Luca Sacchi Ricciardi
369 lines
10 KiB
Python
369 lines
10 KiB
Python
"""
|
|
Models endpoints - Gestione dei modelli Ollama
|
|
"""
|
|
|
|
from fastapi import APIRouter, HTTPException, Query
|
|
from pydantic import BaseModel
|
|
from typing import Any, Dict, List, Optional
|
|
from datetime import datetime
|
|
import requests
|
|
import logging
|
|
from urllib.parse import urlparse
|
|
from app.config import settings
|
|
|
|
logger = logging.getLogger(__name__)
|
|
router = APIRouter()
|
|
|
|
|
|
def ensure_rw_api_enabled() -> None:
|
|
"""Blocca le API di scrittura se non abilitate esplicitamente."""
|
|
if not settings.ENABLE_MODEL_RW_API:
|
|
raise HTTPException(
|
|
status_code=404,
|
|
detail="Endpoint non disponibile"
|
|
)
|
|
|
|
|
|
def resolve_ollama_host(host: Optional[str]) -> str:
|
|
"""Resolve target Ollama host, optionally overridden by query parameter."""
|
|
if not host:
|
|
return settings.OLLAMA_HOST
|
|
|
|
parsed = urlparse(host.strip())
|
|
if parsed.scheme not in {"http", "https"} or not parsed.netloc:
|
|
raise HTTPException(status_code=422, detail="Invalid Ollama host URL")
|
|
|
|
return host.rstrip("/")
|
|
|
|
class ModelInfo(BaseModel):
|
|
"""Informazioni su un modello"""
|
|
name: str
|
|
digest: str
|
|
size: int
|
|
modified_at: datetime
|
|
|
|
class Config:
|
|
json_schema_extra = {
|
|
"example": {
|
|
"name": "llama2",
|
|
"digest": "abc123def456...",
|
|
"size": 3825922048,
|
|
"modified_at": "2024-01-15T10:30:00Z"
|
|
}
|
|
}
|
|
|
|
class ModelsResponse(BaseModel):
|
|
"""Risposta con lista di modelli"""
|
|
models: List[ModelInfo]
|
|
total: int
|
|
|
|
class Config:
|
|
json_schema_extra = {
|
|
"example": {
|
|
"models": [
|
|
{
|
|
"name": "llama2",
|
|
"digest": "abc123def456...",
|
|
"size": 3825922048,
|
|
"modified_at": "2024-01-15T10:30:00Z"
|
|
}
|
|
],
|
|
"total": 1
|
|
}
|
|
}
|
|
|
|
@router.get("/models", response_model=ModelsResponse)
|
|
async def get_models(host: Optional[str] = Query(default=None)):
|
|
"""
|
|
Recupera l'elenco di tutti i modelli caricati in Ollama
|
|
|
|
Returns:
|
|
ModelsResponse: Lista dei modelli disponibili
|
|
|
|
Raises:
|
|
HTTPException: Se Ollama non è disponibile
|
|
"""
|
|
target_host = resolve_ollama_host(host)
|
|
try:
|
|
response = requests.get(
|
|
f"{target_host}/api/tags",
|
|
timeout=settings.OLLAMA_TIMEOUT
|
|
)
|
|
|
|
if response.status_code != 200:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Ollama non disponibile"
|
|
)
|
|
|
|
data = response.json()
|
|
models_data = data.get("models", [])
|
|
|
|
models = [
|
|
ModelInfo(
|
|
name=model.get("name", "unknown"),
|
|
digest=model.get("digest", ""),
|
|
size=model.get("size", 0),
|
|
modified_at=datetime.fromisoformat(
|
|
model.get("modified_at", "").replace("Z", "+00:00")
|
|
) if model.get("modified_at") else datetime.utcnow()
|
|
)
|
|
for model in models_data
|
|
]
|
|
|
|
return ModelsResponse(
|
|
models=models,
|
|
total=len(models)
|
|
)
|
|
|
|
except requests.exceptions.Timeout:
|
|
raise HTTPException(
|
|
status_code=504,
|
|
detail="Timeout: Ollama non ha risposto in tempo"
|
|
)
|
|
except requests.exceptions.ConnectionError:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Impossible connettersi a Ollama"
|
|
)
|
|
except HTTPException:
|
|
raise
|
|
except Exception as e:
|
|
logger.error(f"Error fetching models: {e}")
|
|
raise HTTPException(
|
|
status_code=500,
|
|
detail="Errore nel recupero dei modelli"
|
|
)
|
|
|
|
|
|
@router.get("/models/running")
|
|
async def get_running_models(host: Optional[str] = Query(default=None)) -> Dict[str, Any]:
|
|
"""
|
|
Recupera i modelli attualmente residenti in memoria, equivalenti a `ollama ps`.
