[Quantization][Bugfix] Bump humming-kernels to 0.1.16 (#58054)
Signed-off-by: jinzhen.ljz <jinzhen.ljz@antgroup.com> Co-authored-by: Codex <noreply@openai.com> Co-authored-by: Misha Goin <mgoin64@gmail.com>
Jinzhen Lin committed
Sep 23, 2026 at 23:22 UTC
f69cc7c178d5bd47cd9bc5cf4c2f41530b878aae
3 files changed
+1
-3
requirements/cuda.txt
+1
-1
@@ -33,4 +33,4 @@ quack-kernels==0.6.5 # Required by MSA
33
tokenspeed-mla==0.1.8; platform_system == "Linux"
34
35
# Humming kernels for quantization gemm
36
-humming-kernels[cu13]==0.1.15
36
+humming-kernels[cu13]==0.1.16
tests/evals/gsm8k/configs/humming/Qwen3-0.6B-MXFP8-humming.yaml
-1
@@ -3,6 +3,5 @@ accuracy_threshold: 0.39
3
num_questions: 1319
4
num_fewshot: 5
5
server_args: >-
6
- --enforce-eager
6
--max-model-len 8192
7
--linear-backend humming
tests/evals/gsm8k/configs/humming/Qwen3-30B-A3B-FP8-block-humming.yaml
-1
@@ -3,7 +3,6 @@ accuracy_threshold: 0.86
3
num_questions: 1319
4
num_fewshot: 5
5
server_args: >-
6
- --enforce-eager
6
--max-model-len 8192
7
--moe-backend humming
8
--linear-backend humming