Update app.py
Browse files
app.py
CHANGED
@@ -1,39 +1,48 @@
|
|
1 |
from fastapi import FastAPI, Query, HTTPException
|
2 |
import torch
|
3 |
import re
|
|
|
|
|
4 |
from transformers import AutoTokenizer
|
5 |
from peft import AutoPeftModelForCausalLM
|
6 |
|
7 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
model_name = "habulaj/filter"
|
9 |
-
|
|
|
10 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
11 |
|
12 |
-
# Otimizações de performance
|
13 |
model = AutoPeftModelForCausalLM.from_pretrained(
|
14 |
model_name,
|
15 |
device_map="cpu",
|
16 |
-
torch_dtype=torch.float32,
|
17 |
-
low_cpu_mem_usage=True,
|
18 |
)
|
19 |
model.eval()
|
|
|
20 |
|
21 |
-
# Compilação do modelo para otimizar (PyTorch 2.0+)
|
22 |
try:
|
23 |
model = torch.compile(model, mode="reduce-overhead")
|
24 |
-
|
25 |
except Exception as e:
|
26 |
-
|
27 |
|
28 |
# -------- FASTAPI --------
|
29 |
app = FastAPI(title="News Filter JSON API")
|
30 |
|
31 |
-
# -------- ROOT ENDPOINT --------
|
32 |
@app.get("/")
|
33 |
def read_root():
|
34 |
return {"message": "News Filter JSON API is running!", "docs": "/docs"}
|
35 |
|
36 |
-
#
|
37 |
def infer_filter(title, content):
|
38 |
prompt = f"""Analyze the news title and content, and return the filters in JSON format with the defined fields.
|
39 |
|
@@ -43,44 +52,48 @@ Title: "{title}"
|
|
43 |
Content: "{content}"
|
44 |
"""
|
45 |
|
46 |
-
|
|
|
|
|
47 |
inputs = tokenizer(
|
48 |
-
prompt,
|
49 |
return_tensors="pt",
|
50 |
truncation=True,
|
51 |
-
max_length=512,
|
52 |
-
padding=False
|
53 |
)
|
54 |
input_ids = inputs.input_ids.to("cpu")
|
55 |
-
|
56 |
with torch.no_grad():
|
57 |
-
# Configurações otimizadas para velocidade
|
58 |
outputs = model.generate(
|
59 |
input_ids=input_ids,
|
60 |
-
max_new_tokens=100,
|
61 |
-
temperature=1.0,
|
62 |
do_sample=True,
|
63 |
top_p=0.9,
|
64 |
-
num_beams=1,
|
65 |
-
early_stopping=True,
|
66 |
eos_token_id=tokenizer.eos_token_id,
|
67 |
pad_token_id=tokenizer.eos_token_id,
|
68 |
)
|
69 |
-
|
70 |
decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
71 |
-
|
72 |
-
# Remove prompt do output
|
73 |
generated = decoded[len(prompt):].strip()
|
74 |
-
|
75 |
-
|
|
|
|
|
76 |
match = re.search(r"\{.*\}", generated, re.DOTALL)
|
77 |
if match:
|
78 |
-
|
79 |
-
|
|
|
|
|
80 |
else:
|
|
|
81 |
return "⚠️ Failed to extract JSON. Output:\n" + generated
|
82 |
|
83 |
-
# --------
|
84 |
@app.get("/filter")
|
85 |
def get_filter(
|
86 |
title: str = Query(..., description="Title of the news"),
|
@@ -89,10 +102,10 @@ def get_filter(
|
|
89 |
try:
|
90 |
json_output = infer_filter(title, content)
|
91 |
import json
|
92 |
-
# Retorna como dados brutos (parse do JSON)
|
93 |
return json.loads(json_output)
|
94 |
except json.JSONDecodeError:
|
95 |
-
|
96 |
return {"raw_output": json_output}
|
97 |
except Exception as e:
|
|
|
98 |
raise HTTPException(status_code=422, detail=str(e))
|
|
|
1 |
from fastapi import FastAPI, Query, HTTPException
|
2 |
import torch
|
3 |
import re
|
4 |
+
import time
|
5 |
+
import logging
|
6 |
from transformers import AutoTokenizer
|
7 |
from peft import AutoPeftModelForCausalLM
|
8 |
|
9 |
+
# -------- LOGGING CONFIG --------
|
10 |
+
logging.basicConfig(
|
11 |
+
level=logging.INFO,
|
12 |
+
format="%(asctime)s [%(levelname)s] %(message)s",
|
13 |
+
)
|
14 |
+
log = logging.getLogger("news-filter")
|
15 |
+
|
16 |
+
# -------- CARREGAMENTO DE MODELO --------
|
17 |
model_name = "habulaj/filter"
|
18 |
+
log.info("🚀 Iniciando carregamento do modelo e tokenizer...")
