app.py

import torch
import torch.nn as nn
from torch.nn import functional as F
from fastapi import FastAPI, Request
from fastapi.responses import HTMLResponse
import requests
import json
import os
import re
import uvicorn

# --- CONFIGURATION ---
# Assure-toi que cette variable est bien dans les "Secrets" de ton Space
OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY")
device = 'cpu'
n_embd, n_head, n_layer, block_size, dropout = 384, 6, 6, 256, 0.2

# --- ARCHITECTURE DU MODÈLE ---
class Head(nn.Module):
    def __init__(self, head_size):
        super().__init__()
        self.key = nn.Linear(n_embd, head_size, bias=False)
        self.query = nn.Linear(n_embd, head_size, bias=False)
        self.value = nn.Linear(n_embd, head_size, bias=False)
        self.register_buffer('tril', torch.tril(torch.ones(block_size, block_size)))
        self.dropout = nn.Dropout(dropout)
    def forward(self, x):
        B, T, C = x.shape
        k, q, v = self.key(x), self.query(x), self.value(x)
        wei = q @ k.transpose(-2, -1) * C**-0.5
        wei = wei.masked_fill(self.tril[:T, :T] == 0, float('-inf'))
        wei = F.softmax(wei, dim=-1)
        return self.dropout(wei) @ v

class Block(nn.Module):
    def __init__(self, n_embd, n_head):
        super().__init__()
        head_size = n_embd // n_head
        self.sa = nn.ModuleList([Head(head_size) for _ in range(n_head)])
        self.proj = nn.Linear(n_embd, n_embd)
        self.ffwd = nn.Sequential(
            nn.Linear(n_embd, 4 * n_embd), 
            nn.ReLU(), 
            nn.Linear(4 * n_embd, n_embd), 
            nn.Dropout(dropout)
        )
        self.ln1, self.ln2 = nn.LayerNorm(n_embd), nn.LayerNorm(n_embd)
    def forward(self, x):
        x = x + self.proj(torch.cat([h(self.ln1(x)) for h in self.sa], dim=-1))
        x = x + self.ffwd(self.ln2(x))
        return x

class CygnisNano(nn.Module):
    def __init__(self, vocab_size):
        super().__init__()
        self.token_embedding_table = nn.Embedding(vocab_size, n_embd)
        self.position_embedding_table = nn.Embedding(block_size, n_embd)
        self.blocks = nn.Sequential(*[Block(n_embd, n_head) for _ in range(n_layer)])
        self.ln_f = nn.LayerNorm(n_embd)
        self.lm_head = nn.Linear(n_embd, vocab_size)
    def forward(self, idx, targets=None):
        B, T = idx.shape
        tok_emb = self.token_embedding_table(idx)
        pos_emb = self.position_embedding_table(torch.arange(T, device=device))
        x = self.blocks(tok_emb + pos_emb)
        return self.lm_head(self.ln_f(x)), None

# --- CHARGEMENT ---
with open('vocab.json', 'r', encoding='utf-8') as f:
    vocab = json.load(f)
    stoi = vocab['stoi']
    itos = {int(k): v for k, v in vocab['itos'].items()}

encode = lambda s: [stoi[c] for c in s if c in stoi]
decode = lambda l: ''.join([itos[i] for i in l])

model = CygnisNano(len(stoi))
model.load_state_dict(torch.load('cygnis_nano.pth', map_location=device))
model.eval()

# --- LOGIQUE DE RÉPONSE ---
def logic_cygnis(prompt):
    if not OPENROUTER_API_KEY:
        return "Erreur : La clé API OpenRouter est manquante."

    try:
        res = requests.post(
            url="https://openrouter.ai/api/v1/chat/completions",
            headers={
                "Authorization": f"Bearer {OPENROUTER_API_KEY}",
                "HTTP-Referer": "https://huggingface.co/spaces",
            },
            json={
                "model": "openai/gpt-oss-20b:free",
                "messages": [
                    {"role": "system", "content": "Tu es Cygnis Nano. Réponds directement, sans préambule, et ne mets pas de signature type 'CygnisAI:'."},
                    {"role": "user", "content": prompt}
                ]
            },
            timeout=15
        )
        
        if res.status_code == 200:
            data = res.json()
            raw_text = data['choices'][0]['message']['content']
            # Nettoyage des préfixes si Gemma en met quand même
            clean_text = re.sub(r'^(CygnisAI:|Cygnis Nano:)', '', raw_text, flags=re.IGNORECASE).strip()
            return clean_text
        
        # --- BLOC DE DEBUGGING AJOUTÉ ---
        else:
            try:
                error_details = res.json()
                error_msg = error_details.get('error', {}).get('message', 'Erreur inconnue')
                return f"Erreur API ({res.status_code}): {error_msg}"
            except:
                return f"Erreur de connexion (Code: {res.status_code})"
        # -------------------------------

    except Exception as e:
        return f"Erreur technique : {str(e)}"

# --- APP FASTAPI ---
app = FastAPI()

@app.get("/", response_class=HTMLResponse)
async def serve_index():
    try:
        with open("index.html", "r", encoding="utf-8") as f:
            return f.read()
    except FileNotFoundError:
        return "Fichier index.html introuvable au racine du Space."

@app.post("/ask")
async def ask_api(request: Request):
    data = await request.json()
    user_prompt = data.get("prompt", "")
    response_text = logic_cygnis(user_prompt)
    return {"full_response": response_text}

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=7860)