Removed custom_io passing for bf16 dtype case (quic#983)

- Removed custom_io passing when running through bf16 dtype
- Removed BF16 Warning Emitted by QEfficient Despite BF16

Signed-off-by: Asmita Goswami <asmigosw@qti.qualcomm.com>
Co-authored-by: Hem Agnihotri <hemagnih@qti.qualcomm.com>

Author: asmigosw

QEfficient/base/modeling_qeff.py

@@ -95,9 +95,6 @@ def __init__(self, model: torch.nn.Module, **kwargs) -> None:
         else:
             logger.info(f"Pytorch transforms applied to model: {self.model_name}")
 
-        if self.config.torch_dtype == torch.bfloat16:
-            logger.warning("BFloat16 dtype is not yet supported; converting to float16 precision!")
-
     def _normalize_torch_dtype(self):
         """
         Normalizes torch_dtype across all nested configs to match the top-level config.
@@ -644,7 +641,7 @@ def _compile(
         # Write custom_io.yaml file
         model_in_bfloat16 = hasattr(self, "config") and (self.config.torch_dtype == torch.bfloat16)
         pkv_in_bfloat16 = (custom_io is not None) and any(
-            "past_" in key and "bfloat16" in value for key, value in custom_io.items()
+            ("past_" in key or "pixel_values" in key) and "bfloat16" in value for key, value in custom_io.items()
         )
         if custom_io is not None:
             custom_io_yaml = compile_dir / "custom_io.yaml"