mi355 glm5 fp8 sgl single node

[functionstackx]

@claude create/submit/open an PR to add GLM5 to Mi355 single node in fp8 for tp8 from conc 4 to conc 64.

use the image rocm/sgl-dev:v0.5.8.post1-rocm720-mi35x-20260219

here is example recipe from anush on how to do sglang mi355 glm5

python -m sglang.launch_server \
  --model zai-org/GLM-5-FP8 \
  --tp 8 \
  --mem-fraction-static 0.8 \
  --nsa-prefill-backend tilelang \
  --nsa-decode-backend tilelang

https://gist.github.com/powderluv/8156ec484215f11810532d4a84e7537d

[Klaud-Cold]

Claude finished @functionstackx's task in 4m 24s —— View job • claude/issue-761-20260219-2050 • Create PR ➔

---

Adding GLM5 to MI355 Single Node (FP8, SGLang)

• Read repo structure and existing configs
• Analyze existing MI355 SGLang configs for reference
• Add glm5-fp8-mi355x-sglang config to amd-master.yaml
• Create benchmark script benchmarks/glm5_fp8_mi355x.sh
• Update perf-changelog.yaml
• Commit and push changes
• Provide PR link

Changes

3 files changed:

1. .github/configs/amd-master.yaml - Added glm5-fp8-mi355x-sglang config entry:

   • Image: rocm/sgl-dev:v0.5.8.post1-rocm720-mi35x-20260219
   • Model: zai-org/GLM-5-FP8
   • TP=8, concurrency 4-64 for 1k1k, 1k8k, 8k1k

2. benchmarks/glm5_fp8_mi355x.sh - New benchmark script based on qwen3.5_bf16_mi355x.sh pattern with GLM5-specific args:

   • --attention-backend triton
   • --mem-fraction-static 0.8
   • --nsa-prefill-backend tilelang
   • --nsa-decode-backend tilelang

3. perf-changelog.yaml - Added changelog entry for the new config

|