
app_old.py
ADDED
@@ -0,0 +1,159 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from fastapi import FastAPI, HTTPException
|
2 |
+
from fastapi.middleware.cors import CORSMiddleware
|
3 |
+
from contextlib import asynccontextmanager
|
4 |
+
from typing import List
|
5 |
+
import torch
|
6 |
+
import uvicorn
|
7 |
+
|
8 |
+
from models.schemas import EmbeddingRequest, EmbeddingResponse, ModelInfo
|
9 |
+
from utils.helpers import load_models, get_embeddings, cleanup_memory
|
10 |
+
|
11 |
+
# Global model cache
|
12 |
+
models_cache = {}
|
13 |
+
|
14 |
+
@asynccontextmanager
|
15 |
+
async def lifespan(app: FastAPI):
|
16 |
+
"""Application lifespan handler for startup and shutdown"""
|
17 |
+
# Startup
|
18 |
+
try:
|
19 |
+
global models_cache
|
20 |
+
print("Loading models...")
|
21 |
+
models_cache = load_models()
|
22 |
+
print("All models loaded successfully!")
|
23 |
+
yield
|
24 |
+
except Exception as e:
|
25 |
+
print(f"Failed to load models: {str(e)}")
|
26 |
+
raise
|
27 |
+
finally:
|
28 |
+
# Shutdown - cleanup resources
|
29 |
+
cleanup_memory()
|
30 |
+
|
31 |
+
app = FastAPI(
|
32 |
+
title="Multilingual & Legal Embedding API",
|
33 |
+
description="Multi-model embedding API for Spanish, Catalan, English and Legal texts",
|
34 |
+
version="3.0.0",
|
35 |
+
lifespan=lifespan
|
36 |
+
)
|
37 |
+
|
38 |
+
# Add CORS middleware to allow cross-origin requests
|
39 |
+
app.add_middleware(
|
40 |
+
CORSMiddleware,
|
41 |
+
allow_origins=["*"], # In production, specify actual domains
|
42 |
+
allow_credentials=True,
|
43 |
+
allow_methods=["*"],
|
44 |
+
allow_headers=["*"],
|
45 |
+
)
|
46 |
+
|
47 |
+
@app.get("/")
|
48 |
+
async def root():
|
49 |
+
return {
|
50 |
+
"message": "Multilingual & Legal Embedding API",
|
51 |
+
"models": ["jina", "robertalex", "jina-v3", "legal-bert", "roberta-ca"],
|
52 |
+
"status": "running",
|
53 |
+
"docs": "/docs",
|
54 |
+
"total_models": 5
|
55 |
+
}
|
56 |
+
|
57 |
+
@app.post("/embed", response_model=EmbeddingResponse)
|
58 |
+
async def create_embeddings(request: EmbeddingRequest):
|
59 |
+
"""Generate embeddings for input texts"""
|
60 |
+
try:
|
61 |
+
if not request.texts:
|
62 |
+
raise HTTPException(status_code=400, detail="No texts provided")
|
63 |
+
|
64 |
+
if len(request.texts) > 50: # Rate limiting
|
65 |
+
raise HTTPException(status_code=400, detail="Maximum 50 texts per request")
|
66 |
+
|
67 |
+
embeddings = get_embeddings(
|
68 |
+
request.texts,
|
69 |
+
request.model,
|
70 |
+
models_cache,
|
71 |
+
request.normalize,
|
72 |
+
request.max_length
|
73 |
+
)
|
74 |
+
|
75 |
+
# Cleanup memory after large batches
|
76 |
+
if len(request.texts) > 20:
|
77 |
+
cleanup_memory()
|
78 |
+
|
79 |
+
return EmbeddingResponse(
|
80 |
+
embeddings=embeddings,
|
81 |
+
model_used=request.model,
|
82 |
+
dimensions=len(embeddings[0]) if embeddings else 0,
|
83 |
+
num_texts=len(request.texts)
|
84 |
+
)
|
85 |
+
|
86 |
+
except ValueError as e:
|
87 |
+
raise HTTPException(status_code=400, detail=str(e))
|
88 |
+
except Exception as e:
