Update Dockerfile

Dockerfile

@@ -10,20 +10,19 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     ca-certificates \
     && rm -rf /var/lib/apt/lists/*
 
-# ---------------- Python deps ----------------
-# Install torch FIRST, pinned, CPU by default (much smaller + stable)
+# ---------------- Python deps (PINNED & COMPATIBLE) ----------------
 RUN pip install --no-cache-dir --upgrade pip && \
     pip install --no-cache-dir \
-    torch==2.1.2 \
-    torchvision==0.16.2 \
-    torchaudio==2.1.2 \
+    numpy<2 \
+    torch==2.0.1 \
+    torchvision==0.15.2 \
+    torchaudio==2.0.2 \
     --index-url https://download.pytorch.org/whl/cpu
 
-# Then the rest
 RUN pip install --no-cache-dir \
-    diffusers[torch] \
-    transformers \
-    accelerate \
+    diffusers==0.24.0 \
+    transformers==4.36.2 \
+    accelerate==0.25.0 \
     safetensors \
     flask \
     flask-cors \
@@ -34,7 +33,7 @@ ENV HOME=/home/sd
 ENV HF_HOME=/home/sd/.cache
 ENV OMP_NUM_THREADS=1
 ENV MKL_NUM_THREADS=1
-ENV PYTORCH_ENABLE_MPS_FALLBACK=1
+ENV NUMPY_EXPERIMENTAL_ARRAY_FUNCTION=0
 
 # ---------------- Storage ----------------
 RUN mkdir -p /home/sd && chmod -R 777 /home/sd
@@ -46,6 +45,7 @@ from flask_cors import CORS
 from diffusers import DiffusionPipeline, LCMScheduler
 import torch, os, json, secrets
 from io import BytesIO
+from datetime import datetime
 
 app = Flask(__name__)
 CORS(app)
@@ -63,32 +63,24 @@ for p in [WL_PATH, USAGE_PATH, LIMITS_PATH]:
     if not os.path.exists(p):
         open(p, "w").write("{}" if p.endswith(".json") else "")
 
-print(f"Loading {MODEL_ID}...")
+print(f"Loading model: {MODEL_ID}")
 
 torch.set_grad_enabled(False)
-torch.backends.cuda.matmul.allow_tf32 = True
-torch.backends.cudnn.allow_tf32 = True
 
 pipe = DiffusionPipeline.from_pretrained(
     MODEL_ID,
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    torch_dtype=torch.float32,
     safety_checker=None
 )
 
 pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
+pipe = pipe.to("cpu")
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-pipe = pipe.to(device)
-
-# ---- SPEED + MEMORY OPTS ----
+# ---- MEMORY + SPEED ----
 pipe.enable_attention_slicing()
 pipe.enable_vae_slicing()
 
-if device == "cuda":
-    pipe.enable_model_cpu_offload()
-    pipe.unet.to(memory_format=torch.channels_last)
-
-print(f"Model ready on {device}")
+print("Model loaded (CPU, optimized)")
 
 def whitelist():
     try:
@@ -113,12 +105,14 @@ def health():
 def generate_key():
     data = request.get_json() or {}
     key = "sk-" + secrets.token_hex(16)
+
     with open(WL_PATH, "a") as f:
         f.write(key + "\n")
 
     limits = load_json(LIMITS_PATH)
     limits[key] = "unlimited" if data.get("unlimited") else int(data.get("limit", DEFAULT_LIMIT))
     save_json(LIMITS_PATH, limits)
+
     return jsonify({"key": key, "limit": limits[key]})
 
 @app.route("/api/generate", methods=["POST"])
@@ -134,11 +128,10 @@ def generate():
 
     limits = load_json(LIMITS_PATH)
     usage = load_json(USAGE_PATH)
-    limit = limits.get(key, DEFAULT_LIMIT)
 
-    from datetime import datetime
     month = datetime.now().strftime("%Y-%m")
     used = usage.get(key, {}).get(month, 0)
+    limit = limits.get(key, DEFAULT_LIMIT)
 
     if limit != "unlimited" and used >= limit:
         return jsonify({"error": "Monthly limit reached"}), 429