Upload Dockerfile with huggingface_hub
Browse files- Dockerfile +145 -0
Dockerfile
ADDED
|
@@ -0,0 +1,145 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Dockerfile ottimizzato per Hugging Face Spaces - Jaksel AI
|
| 2 |
+
# Usa immagine base con Python e CUDA per GPU
|
| 3 |
+
FROM python:3.11-slim
|
| 4 |
+
|
| 5 |
+
# Imposta environment variables
|
| 6 |
+
ENV DEBIAN_FRONTEND=noninteractive
|
| 7 |
+
ENV PYTHONUNBUFFERED=1
|
| 8 |
+
ENV OLLAMA_HOST=0.0.0.0
|
| 9 |
+
ENV PORT=7860
|
| 10 |
+
ENV HF_HOME=/data
|
| 11 |
+
ENV TRANSFORMERS_CACHE=/data/transformers_cache
|
| 12 |
+
|
| 13 |
+
# Installa dipendenze di sistema
|
| 14 |
+
RUN apt-get update && apt-get install -y \
|
| 15 |
+
curl \
|
| 16 |
+
ca-certificates \
|
| 17 |
+
&& rm -rf /var/lib/apt/lists/*
|
| 18 |
+
|
| 19 |
+
# Installa Ollama
|
| 20 |
+
RUN curl -fsSL https://ollama.ai/install.sh | sh
|
| 21 |
+
|
| 22 |
+
# Copia l'applicazione
|
| 23 |
+
WORKDIR /app
|
| 24 |
+
COPY . .
|
| 25 |
+
|
| 26 |
+
# Installa dipendenze Python
|
| 27 |
+
RUN pip install --no-cache-dir fastapi uvicorn python-multipart aiohttp
|
| 28 |
+
|
| 29 |
+
# Crea directory per il modello
|
| 30 |
+
RUN mkdir -p /root/.ollama
|
| 31 |
+
|
| 32 |
+
# Scarica Jaksel model ( questo avviene al runtime per evitare problemi con HF cache )
|
| 33 |
+
# Il modello verrà scaricato al primo avvio
|
| 34 |
+
|
| 35 |
+
# Script di startup che gestisce il download del modello e avvia Ollama
|
| 36 |
+
COPY <<EOF /app/start.sh
|
| 37 |
+
#!/bin/bash
|
| 38 |
+
echo "🚀 Starting Jaksel AI on Hugging Face Spaces..."
|
| 39 |
+
|
| 40 |
+
# Assicura che la directory esista
|
| 41 |
+
mkdir -p /root/.ollama
|
| 42 |
+
|
| 43 |
+
# Avvia Ollama in background
|
| 44 |
+
echo "📥 Starting Ollama server..."
|
| 45 |
+
ollama serve --host 0.0.0.0 --port 11434 &
|
| 46 |
+
|
| 47 |
+
# Attendi che Ollama sia avviato
|
| 48 |
+
echo "⏳ Waiting for Ollama to start..."
|
| 49 |
+
sleep 10
|
| 50 |
+
|
| 51 |
+
# Pull Jaksel model ( se non già presente )
|
| 52 |
+
echo "🤖 Pulling Jaksel model..."
|
| 53 |
+
ollama pull zantara-jaksel:latest || echo "Model already exists or download failed, continuing..."
|
| 54 |
+
|
| 55 |
+
# Tenta di nuovo se il primo download fallisce
|
| 56 |
+
if ! ollama list | grep -q "zantara-jaksel"; then
|
| 57 |
+
echo "🔄 Retrying model download..."
