Fix upstream PR 22668 that added additional arg to is_kv_cache_dtype_supported (#96)

Fixes vllm-project/vllm#22668 - we need to take
one more arg.

Signed-off-by: Marcin Swiniarski <mswiniarski@habana.ai>
Signed-off-by: Marcin Swiniarski <marcin.swiniarski@intel.com>

Author: mswiniarsk

vllm_gaudi/platform.py

@@ -154,7 +154,8 @@ def set_torch_compile(cls) -> None:
             os.environ['PT_HPU_ENABLE_LAZY_COLLECTIVES'] = 'true'
 
     @classmethod
-    def is_kv_cache_dtype_supported(cls, kv_cache_dtype: str) -> bool:
+    def is_kv_cache_dtype_supported(cls, kv_cache_dtype: str,
+                                    model_config: ModelConfig) -> bool:
         return kv_cache_dtype == "fp8_inc"
 
     @classmethod