Comrade Chief Designer
bd8c72ff47
A torch-profiler kernel-time window (496 decode tokens, rank0 trace parsed for
device-activity time) on the deployed GLM-5.3-Flash shape falsifies this page's
own "the expert gather is the bound" explanation:
46.18 s of kernel time / 496 tokens = ~93 ms per token (actual step ~104 ms)
34.8 % vectorized_elementwise_kernel 708,500 launches (1,428/token)
10.0 % r4d_gemm_moe_w4a16_nt_b16_m64 (the MoE GEMM)
6.7 % elementwise_kernel_manual_unroll
6.4 % wvSplitK_hf_sml_ (split-K GEMM)
8.9 % router: sbtopk + warpMergeSortTopK + bitonicSortKVInPlace
6.5 % r4d_mhc_* (hyper-connections)
1.05 % r4d_moe_lru_gather_kernel <-- the retracted suspect
The earlier arithmetic was circular (it assumed the gather consumed the whole
step). The real bound is unfused pointwise glue: the fork compiles its own Qwen
trunk (support_torch_compile in models/qwen4_exp/amd/model.py) and not its GLM
trunk, so the engine boots with "provides neither a compiled submodule nor
breakable CUDA graphs. Overriding to FULL_DECODE_ONLY." That also explains why
MTP4 measured flat — pointwise cost scales with the token count, so spec decode
amortizes nothing here — and it retires arena/hit-rate tuning (the gather is 1 %).
Staged as documented patch artifacts (not wired into any serve script; the
engine does not yet boot with them) — scripts/patches/glm53-compile-0{1,2,3}*.patch:
-01 models/glm5next/common/model.py decorate Glm5NextModel (R4D_GLM_COMPILE)
-02 model_executor/kernels/r4d_lib.py freeze 10 capability probes that do
os.path.exists/.so lookups inside the traced forward. functools.lru_cache
does NOT protect them: dynamo re-traces the wrapped body even when warm
(verified in-image); a plain module-global read compiles.
-03 compilation/wrapper.py make fullgraph=True opt-out, because
suppress_errors does NOT cover graph breaks and this trunk can never
satisfy fullgraph (indexer metadata syncs to host — .item() at
glm5next/amd/sparse_indexer.py:263,419).
With all three the trunk traces (Dynamo bytecode transform time 1.08 s) and dies
on KeyError: in_proj_qkvbfg_a in the mHC/KDA path — a latent bug the compile
exposes, and the remaining blocker.
Also records two structural constraints: no new HIP kernels are possible (r4d.so
ships without sources), which binary-blocks the fused router at GLM's 288
experts; and DFlash2 for GLM is a TARGET-side port, not a drafter port
(incoai/GLM-5.3-Flash-DFlash2 is architectures ["DFlash2DraftModel"] +
model_type qwen3, already mapped at registry.py:636; glm5next lacks only the
aux-hidden-state plumbing deepseek_v2.py demonstrates).
Runbook frontmatter, intro, mermaid and wall section, AGENTS.md and the work log
corrected to match the measurement. tests/run.sh: 362 seams pass.
Server App Configs
The single repository for the two homelab servers: the Podman container stacks and the LLM inference fleet. One repo, one checkout per host, one deploy script.
2026-08-22: the former
vllmScriptsrepository (LLM fleet) is merged into this repo with its full git history. See AGENTS.md for the merge note.
Targets
| Host | User | IP | Podman stack | LLM service | LLM port |
|---|---|---|---|---|---|
| dankserver | dank | 192.168.0.235 | media and library | frank-medium (ROCmFPX llama.cpp, Qwen3.8-27B AGENT) | 7770 |
| dankclaw | dank | 192.168.0.143 | AI, search, development, observability | frank-large (gufo, Qwen3.8-27B UD-Q8_K_XL) | 7770 |
Active Services
| Service | Host | Hostname | Port |
|---|---|---|---|
| RomM | dankserver | romm.dankserver.local | 8085 |
| Copyparty | dankserver | files.dankserver.local | 3923 |
| Calibre-Web | dankserver | books.dankserver.local | 8083 |
| frank-medium | dankserver | — (bare endpoint) | 7770 |
| SearXNG | dankclaw | search.dankclaw.local | 8081 |
| ComfyUI | dankclaw | comfyui.dankclaw.local | 8188 |
| OpenWebUI | dankclaw | chat.dankclaw.local | 3000 |
| Hermes | dankclaw | hermes.dankclaw.local | 8090 |
| Gitea | dankclaw | gitea.dankclaw.local | 3001 |
| Grafana | dankclaw | grafana.dankclaw.local | 3002 |
| Prometheus | dankclaw | prometheus.dankclaw.local | 9090 |
| frank-large | dankclaw | — (bare endpoint) | 7770 |
Structure
dankserver/— compose.yaml + Caddyfile + .env.example for the Dankserver stackdankclaw/— compose.yaml + Caddyfile + .env.example for the Dankclaw stackshared/— homepage, prometheus config, deploy helpersscripts/—deploy.shanddeploy-remote.sh(service-scoped deployment), the planner, LLM serve scripts, systemd units, env tunables, and bench harnessesold/— retired PP/NPU-era LLM scripts (reference only)serverAppConfigs-docs/— project knowledge base (Obsidian vault, wikilinks).serverAppConfigs-docs/LLM/is the LLM fleettests/— bats test suite (see Testing Policy)
Deployment
See serverAppConfigs-docs/Architecture/Deployment Flow.md
and serverAppConfigs-docs/LLM/deployment.md. Run
./scripts/deploy.sh <dankclaw|dankserver> after committing changes. It
plans and deploys only the changed services. Use --service NAME for an
explicit single-service retry; inference is restarted only when its own files
(or the shared chat template) changed.
Knowledge Base
The serverAppConfigs-docs/ directory is an Obsidian vault. Start at
serverAppConfigs-docs/index.md. The LLM fleet knowledge base lives at
serverAppConfigs-docs/LLM/index.md.
Testing
Run ./tests/run.sh before every push. See
serverAppConfigs-docs/Architecture/Testing Policy.md.