diag: add NaN/Inf + FP8-dtype checks after NVFP4 dequant

patches/deepseek_v4.py

@@ -1970,6 +1970,15 @@ class DeepseekV4Model(nn.Module):
             w_dequant = w_bf16
         
         # Replace weight with bf16 version
+        # Diagnostics: check for NaN/Inf from bad dequant
+        if w_dequant.isnan().any() or w_dequant.isinf().any():
+            nan_count = w_dequant.isnan().sum().item()
+            inf_count = w_dequant.isinf().sum().item()
+            print(f"[NVFP4-DEQUANT-WARN] {getattr(mod, 'prefix', 'unknown')}: "
+                  f"shape={w_dequant.shape}, dtype={w_dequant.dtype}, "
+                  f"NaN={nan_count}, Inf={inf_count}, "
+                  f"block_scale range=[{block_scale.min().item():.6f}, {block_scale.max().item():.6f}], "
+                  f"global_scale={global_scale:.6f}")
         mod.weight = torch.nn.Parameter(w_dequant, requires_grad=False)
         from vllm.model_executor.layers.linear import UnquantizedLinearMethod
         mod.quant_method = UnquantizedLinearMethod()
@@ -2066,6 +2075,14 @@ class DeepseekV4Model(nn.Module):
                 delattr(mod, attr)
         from vllm.model_executor.layers.linear import UnquantizedLinearMethod
         mod.quant_method = UnquantizedLinearMethod()
+        # Safety check: UnquantizedLinearMethod with FP8 weight will crash CUBLAS
+        if mod.weight.dtype in (torch.float8_e4m3fn, torch.float8_e5m2):
+            print(f"[NVFP4-FP8-CRASH] {getattr(mod, 'prefix', 'unknown')}: "
+                  f"weight is {mod.weight.dtype} but quant_method=UnquantizedLinearMethod! "
+                  f"This will crash CUBLAS.")
+        for attr in ("weight_scale", "weight_scale_2", "input_scale"):
+            if hasattr(mod, attr):
+                delattr(mod, attr)
 
     def _reconstruct_compressor_weight(self, fused_mod, parent_mod, layer_idx, e2m1_lut, sub_path=""):
         """Reconstruct compressor fused_wkv_wgate from checkpoint.