disable prefix on random for qwen3.5 & glm

[functionstackx]

@Oseltamivir

https://github.com/SemiAnalysisAI/InferenceX/blob/main/benchmarks/single_node/qwen3.5_bf16_mi300x.sh
https://github.com/SemiAnalysisAI/InferenceX/blob/main/benchmarks/single_node/qwen3.5_bf16_mi325x.sh
https://github.com/SemiAnalysisAI/InferenceX/blob/main/benchmarks/single_node/qwen3.5_bf16_mi355x.sh
https://github.com/SemiAnalysisAI/InferenceX/blob/main/benchmarks/single_node/qwen3.5_fp8_mi300x.sh
https://github.com/SemiAnalysisAI/InferenceX/blob/main/benchmarks/single_node/qwen3.5_fp8_mi325x.sh
https://github.com/SemiAnalysisAI/InferenceX/blob/main/benchmarks/single_node/qwen3.5_fp8_mi355x.sh
https://github.com/SemiAnalysisAI/InferenceX/blob/main/benchmarks/single_node/glm5_fp8_mi355x.sh

[functionstackx]

@claude can u draft up an inital PR that bryan can use to validate & build on

[Klaud-Cold]

Claude finished @functionstackx's task in 1m 45s —— View job • claude/issue-968-20260328-0530 • Create PR ➔

---

Disable prefix on random for qwen3.5 & glm

• Gather context — read referenced files and recent PR Disable prefix minimax #966
• Understand the pattern — SGLang uses --disable-radix-cache
• Apply changes to all 7 listed benchmark scripts
• Commit and push changes
• Provide PR link

Added --disable-radix-cache to the SGLang server launch command in all 7 scripts listed in the issue. This matches the pattern used in other scripts (e.g., qwen3.5_bf16_b200.sh, qwen3.5_fp8_h200.sh, glm5_fp8_b200.sh).

Note: glm5_fp8_h200.sh also appears to be missing --disable-radix-cache but wasn't listed in the issue — Bryan may want to add it as well.

|

[functionstackx]

@claude redo disabling prefix cache for these models but do it off the latest main branch

[Klaud-Cold]

Claude finished @functionstackx's task in 1m 2s —— View job • claude/issue-968-20260328-0808 • Create PR ➔

---

Disable prefix caching for qwen3.5 & glm (rebased on latest main)

• Reset branch to latest main (626932a)
• Read all 7 benchmark scripts
• Add --disable-radix-cache to each script's SGLang server launch command
• Commit and push changes
• Provide PR link

Added --disable-radix-cache to the SGLang server launch command in all 7 scripts, based off the latest main branch:

• qwen3.5_bf16_mi300x.sh
• qwen3.5_bf16_mi325x.sh
• qwen3.5_bf16_mi355x.sh
• qwen3.5_fp8_mi300x.sh
• qwen3.5_fp8_mi325x.sh
• qwen3.5_fp8_mi355x.sh
• glm5_fp8_mi355x.sh

Note: glm5_fp8_h200.sh also appears to be missing --disable-radix-cache but wasn't listed in the issue — Bryan may want to add it as well.

• claude/issue-968-20260328-0808