TRTLLM-MLA FP8 path (#8638)

Signed-off-by: Faraz Khoubsirat <58580514+farazkh80@users.noreply.github.com>

python/sglang/srt/server_args.py

@@ -432,7 +432,10 @@ class ServerArgs:
             )
             self.page_size = 128
 
-        if self.attention_backend == "trtllm_mla":
+        if (
+            self.attention_backend == "trtllm_mla"
+            or self.decode_attention_backend == "trtllm_mla"
+        ):
             if not is_sm100_supported():
                 raise ValueError(
                     "TRTLLM MLA backend is only supported on Blackwell GPUs (SM100). Please use a different backend."
@@ -443,11 +446,17 @@ class ServerArgs:
                     f"TensorRT-LLM MLA only supports page_size of 32 or 64, changing page_size from {self.page_size} to 64."
                 )
                 self.page_size = 64
+
             if self.speculative_algorithm is not None:
                 raise ValueError(
                     "trtllm_mla backend does not support speculative decoding yet."
                 )
 
+            if self.kv_cache_dtype not in ["fp8_e4m3", "auto"]:
+                raise ValueError(
+                    "TensorRT-LLM MLA backend only supports kv-cache-dtype of fp8_e4m3 or auto."
+                )
+
         if (
             self.attention_backend == "trtllm_mha"
             or self.decode_attention_backend == "trtllm_mha"