Spaces:
Sleeping
Sleeping
File size: 4,124 Bytes
abbab7a 404886a b605fd6 404886a b605fd6 abbab7a b605fd6 abbab7a 404886a b605fd6 404886a b605fd6 6161aaf b605fd6 404886a b605fd6 404886a b605fd6 404886a b605fd6 404886a |
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 |
import os
import gradio as gr
from huggingface_hub import InferenceClient
from huggingface_hub.utils import HfHubHTTPError
# Modelo Mistral Instruct disponível no Hub
MODEL_ID = "mistralai/Mistral-7B-Instruct-v0.2"
# token vindo do secret HF_TOKEN do Space (ou env local)
token = os.environ.get("HF_TOKEN")
# Cliente (se token for None, o client tenta usar config local)
client = InferenceClient(model=MODEL_ID, token=token)
def _extract_text_from_response(resp):
"""
Tenta extrair texto de várias possíveis formas de retorno da API.
Retorna string sempre.
"""
# string direta
if isinstance(resp, str):
return resp
# dataclass-like (possível)
try:
# alguns SDKs retornam objeto com atributo 'generated_text' ou 'text'
if hasattr(resp, "generated_text"):
return getattr(resp, "generated_text") or ""
if hasattr(resp, "text"):
return getattr(resp, "text") or ""
except Exception:
pass
# dict-like formas comuns
if isinstance(resp, dict):
# chaves óbvias
for key in ("generated_text", "generated_texts", "text", "output_text", "result"):
if key in resp:
v = resp[key]
if isinstance(v, list) and v:
return v[0] if isinstance(v[0], str) else str(v[0])
if isinstance(v, str):
return v
# choices -> message -> content (formato chat-like)
if "choices" in resp and isinstance(resp["choices"], list) and resp["choices"]:
first = resp["choices"][0]
if isinstance(first, dict):
# try message.content
if "message" in first and isinstance(first["message"], dict) and "content" in first["message"]:
maybe = first["message"]["content"]
if isinstance(maybe, str):
return maybe
# try text or content directly
for k in ("text", "content", "generated_text"):
if k in first and isinstance(first[k], str):
return first[k]
# fallback
try:
return str(resp)
except Exception:
return "<unable to decode response>"
def respond(
message,
history: list[tuple[str, str]],
system_message,
max_tokens,
temperature,
top_p,
):
# valida token
if not token:
yield "ERRO: variável de ambiente HF_TOKEN não encontrada. Adicione um secret HF_TOKEN no Settings do Space."
return
# monta prompt estilo chat (simples)
prompt = f"{system_message}\n\n"
for user_msg, bot_msg in history:
if user_msg:
prompt += f"User: {user_msg}\n"
if bot_msg:
prompt += f"Assistant: {bot_msg}\n"
prompt += f"User: {message}\nAssistant:"
try:
# chamada sem streaming (resposta completa)
out = client.text_generation(
prompt,
max_new_tokens=int(max_tokens),
temperature=float(temperature),
top_p=float(top_p),
do_sample=True,
)
except HfHubHTTPError as e:
# captura erros HTTP da Hugging Face e retorna mensagem legível
yield f"ERRO na chamada de inferência: {e}\n(verifique HF_TOKEN, permissões e se o modelo está disponível via Inference API)"
return
except Exception as e:
yield f"Erro inesperado ao chamar a API: {e}"
return
# extrai texto (robusto a vários formatos de retorno)
text = _extract_text_from_response(out)
yield text
demo = gr.ChatInterface(
respond,
additional_inputs=[
gr.Textbox(value="You are a helpful assistant.", label="System message"),
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
gr.Slider(minimum=0.0, maximum=2.0, value=0.7, step=0.05, label="Temperature"),
gr.Slider(minimum=0.0, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
],
title="Chat com Mistral-7B",
)
if __name__ == "__main__":
demo.launch()
|