[Bugfix][Hardware][POWERPC] Fix auto dtype failure in case of POWER10 (…

…vllm-project#11331)

Signed-off-by: Akash Kaothalkar <[email protected]>

vllm/config.py

@@ -22,7 +22,7 @@
 from vllm.model_executor.layers.quantization import (QUANTIZATION_METHODS,
                                                      get_quantization_config)
 from vllm.model_executor.models import ModelRegistry
-from vllm.platforms import current_platform
+from vllm.platforms import current_platform, interface
 from vllm.tracing import is_otel_available, otel_import_error_traceback
 from vllm.transformers_utils.config import (
     ConfigFormat, get_config, get_hf_image_processor_config,
@@ -2199,6 +2199,17 @@ def _get_and_verify_dtype(
             else:
                 torch_dtype = config_dtype
 
+            if (current_platform.is_cpu()
+                    and current_platform.get_cpu_architecture()
+                    == interface.CpuArchEnum.POWERPC
+                    and (config_dtype == torch.float16
+                         or config_dtype == torch.float32)):
+                logger.info(
+                    "For POWERPC, we cast models to bfloat16 instead of "
+                    "using float16 by default. Float16 is not currently "
+                    "supported for POWERPC.")
+                torch_dtype = torch.bfloat16
+
             if current_platform.is_hpu() and config_dtype == torch.float16:
                 logger.info(
                     "For HPU, we cast models to bfloat16 instead of"