|
|
|
|
Returns:
|
|
Dict[str, Any]: Payload con modelli running e conteggio
|
|
"""
|
|
target_host = resolve_ollama_host(host)
|
|
try:
|
|
response = requests.get(
|
|
f"{target_host}/api/ps",
|
|
timeout=settings.OLLAMA_TIMEOUT
|
|
)
|
|
|
|
if response.status_code != 200:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Ollama non disponibile"
|
|
)
|
|
|
|
data = response.json()
|
|
models_data = data.get("models", [])
|
|
|
|
return {
|
|
"models": models_data,
|
|
"total": len(models_data)
|
|
}
|
|
|
|
except requests.exceptions.Timeout:
|
|
raise HTTPException(
|
|
status_code=504,
|
|
detail="Timeout: Ollama non ha risposto in tempo"
|
|
)
|
|
except requests.exceptions.ConnectionError:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Impossible connettersi a Ollama"
|
|
)
|
|
except HTTPException:
|
|
raise
|
|
except Exception as e:
|
|
logger.error(f"Error fetching running models: {e}")
|
|
raise HTTPException(
|
|
status_code=500,
|
|
detail="Errore nel recupero dei modelli residenti"
|
|
)
|
|
|
|
@router.get("/models/{model_name}", response_model=ModelInfo)
|
|
async def get_model(model_name: str, host: Optional[str] = Query(default=None)):
|
|
"""
|
|
Recupera le informazioni di un modello specifico
|
|
|
|
Args:
|
|
model_name: Nome del modello da cercare
|
|
|
|
Returns:
|
|
ModelInfo: Informazioni del modello
|
|
|
|
Raises:
|
|
HTTPException: Se il modello non esiste o Ollama non è disponibile
|
|
"""
|
|
target_host = resolve_ollama_host(host)
|
|
try:
|
|
response = requests.get(
|
|
f"{target_host}/api/tags",
|
|
timeout=settings.OLLAMA_TIMEOUT
|
|
)
|
|
|
|
if response.status_code != 200:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Ollama non disponibile"
|
|
)
|
|
|
|
data = response.json()
|
|
models_data = data.get("models", [])
|
|
|
|
# Cercare il modello
|
|
for model in models_data:
|
|
if model.get("name") == model_name:
|
|
return ModelInfo(
|
|
name=model.get("name", "unknown"),
|
|
digest=model.get("digest", ""),
|
|
size=model.get("size", 0),
|
|
modified_at=datetime.fromisoformat(
|
|
model.get("modified_at", "").replace("Z", "+00:00")
|
|
) if model.get("modified_at") else datetime.utcnow()
|
|
)
|
|
|
|
raise HTTPException(
|
|
status_code=404,
|
|
detail=f"Modello '{model_name}' non trovato"
|
|
)
|
|
|
|
except HTTPException:
|
|
raise
|
|
except Exception as e:
|
|
logger.error(f"Error fetching model: {e}")
|
|
raise HTTPException(
|
|
status_code=500,
|
|
detail="Errore nel recupero del modello"
|
|
)
|
|
|
|
|
|
@router.get("/models/{model_name}/show")
|
|
async def get_model_show(model_name: str, host: Optional[str] = Query(default=None)) -> Dict[str, Any]:
|
|
"""
|
|
Recupera le informazioni estese di un modello tramite endpoint Ollama /api/show.
