Update app.py
Browse files
app.py
CHANGED
@@ -1,16 +1,21 @@
|
|
1 |
-
import
|
|
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import torch
|
4 |
import os
|
|
|
|
|
5 |
|
6 |
-
#
|
7 |
os.environ["TRANSFORMERS_VERBOSITY"] = "error"
|
|
|
8 |
|
9 |
-
#
|
10 |
MODEL_NAME = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
|
11 |
|
12 |
-
print("Carregando TinyLlama
|
13 |
|
|
|
14 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
15 |
model = AutoModelForCausalLM.from_pretrained(
|
16 |
MODEL_NAME,
|
@@ -19,95 +24,142 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
19 |
low_cpu_mem_usage=True
|
20 |
)
|
21 |
|
22 |
-
# Configurar pad token
|
23 |
if tokenizer.pad_token is None:
|
24 |
tokenizer.pad_token = tokenizer.eos_token
|
25 |
|
26 |
-
print("✅ Modelo carregado!
|
27 |
|
28 |
-
|
29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
try:
|
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 |
-
new_tokens = outputs[0][len(inputs.input_ids[0]):]
|
59 |
-
response = tokenizer.decode(new_tokens, skip_special_tokens=True)
|
60 |
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
|
66 |
-
return response
|
67 |
|
68 |
except Exception as e:
|
69 |
-
|
70 |
-
|
71 |
-
# Interface Gradio simples e funcional
|
72 |
-
interface = gr.Interface(
|
73 |
-
fn=chat_response,
|
74 |
-
inputs=[
|
75 |
-
gr.Textbox(
|
76 |
-
label="💬 Sua pergunta",
|
77 |
-
placeholder="Digite sua pergunta aqui...",
|
78 |
-
lines=2
|
79 |
-
),
|
80 |
-
gr.Slider(
|
81 |
-
minimum=50,
|
82 |
-
maximum=400,
|
83 |
-
value=200,
|
84 |
-
step=10,
|
85 |
-
label="🔢 Tokens máximos"
|
86 |
-
),
|
87 |
-
gr.Slider(
|
88 |
-
minimum=0.1,
|
89 |
-
maximum=1.2,
|
90 |
-
value=0.7,
|
91 |
-
step=0.1,
|
92 |
-
label="🌡️ Criatividade"
|
93 |
-
)
|
94 |
-
],
|
95 |
-
outputs=gr.Textbox(
|
96 |
-
label="🤖 Resposta do TinyLlama",
|
97 |
-
lines=5
|
98 |
-
),
|
99 |
-
title="🦙 TinyLlama Chat API",
|
100 |
-
description="Modelo de IA leve (2.2GB) otimizado para Hugging Face Spaces gratuito",
|
101 |
-
theme="default",
|
102 |
-
# Sem examples para evitar cache/erros
|
103 |
-
allow_flagging="never"
|
104 |
-
)
|
105 |
|
106 |
if __name__ == "__main__":
|
107 |
-
print("🚀 Iniciando servidor...")
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
)
|
|
|
1 |
+
from fastapi import FastAPI, HTTPException
|
2 |
+
from pydantic import BaseModel
|
3 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
4 |
import torch
|
5 |
import os
|
6 |
+
import uvicorn
|
7 |
+
import threading
|
8 |
|
9 |
+
# Configurações
|
10 |
os.environ["TRANSFORMERS_VERBOSITY"] = "error"
|
11 |
+
os.environ["TOKENIZERS_PARALLELISM"] = "false"
|
12 |
|
13 |
+
# Modelo
|
14 |
MODEL_NAME = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
|
15 |
|
16 |
+
print("🦙 Carregando TinyLlama para API...")
|
17 |
|
18 |
+
# Carregar modelo
|
19 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
20 |
model = AutoModelForCausalLM.from_pretrained(
|
21 |
MODEL_NAME,
|
|
|
24 |
low_cpu_mem_usage=True
|
25 |
)
|
26 |
|
|
|
27 |
if tokenizer.pad_token is None:
|
28 |
tokenizer.pad_token = tokenizer.eos_token
|
29 |
|
30 |
+
print("✅ Modelo carregado! API iniciando...")
