Skip to content

Commit d648774

Browse files
JordanNanosclaude
andcommitted
Switch to v0.5.10-bnxt-patched (PD fixes + transformers patch)
SGLang main broke the disagg router for all models. Use v0.5.10 (has PD fixes #19086 merged) + transformers patch for model type recognition. v0.5.10 has working router + PD KV cache fixes. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1 parent 947e339 commit d648774

File tree

1 file changed

+3
-3
lines changed

1 file changed

+3
-3
lines changed

.github/configs/amd-master.yaml

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -113,7 +113,7 @@ dsr1-fp8-mi325x-sglang-mtp:
113113
- { tp: 8, conc-start: 4, conc-end: 64, spec-decoding: mtp }
114114

115115
glm5-fp8-mi325x-sglang:
116-
image: semianalysiswork/sgl-bnxt-cdna3:main-bnxt
116+
image: semianalysiswork/sgl-bnxt-cdna3:v0.5.10-bnxt-patched
117117
model: zai-org/GLM-5-FP8
118118
model-prefix: glm5
119119
runner: mi325x
@@ -1619,7 +1619,7 @@ dsr1-fp8-mi325x-sglang-disagg-mtp:
16191619
- "DECODE_MTP_SIZE=3"
16201620

16211621
qwen3.5-fp8-mi325x-sglang-disagg:
1622-
image: semianalysiswork/sgl-bnxt-cdna3:main-bnxt
1622+
image: semianalysiswork/sgl-bnxt-cdna3:v0.5.10-bnxt-patched
16231623
model: Qwen/Qwen3.5-397B-A17B-FP8
16241624
model-prefix: qwen3.5
16251625
runner: mi325x-disagg
@@ -1802,7 +1802,7 @@ qwen3.5-fp8-mi325x-sglang-disagg:
18021802
- "MODEL_YAML_KEY=Qwen3.5-397B-A17B-FP8-no-mori"
18031803

18041804
glm5-fp8-mi325x-sglang-disagg:
1805-
image: semianalysiswork/sgl-bnxt-cdna3:main-bnxt
1805+
image: semianalysiswork/sgl-bnxt-cdna3:v0.5.10-bnxt-patched
18061806
model: zai-org/GLM-5-FP8
18071807
model-prefix: glm5
18081808
runner: mi325x-disagg

0 commit comments

Comments
 (0)