# GLM-5.3-Flash NVFP4 — three quant variants measured on 2× DGX Spark (ds4)

Measured 2026-08-27/28 on cluster `ds4` (dgx3+dgx4, switched fabric), all three
served by sm121 through the launcher, same benchmark invocation:

```
tool-eval-bench run --base-url http://<head>:30000/v1 --model glm53 \
  --seed 42 --hardmode --perf --concurrency 1,2
```

All three share served name **`glm53`** — clients never changed between runs.

## Results

| recipe | quant | quality | points | decode tg t/s (c1) | tg c2 | TTFT | deployability | MTP |
|---|---|---|---|---|---|---|---|---|
| `glm53-flash-nvfp4-lab` | local-inference-lab mixed NVFP4/MXFP8-MTP | **89** | 155/174 | **28.4** | **43.4** | 3.9 s | 70 | ✅ k=3, accept len ~2.9–3.5, ~63–83 % |
| `glm53-flash-nvfp4-redhat` | RedHatAI W4A4 compressed-tensors | **90** | 159/176 | 13.9 | 13.8 | 2.2 s | 67 | ❌ (no MTP weights in checkpoint) |
| `glm53-flash-nvfp4` | LibertAIDAI NVFP4-A16 | **88** | 154/176 | 26.2 | 22.7 | 6.3 s | 69 | ✅ k=5 |

(lab scored /174 — one scenario skipped by the harness that run; quality % is
comparable. LibertAIDAI is the only variant whose MTP *regresses* under c2
(26.2→22.7) — lab's k=3 draft scales, LibertAIDAI's k=5 does not.)

Prefill (pp2048) 0.3–1.7 K t/s depending on depth/concurrency — noisy, all three
in the same band (NAS load-influenced); decode is the stable discriminator.

## Read of the numbers

- **lab** is the throughput pick: MTP doubles decode (28–43 t/s vs ~14 baseline)
  at −1 quality point. Acceptance lengths matched upstream's 2.55–3.5 range.
- **redhat** is the quality pick (90/100 — best of the three) but is a
  **baseline-decode dead-end on this box: its checkpoint ships zero MTP tensors**
  (verified index: 146,745 tensors, none under layer 45), so it is stuck at ~14 t/s.
- **lab = best agent-throughput/day-driver; redhat = eval ceiling;
  LibertAIDAI = solid fallback** (proven first, but k=5 MTP regresses at c2).

## Images & patches

| image dir | base | patches |
|---|---|---|
| `glm53-flash-nvfp4` | vllm/vllm-openai:glm53-flash-arm64-cu130 | kingjones30 NoPE-MLA mod (4-part, anchor-asserted) |
| `glm53-flash-nvfp4-lab` | same | + 4 FujitsuPolycom patches (Apache-2.0, see patches/UPSTREAM.md): model.py naming shim, modelopt MTP namespace fix, persistent_topk CC-12.x guards ×2 |
| redhat recipe reuses `glm53-flash-nvfp4:local` (mod is quant-independent) | | |

Boot markers `[quantprobe] layer=RoutedExperts prefix=model.layers.45.mlp.experts
algo=MXFP8` confirm the MTP fix is live; `algo=None` there = broken load.

## Operational notes (measured)

- lab weight load: 92.19 GiB / 700 s sequential from the NAS HF cache;
  `--kv-cache-memory-bytes 10737418240` → KV 1,164,369 tokens / 2.22× @ 512K
  (matches upstream to the digit — do NOT switch back to GMU sizing).
- First launcher serve of lab died on both ranks at ~54 s (exit 1, logs lost
  to `--rm`); re-run was clean twice. Suspected transient (cold imports +
  simultaneous launch). If it recurs: re-launch without `--rm` and keep logs.
- `sm121 serve` immediately after `sm121 stop` can fail with "cluster has only
  1 nodes" — occupancy race, nodes not yet released; wait ~10 s and retry.
- LibertAIDAI/redhat recipes still run FlashInfer autotune (60-min possible
  boot, `VLLM_ENGINE_READY_TIMEOUT_S=3600` set); the lab recipe disables it via
  `--kernel-config` per upstream's proven boot.
- Chat-template caveat: `reasoning_effort:"low"` still emits a reasoning trace
  (budget ≠ 0); `max_tokens:48` can yield `content: null, finish: length` —
  use ≥256 for smoke tests.

## Follow-ups

- Vendor instanttensor loader into the lab image (~30 s vs 700 s weight load).
- Consider adding `--kernel-config` autotune-off + pinned KV to the redhat and
  LibertAIDAI recipes (their boot could be minutes shorter; verify before trusting).
- MTP for redhat is impossible unless someone trains/exports a drafter into
  the checkpoint — don't chase it.