|
89 |
+
raise HTTPException(status_code=500, detail=f"Internal error: {str(e)}")
|
90 |
+
|
91 |
+
@app.get("/models", response_model=List[ModelInfo])
|
92 |
+
async def list_models():
|
93 |
+
"""List available models and their specifications"""
|
94 |
+
return [
|
95 |
+
ModelInfo(
|
96 |
+
model_id="jina",
|
97 |
+
name="jinaai/jina-embeddings-v2-base-es",
|
98 |
+
dimensions=768,
|
99 |
+
max_sequence_length=8192,
|
100 |
+
languages=["Spanish", "English"],
|
101 |
+
model_type="bilingual",
|
102 |
+
description="Bilingual Spanish-English embeddings with long context support"
|
103 |
+
),
|
104 |
+
ModelInfo(
|
105 |
+
model_id="robertalex",
|
106 |
+
name="PlanTL-GOB-ES/RoBERTalex",
|
107 |
+
dimensions=768,
|
108 |
+
max_sequence_length=512,
|
109 |
+
languages=["Spanish"],
|
110 |
+
model_type="legal domain",
|
111 |
+
description="Spanish legal domain specialized embeddings"
|
112 |
+
),
|
113 |
+
ModelInfo(
|
114 |
+
model_id="jina-v3",
|
115 |
+
name="jinaai/jina-embeddings-v3",
|
116 |
+
dimensions=1024,
|
117 |
+
max_sequence_length=8192,
|
118 |
+
languages=["Multilingual"],
|
119 |
+
model_type="multilingual",
|
120 |
+
description="Latest Jina v3 with superior multilingual performance"
|
121 |
+
),
|
122 |
+
ModelInfo(
|
123 |
+
model_id="legal-bert",
|
124 |
+
name="nlpaueb/legal-bert-base-uncased",
|
125 |
+
dimensions=768,
|
126 |
+
max_sequence_length=512,
|
127 |
+
languages=["English"],
|
128 |
+
model_type="legal domain",
|
129 |
+
description="English legal domain BERT model"
|
130 |
+
),
|
131 |
+
ModelInfo(
|
132 |
+
model_id="roberta-ca",
|
133 |
+
name="projecte-aina/roberta-large-ca-v2",
|
134 |
+
dimensions=1024,
|
135 |
+
max_sequence_length=512,
|
136 |
+
languages=["Catalan"],
|
137 |
+
model_type="general",
|
138 |
+
description="Catalan RoBERTa-large model trained on large corpus"
|
139 |
+
)
|
140 |
+
]
|
141 |
+
|
142 |
+
@app.get("/health")
|
143 |
+
async def health_check():
|
144 |
+
"""Health check endpoint"""
|
145 |
+
models_loaded = len(models_cache) == 5
|
146 |
+
return {
|
147 |
+
"status": "healthy" if models_loaded else "degraded",
|
148 |
+
"models_loaded": models_loaded,
|
149 |
+
"available_models": list(models_cache.keys()),
|
150 |
+
"expected_models": ["jina", "robertalex", "jina-v3", "legal-bert", "roberta-ca"],
|
151 |
+
"models_count": len(models_cache)
|
152 |
+
}
|
153 |
+
|
154 |
+
if __name__ == "__main__":
|
155 |
+
# Set multi-threading for CPU
|
156 |
+
torch.set_num_threads(8)
|
157 |
+
torch.set_num_interop_threads(1)
|
158 |
+
|
159 |
+
uvicorn.run(app, host="0.0.0.0", port=7860)
|
test_api.py
CHANGED
@@ -7,7 +7,7 @@ import requests
|
|
7 |
import json
|
8 |
import time
|
9 |
|
10 |
-
def test_api(base_url="
|
11 |
"""Test the API endpoints"""
|
12 |
|
13 |
print(f"Testing API at {base_url}")
|
|
|
7 |
import json
|
8 |
import time
|
9 |
|
10 |
+
def test_api(base_url="https://aurasystems-spanish-embeddings-api.hf.space"):
|
11 |
"""Test the API endpoints"""
|
12 |
|
13 |
print(f"Testing API at {base_url}")
|