|
|
|
|
Args:
|
|
model_name: Nome del modello da interrogare
|
|
|
|
Returns:
|
|
Dict[str, Any]: Dati estesi del modello
|
|
"""
|
|
target_host = resolve_ollama_host(host)
|
|
try:
|
|
response = requests.post(
|
|
f"{target_host}/api/show",
|
|
json={"model": model_name},
|
|
timeout=settings.OLLAMA_TIMEOUT
|
|
)
|
|
|
|
if response.status_code == 404:
|
|
raise HTTPException(
|
|
status_code=404,
|
|
detail=f"Modello '{model_name}' non trovato"
|
|
)
|
|
|
|
if response.status_code != 200:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Errore durante il recupero dettagli modello"
|
|
)
|
|
|
|
return response.json()
|
|
|
|
except requests.exceptions.Timeout:
|
|
raise HTTPException(
|
|
status_code=504,
|
|
detail="Timeout: Ollama non ha risposto in tempo"
|
|
)
|
|
except requests.exceptions.ConnectionError:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Impossible connettersi a Ollama"
|
|
)
|
|
except HTTPException:
|
|
raise
|
|
except Exception as e:
|
|
logger.error(f"Error fetching model show data: {e}")
|
|
raise HTTPException(
|
|
status_code=500,
|
|
detail="Errore nel recupero dei dettagli modello"
|
|
)
|
|
|
|
@router.post(
|
|
"/models/{model_name}/pull",
|
|
include_in_schema=settings.ENABLE_MODEL_RW_API
|
|
)
|
|
async def pull_model(model_name: str):
|
|
"""
|
|
Scarica/carica un modello in Ollama
|
|
|
|
Args:
|
|
model_name: Nome del modello da scaricare
|
|
|
|
Returns:
|
|
dict: Status del download
|
|
"""
|
|
ensure_rw_api_enabled()
|
|
try:
|
|
response = requests.post(
|
|
f"{settings.OLLAMA_HOST}/api/pull",
|
|
json={"name": model_name},
|
|
timeout=None # Pull può essere lungo
|
|
)
|
|
|
|
if response.status_code not in [200, 201]:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Errore nel pull del modello"
|
|
)
|
|
|
|
return {"status": "pulling", "model": model_name}
|
|
|
|
except Exception as e:
|
|
logger.error(f"Error pulling model: {e}")
|
|
raise HTTPException(
|
|
status_code=500,
|
|
detail="Errore nel pull del modello"
|
|
)
|
|
|
|
@router.delete(
|
|
"/models/{model_name}",
|
|
include_in_schema=settings.ENABLE_MODEL_RW_API
|
|
)
|
|
async def delete_model(model_name: str):
|
|
"""
|
|
Elimina un modello da Ollama
|
|
|
|
Args:
|
|
model_name: Nome del modello da eliminare
|
|
|
|
Returns:
|
|
dict: Confirmazione eliminazione
|
|
"""
|
|
ensure_rw_api_enabled()
|
|
try:
|
|
response = requests.delete(
|
|
f"{settings.OLLAMA_HOST}/api/delete",
|
|
json={"name": model_name},
|
|
timeout=settings.OLLAMA_TIMEOUT
|
|
)
|
|
|
|
if response.status_code not in [200, 204]:
|
|
raise HTTPException(
|
|
status_code=502,
|
|
detail="Errore nell'eliminazione del modello"
|
|
)
|
|
|
|
return {"status": "deleted", "model": model_name}
|
|
|
|
except Exception as e:
|
|
logger.error(f"Error deleting model: {e}")
|
|
raise HTTPException(
|
|
status_code=500,
|
|
detail="Errore nell'eliminazione del modello"
|
|
)
|