fix

LucasWilkinson · LucasWilkinson · commit a3f6647a6155 · 2025-11-09T22:27:15.000-08:00
Signed-off-by: Lucas Wilkinson &lt;lwilkins@redhat.com&gt;
diff --git a/vllm/v1/attention/backends/mla/flashmla_sparse.py b/vllm/v1/attention/backends/mla/flashmla_sparse.py
@@ -636,8 +636,10 @@ def __init__(
                 vllm_config.model_config.max_model_len
             )
             self.prefill_workspace_shape = (prefill_workspace_size, head_size)
-            self.prefill_bf16_workspace = current_workspace_manager().get_simultaneous(
-                (self.prefill_workspace_shape, torch.bfloat16)
+            (self.prefill_bf16_workspace,) = (
+                current_workspace_manager().get_simultaneous(
+                    (self.prefill_workspace_shape, torch.bfloat16)
+                )
             )
 
     def _forward_bf16_kv(

Original file line number	Diff line number	Diff line change
`@@ -636,8 +636,10 @@ def __init__(`
`636`	`636`	`vllm_config.model_config.max_model_len`
`637`	`637`	`)`
`638`	`638`	`self.prefill_workspace_shape = (prefill_workspace_size, head_size)`
`639`		`- self.prefill_bf16_workspace = current_workspace_manager().get_simultaneous(`
`640`		`- (self.prefill_workspace_shape, torch.bfloat16)`
	`639`	`+ (self.prefill_bf16_workspace,) = (`
	`640`	`+ current_workspace_manager().get_simultaneous(`
	`641`	`+ (self.prefill_workspace_shape, torch.bfloat16)`
	`642`	`+ )`
`641`	`643`	`)`
`642`	`644`
`643`	`645`	`def _forward_bf16_kv(`