From 5d8e1c9279678b3342d9618167121e758ed00c05 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Antonio=20G=C3=B3mez?= Date: Tue, 25 Mar 2025 18:59:25 +0100 Subject: [PATCH] [Bugfix] Support triton==3.3.0+git95326d9f for RTX 5090 (Unsloth + vLLM compatibility) (#15471) Co-authored-by: ServerAI --- vllm/lora/ops/triton_ops/kernel_utils.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/lora/ops/triton_ops/kernel_utils.py b/vllm/lora/ops/triton_ops/kernel_utils.py index 3572d3018622a..5b8c19376106a 100644 --- a/vllm/lora/ops/triton_ops/kernel_utils.py +++ b/vllm/lora/ops/triton_ops/kernel_utils.py @@ -130,7 +130,7 @@ def do_expand_kernel( # Identify A and B block pointers offset_k = tl.arange(0, BLOCK_K) a_ptr = (cur_input_ptr + ram[:, None] * input_d1_stride + - offset_k[None, :] * input_d2_stride, ) + offset_k[None, :] * input_d2_stride) b_ptr = (cur_lora_ptr + cur_lora_d0_stride * lora_index + offset_k[:, None] * cur_lora_d2_stride + rbn[None, :] * cur_lora_d1_stride)