|
19 |
+
|
20 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
21 |
+
log.info("✅ Tokenizer carregado.")
|
22 |
|
|
|
23 |
model = AutoPeftModelForCausalLM.from_pretrained(
|
24 |
model_name,
|
25 |
device_map="cpu",
|
26 |
+
torch_dtype=torch.float32,
|
27 |
+
low_cpu_mem_usage=True,
|
28 |
)
|
29 |
model.eval()
|
30 |
+
log.info("✅ Modelo carregado e em modo eval.")
|
31 |
|
|
|
32 |
try:
|
33 |
model = torch.compile(model, mode="reduce-overhead")
|
34 |
+
log.info("✅ Modelo compilado com torch.compile.")
|
35 |
except Exception as e:
|
36 |
+
log.warning(f"⚠️ torch.compile indisponível: {e}")
|
37 |
|
38 |
# -------- FASTAPI --------
|
39 |
app = FastAPI(title="News Filter JSON API")
|
40 |
|
|
|
41 |
@app.get("/")
|
42 |
def read_root():
|
43 |
return {"message": "News Filter JSON API is running!", "docs": "/docs"}
|
44 |
|
45 |
+
# -------- INFERÊNCIA --------
|
46 |
def infer_filter(title, content):
|
47 |
prompt = f"""Analyze the news title and content, and return the filters in JSON format with the defined fields.
|
48 |
|
|
|
52 |
Content: "{content}"
|
53 |
"""
|
54 |
|
55 |
+
log.info(f"🧠 Iniciando inferência para notícia:\n📰 Title: {title}\n📝 Content: {content[:100]}...")
|
56 |
+
start_time = time.time()
|
57 |
+
|
58 |
inputs = tokenizer(
|
59 |
+
prompt,
|
60 |
return_tensors="pt",
|
61 |
truncation=True,
|
62 |
+
max_length=512,
|
63 |
+
padding=False,
|
64 |
)
|
65 |
input_ids = inputs.input_ids.to("cpu")
|
66 |
+
|
67 |
with torch.no_grad():
|
|
|
68 |
outputs = model.generate(
|
69 |
input_ids=input_ids,
|
70 |
+
max_new_tokens=100,
|
71 |
+
temperature=1.0,
|
72 |
do_sample=True,
|
73 |
top_p=0.9,
|
74 |
+
num_beams=1,
|
75 |
+
early_stopping=True,
|
76 |
eos_token_id=tokenizer.eos_token_id,
|
77 |
pad_token_id=tokenizer.eos_token_id,
|
78 |
)
|
79 |
+
|
80 |
decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
|
|
|
81 |
generated = decoded[len(prompt):].strip()
|
82 |
+
|
83 |
+
log.info("📤 Resposta bruta decodificada:")
|
84 |
+
log.info(generated)
|
85 |
+
|
86 |
match = re.search(r"\{.*\}", generated, re.DOTALL)
|
87 |
if match:
|
88 |
+
json_result = match.group(0)
|
89 |
+
duration = time.time() - start_time
|
90 |
+
log.info(f"✅ JSON extraído com sucesso em {duration:.2f}s")
|
91 |
+
return json_result
|
92 |
else:
|
93 |
+
log.warning("⚠️ Não foi possível extrair JSON.")
|
94 |
return "⚠️ Failed to extract JSON. Output:\n" + generated
|
95 |
|
96 |
+
# -------- ENDPOINT --------
|
97 |
@app.get("/filter")
|
98 |
def get_filter(
|
99 |
title: str = Query(..., description="Title of the news"),
|
|
|
102 |
try:
|
103 |
json_output = infer_filter(title, content)
|
104 |
import json
|
|
|
105 |
return json.loads(json_output)
|
106 |
except json.JSONDecodeError:
|
107 |
+
log.error("❌ Erro ao fazer parse do JSON retornado.")
|
108 |
return {"raw_output": json_output}
|
109 |
except Exception as e:
|
110 |
+
log.exception("❌ Erro inesperado durante a inferência:")
|
111 |
raise HTTPException(status_code=422, detail=str(e))
|