A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 17:15 UTC
[Bugfix] Support triton==3.3.0+git95326d9f for RTX 5090 (Unsloth + vLLM compatibility) (#15471)
Co-authored-by: ServerAI <ai@exc-mad-ai.com>
Antonio Gómez committed
Mar 25, 2025 at 18:59 UTC
5d8e1c9279678b3342d9618167121e758ed00c05
1 file changed
+1
-1
vllm/lora/ops/triton_ops/kernel_utils.py
+1
-1
@@ -130,7 +130,7 @@ def do_expand_kernel(
130
# Identify A and B block pointers
131
offset_k = tl.arange(0, BLOCK_K)
132
a_ptr = (cur_input_ptr + ram[:, None] * input_d1_stride +
133
- offset_k[None, :] * input_d2_stride, )
133
+ offset_k[None, :] * input_d2_stride)
134
b_ptr = (cur_lora_ptr + cur_lora_d0_stride * lora_index +
135
offset_k[:, None] * cur_lora_d2_stride +
136
rbn[None, :] * cur_lora_d1_stride)