[V1] [Hybrid] Support Minimax-Text-01 in V1 (#22151)

Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
2025-08-09 08:08:48 +02:00
parent 3157aebb63
commit 6ade99eafa
5 changed files with 234 additions and 42 deletions
--- a/vllm/model_executor/layers/lightning_attn.py
+++ b/vllm/model_executor/layers/lightning_attn.py
@@ -532,7 +532,7 @@ def _linear_attn_decode_kernel(
    pid_d = tl.program_id(2)  # dimension block index

    # Load slot index for the current batch
-    slot_id = tl.load(slot_idx + pid_b)
+    slot_id = tl.load(slot_idx + pid_b).to(tl.int64)

    # Skip if slot_id is -1 (padding)
    if slot_id == -1: