[Attention][UX][1/N] Add AttentionConfig and change attention env vars to CLI arguments (#26315)

Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>

vllm/v1/attention/backends/rocm_attn.py

@@ -165,7 +165,7 @@ class RocmAttentionBackend(AttentionBackend):
             raise ValueError(
                 f"Head size {head_size} is not supported by {attn_type}. "
                 f"Supported head sizes are: {cls.get_supported_head_sizes()}. "
-                "Set VLLM_ATTENTION_BACKEND=FLEX_ATTENTION to use "
+                "Set --attention-config.backend=FLEX_ATTENTION to use "
                 "FlexAttention backend which supports all head sizes."
             )
 

vllm/v1/attention/backends/triton_attn.py

@@ -210,9 +210,6 @@ class TritonAttentionImpl(AttentionImpl):
     def fused_output_quant_supported(self, quant_key: QuantKey):
         return quant_key == kFp8StaticTensorSym
 
-    def supports_quant_query_input(self) -> bool:
-        return current_platform.is_cuda()
-
     def __init__(
         self,
         num_heads: int,
@@ -262,6 +259,8 @@ class TritonAttentionImpl(AttentionImpl):
                 f"num_heads: {num_heads}."
             )
 
+        self.supports_quant_query_input = current_platform.is_cuda()
+
     def forward(
         self,
         layer: torch.nn.Module,