|
| 58 |
+
sleep 5
|
| 59 |
+
ollama pull zantara-jaksel:latest
|
| 60 |
+
fi
|
| 61 |
+
|
| 62 |
+
# Crea proxy server per gli endpoint HF
|
| 63 |
+
python <<PYTHON
|
| 64 |
+
import subprocess
|
| 65 |
+
import time
|
| 66 |
+
import requests
|
| 67 |
+
from fastapi import FastAPI, Request
|
| 68 |
+
import uvicorn
|
| 69 |
+
import json
|
| 70 |
+
|
| 71 |
+
app = FastAPI(title="Jaksel AI Proxy")
|
| 72 |
+
|
| 73 |
+
@app.get("/")
|
| 74 |
+
async def root():
|
| 75 |
+
return {"message": "Jaksel AI is running!", "status": "healthy"}
|
| 76 |
+
|
| 77 |
+
@app.get("/health")
|
| 78 |
+
async def health():
|
| 79 |
+
try:
|
| 80 |
+
response = requests.get("http://127.0.0.1:11434/api/tags", timeout=5)
|
| 81 |
+
if response.status_code == 200:
|
| 82 |
+
models = response.json().get("models", [])
|
| 83 |
+
jaksel_found = any("zantara-jaksel" in m.get("name", "") for m in models)
|
| 84 |
+
return {
|
| 85 |
+
"status": "healthy",
|
| 86 |
+
"ollama": "connected",
|
| 87 |
+
"jaksel_loaded": jaksel_found,
|
| 88 |
+
"models": [m.get("name") for m in models]
|
| 89 |
+
}
|
| 90 |
+
else:
|
| 91 |
+
return {"status": "unhealthy", "ollama": "error"}
|
| 92 |
+
except Exception as e:
|
| 93 |
+
return {"status": "unhealthy", "error": str(e)}
|
| 94 |
+
|
| 95 |
+
@app.post("/api/generate")
|
| 96 |
+
@app.post("/api/chat")
|
| 97 |
+
async def proxy_ollama(request: Request):
|
| 98 |
+
"""Proxy per richieste a Ollama"""
|
| 99 |
+
try:
|
| 100 |
+
# Ottieni body della richiesta
|
| 101 |
+
body = await request.json()
|
| 102 |
+
|
| 103 |
+
# Forward a Ollama
|
| 104 |
+
ollama_url = "http://127.0.0.1:11434" + request.scope.get("path", "")
|
| 105 |
+
|
| 106 |
+
response = requests.post(
|
| 107 |
+
ollama_url,
|
| 108 |
+
json=body,
|
| 109 |
+
headers={
|
| 110 |
+
"Content-Type": "application/json",
|
| 111 |
+
},
|
| 112 |
+
timeout=120 # Timeout più lungo per HF Spaces
|
| 113 |
+
)
|
| 114 |
+
|
| 115 |
+
return Response(
|
| 116 |
+
content=response.content,
|
| 117 |
+
status_code=response.status_code,
|
| 118 |
+
headers={
|
| 119 |
+
"Content-Type": "application/json",
|
| 120 |
+
}
|
| 121 |
+
)
|
| 122 |
+
except Exception as e:
|
| 123 |
+
return {
|
| 124 |
+
"error": f"Proxy error: {str(e)}",
|
| 125 |
+
"response": "Maaf, Jaksel lagi nggak bisa merespon. Coba lagi ya!"
|
| 126 |
+
}
|
| 127 |
+
|
| 128 |
+
from fastapi.responses import Response
|
| 129 |
+
|
| 130 |
+
print("🌐 Starting proxy server on port 7860...")
|
| 131 |
+
uvicorn.run(app, host="0.0.0.0", port=7860)
|
| 132 |
+
PYTHON
|
| 133 |
+
|
| 134 |
+
# Se il processo termina, aspetta un po' prima di uscire
|
| 135 |
+
sleep 30
|
| 136 |
+
EOF
|
| 137 |
+
|
| 138 |
+
# Rendi eseguibile lo script
|
| 139 |
+
RUN chmod +x /app/start.sh
|
| 140 |
+
|
| 141 |
+
# Esponi le porte
|
| 142 |
+
EXPOSE 7860 11434
|
| 143 |
+
|
| 144 |
+
# Comando di avvio
|
| 145 |
+
CMD ["/app/start.sh"]
|