|
31 |
|
32 |
+
# FastAPI app
|
33 |
+
app = FastAPI(
|
34 |
+
title="TinyLlama Chat API",
|
35 |
+
description="API REST para TinyLlama 1.1B",
|
36 |
+
version="1.0.0"
|
37 |
+
)
|
38 |
+
|
39 |
+
# Modelos Pydantic
|
40 |
+
class ChatRequest(BaseModel):
|
41 |
+
message: str
|
42 |
+
max_tokens: int = 200
|
43 |
+
temperature: float = 0.7
|
44 |
+
|
45 |
+
class ChatResponse(BaseModel):
|
46 |
+
response: str
|
47 |
+
status: str = "success"
|
48 |
+
|
49 |
+
# Lock para thread safety
|
50 |
+
model_lock = threading.Lock()
|
51 |
+
|
52 |
+
def generate_response(message: str, max_tokens: int = 200, temperature: float = 0.7) -> str:
|
53 |
+
"""Gerar resposta com o modelo"""
|
54 |
try:
|
55 |
+
with model_lock:
|
56 |
+
prompt = f"<|system|>\nVocê é um assistente útil. Responda de forma clara e concisa.<|user|>\n{message}<|assistant|>\n"
|
57 |
+
|
58 |
+
inputs = tokenizer(
|
59 |
+
prompt,
|
60 |
+
return_tensors="pt",
|
61 |
+
truncation=True,
|
62 |
+
max_length=1000,
|
63 |
+
padding=False
|
64 |
+
)
|
65 |
+
|
66 |
+
with torch.no_grad():
|
67 |
+
outputs = model.generate(
|
68 |
+
inputs.input_ids,
|
69 |
+
max_new_tokens=min(max_tokens, 300),
|
70 |
+
temperature=max(0.1, min(temperature, 1.0)),
|
71 |
+
do_sample=True,
|
72 |
+
top_p=0.9,
|
73 |
+
repetition_penalty=1.1,
|
74 |
+
pad_token_id=tokenizer.eos_token_id,
|
75 |
+
eos_token_id=tokenizer.eos_token_id
|
76 |
+
)
|
77 |
+
|
78 |
+
response = tokenizer.decode(
|
79 |
+
outputs[0][len(inputs.input_ids[0]):],
|
80 |
+
skip_special_tokens=True
|
81 |
)
|
82 |
+
|
83 |
+
# Limpar resposta
|
84 |
+
response = response.split("<|user|>")[0].split("<|system|>")[0].strip()
|
85 |
+
|
86 |
+
return response if response else "Não consegui gerar uma resposta."
|
87 |
+
|
88 |
+
except Exception as e:
|
89 |
+
raise HTTPException(status_code=500, detail=f"Erro na geração: {str(e)}")
|
90 |
+
|
91 |
+
# Endpoints da API
|
92 |
+
|
93 |
+
@app.get("/")
|
94 |
+
async def root():
|
95 |
+
"""Endpoint raiz - informações da API"""
|
96 |
+
return {
|
97 |
+
"message": "TinyLlama Chat API",
|
98 |
+
"model": MODEL_NAME,
|
99 |
+
"endpoints": {
|
100 |
+
"POST /chat": "Enviar mensagem para o modelo",
|
101 |
+
"GET /health": "Verificar status da API",
|
102 |
+
"GET /docs": "Documentação interativa"
|
103 |
+
}
|
104 |
+
}
|
105 |
+
|
106 |
+
@app.get("/health")
|
107 |
+
async def health_check():
|
108 |
+
"""Verificar se a API está funcionando"""
|
109 |
+
return {
|
110 |
+
"status": "healthy",
|
111 |
+
"model_loaded": True,
|
112 |
+
"model_name": MODEL_NAME
|
113 |
+
}
|
114 |
+
|
115 |
+
@app.post("/chat", response_model=ChatResponse)
|
116 |
+
async def chat_endpoint(request: ChatRequest):
|
117 |
+
"""Endpoint principal para chat"""
|
118 |
+
if not request.message or not request.message.strip():
|
119 |
+
raise HTTPException(status_code=400, detail="Mensagem não pode estar vazia")
|
120 |
+
|
121 |
+
try:
|
122 |
+
response = generate_response(
|
123 |
+
message=request.message,
|
124 |
+
max_tokens=request.max_tokens,
|
125 |
+
temperature=request.temperature
|
126 |
+
)
|
127 |
|
128 |
+
return ChatResponse(response=response)
|
|
|
|
|
129 |
|
130 |
+
except Exception as e:
|
131 |
+
raise HTTPException(status_code=500, detail=str(e))
|
132 |
+
|
133 |
+
@app.get("/chat")
|
134 |
+
async def chat_get(message: str, max_tokens: int = 200, temperature: float = 0.7):
|
135 |
+
"""Endpoint GET para chat (mais simples de testar)"""
|
136 |
+
if not message or not message.strip():
|
137 |
+
raise HTTPException(status_code=400, detail="Parâmetro 'message' é obrigatório")
|
138 |
+
|
139 |
+
try:
|
140 |
+
response = generate_response(
|
141 |
+
message=message,
|
142 |
+
max_tokens=max_tokens,
|
143 |
+
temperature=temperature
|
144 |
+
)
|
145 |
|
146 |
+
return {"response": response, "status": "success"}
|
147 |
|
148 |
except Exception as e:
|
149 |
+
raise HTTPException(status_code=500, detail=str(e))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
150 |
|
151 |
if __name__ == "__main__":
|
152 |
+
print("🚀 Iniciando servidor FastAPI...")
|
153 |
+
print("📡 API estará disponível em:")
|
154 |
+
print(" - GET / (informações)")
|
155 |
+
print(" - GET /health (status)")
|
156 |
+
print(" - POST /chat (principal)")
|
157 |
+
print(" - GET /chat (teste simples)")
|
158 |
+
print(" - GET /docs (documentação)")
|
159 |
+
|
160 |
+
uvicorn.run(
|
161 |
+
app,
|
162 |
+
host="0.0.0.0",
|
163 |
+
port=7860,
|
164 |
+
log_level="error" # Reduzir logs
|
165 |
)
|