Add GLM-4.7-FP8 tuned/untuned BF16 GEMM configs (gfx950)

[omirosh]

Summary

• Adds tuned BF16 GEMM configurations for GLM-4.7-FP8 detected from vLLM server log untuned warnings on gfx950.
• New files: aiter/configs/model_configs/glm47_bf16_tuned_gemm.csv and aiter/configs/model_configs/glm47_bf16_untuned_gemm.csv.

Test plan

• Verify the new configs are picked up by the GEMM tuner / dispatcher on gfx950.
• Re-run the vLLM server scenario that produced the untuned warnings and confirm they no longer appear.

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests on MI35X (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:triton-300x	Run an additional Triton test job on MI300X in PRs; main branch always runs both MI35X and MI300X
ci:sglang	SGLang integration tests: DeepSeek-R1-MXFP4 accuracy, Qwen 3.5 accuracy
ci:atom	ATOM benchmark: DeepSeek-R1-0528, GPT-OSS-120B
ci:atom_full	ATOM accuracy suite for PR and main models from ATOM models_accuracy.json
ci:vllm	vLLM benchmark: GPT-OSS-120B, DeepSeek-R1-0528, Kimi-K2.5
ci:all	All standard extended tests (excludes ci:atom_full)

Only add ci:atom_full for FlyDSL or Triton upgrades.
Add labels via the sidebar or gh pr edit 3285 --add-label <label>

[Copilot]

Pull request overview

Adds model-specific BF16 GEMM shape lists and their tuned kernel selections for the GLM-4.7-FP8 deployment scenario on gfx950, so the runtime GEMM dispatcher can pick optimized kernels and avoid “untuned” warnings.

Changes:

• Add a GLM-4.7 BF16 untuned GEMM shape list (M,N,K,...) for reproducing/tuning and avoiding missing-shape warnings.
• Add a GLM-4.7 BF16 tuned GEMM config (gfx,cu_num,...,libtype,solidx,...) targeting gfx950 with cu_num=256.

Reviewed changes

Copilot reviewed 1 out of 2 changed files in this pull request and generated no comments.

File	Description
aiter/configs/model_configs/glm47_bf16_untuned_gemm.csv	Adds the set of BF16 GEMM shapes observed in the target workload so tuning/coverage can be driven from a stable CSV.
aiter/configs/model_configs/glm47_bf16_tuned_gemm.csv	Provides tuned kernel selections for the above shapes on gfx950, enabling the dispatcher to select optimized implementations.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.