Upload ZETAGRID_INFERENCE.py with huggingface_hub

ZETAGRID_INFERENCE.py

@@ -21,10 +21,29 @@ print("=" * 70)
 # CONFIG (must match training)
 # ============================================================
 
-BASE_DIR = "/workspace/zetagrid_50b"
+# Detect Environment: RunPod vs Local Windows
+if os.name == 'nt':
+    print("🖥️  Running on Local Windows (E:/ZETAGRID)")
+    BASE_DIR = r"E:/ZETAGRID"
+    CKPT_DIR = BASE_DIR # Checkpoints are in root
+    DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+else:
+    print("☁️  Running on Linux/RunPod")
+    BASE_DIR = "/workspace/zetagrid_50b"
+    # Fallback to standard check path
+    CKPT_DIR = f"{BASE_DIR}/phase2_checkpoints"
+    DEVICE = "cuda"
+
 GENOME_PATH = f"{BASE_DIR}/zetagrid_25b_production.npy"
 DTYPE = torch.bfloat16
-DEVICE = "cuda"
+
+# PRE-FLIGHT CHECK
+if not os.path.exists(GENOME_PATH):
+    print(f"❌ CRITICAL ERROR: Genome file not found at {GENOME_PATH}")
+    print("   Please ensure correct path or download the model.")
+    if os.name == 'nt':
+        print("   On Windows, ensure E:/ZETAGRID is mounted/accessible.")
+    exit(1)
 
 VOCAB_SIZE = 256
 D_MODEL = 4096
@@ -34,9 +53,6 @@ KERNEL_SIZE = 3
 LORA_RANK = 128
 DILATION_CYCLE = [1, 2, 4, 8, 16, 32, 64, 128]
 
-# Find best checkpoint
-CKPT_DIR = f"{BASE_DIR}/phase2_checkpoints"
-
 # ============================================================
 # MODEL (same as training)
 # ============================================================