---
license: mit
base_model: zai-org/GLM-5.3-Flash-BF16
base_model_relation: quantized
pipeline_tag: image-text-to-text
library_name: transformers
tags:
- glm
- glm-5
- glm5_next
- tr3
- trellis
- mcg
- quantized
- 8-bit
- moe
- reasoning
- text-generation
- fidelity
- kl-divergence
- exllamav3
- fidelity-provenance
datasets:
- brandonmusic/GLM-5.3-Flash-BF16-Teacher-Logits
- malaiwah/GLM-5.3-Flash-fidelity-suite-v1
metrics:
- kl_divergence
- top1_agreement
model-index:
- name: GLM-5.3-Flash-TR3-8bpw
  results:
  - task:
      type: text-generation
      name: Distribution fidelity (KL divergence vs BF16 reference)
    dataset:
      type: brandonmusic/GLM-5.3-Flash-BF16-Teacher-Logits
      name: brandonmusic GLM-5.3-Flash sealed qualification panel v1 -- 25 final windows -- panel25 subset
      config: final25-panel25
      split: streaming
      revision: 95f4fdd94bf29989db2e0d1054e4931f55edb6aa
      args:
        panel_id: panel--glm53.brandonmusic.final25
        panel_token_sha256: 6bafe3283c54bc9342d0f30aa3199d36032d103feb92c31715be8545362790ff
        contexts: 25
        scored_positions: 51175
        context_length: 2048
        tokenizer: zai-org/GLM-5.3-Flash-BF16@a6c167b62691b2bac901344b65cb651a70f53e43
        vocab_size: 154880
        reference_id: reference--brandonmusic.glm53-bf16-fp32-logits.final25
        scope_name: panel25
        covers_full_panel: true
        scope_selection_sha256: 180678c464c8c325a4675a9fd5c54247e99624f0fc911580805e98fd960f79fe
    metrics:
    - type: kl_divergence
      name: Mean tokenwise KLD (reference || candidate), nats
      value: 0.012384191023436866
      args:
        units: nats
        higher_is_better: false
        direction: reference_to_candidate
        estimator: full_vocabulary_fp64
        accumulation_dtype: float64
        logits_dtype: fp32
        head_policy: native_head
        stack_relation: same_stack
        lane: streaming
        run_count: 2
        population_stddev_of_run_means: 0.0
        determinism: bitwise_identical_across_runs
        measurement_id: measurement--glm53.k8-8bpw-stream.brandonmusic-final25
        comparability_key: cmp--202b717f3219c414
    - type: kl_divergence_quantization_attributable
      name: KLD attributable to quantization (same-lane floor removed), nats
      value: 0.0008782684041065674
      args:
        units: nats
        higher_is_better: false
        derived: true
        derivation: candidate_minus_same_lane_floor
        floor_value: 0.011505922619330299
        floor_measurement_id: measurement--glm53.bf16-stream-floor.brandonmusic-final25
        floor_lane: streaming
        caveat: KL is not additive; this is an estimate, valid only against a floor measured on the same
          lane, same panel and same reference.
    source:
      name: quant-fidelity-registry
      url: https://huggingface.co/datasets/malaiwah/quant-fidelity-registry/viewer/measurements?q=measurement--glm53.k8-8bpw-stream.brandonmusic-final25
  - task:
      type: text-generation
      name: Distribution fidelity (KL divergence vs BF16 reference)
    dataset:
      type: brandonmusic/GLM-5.3-Flash-BF16-Teacher-Logits
      name: brandonmusic panel v1, calibration-clean subset -- 17 of 25 final windows -- clean17 subset
      config: final25-clean17
      split: streaming
      revision: 95f4fdd94bf29989db2e0d1054e4931f55edb6aa
      args:
        panel_id: panel--glm53.brandonmusic.final25-clean17
        panel_token_sha256: ecfce5997ab9106cb544d41f26c3c78730d246f3d3bf3ed8fe6f20bbd0847d83
        contexts: 17
        scored_positions: 34799
        context_length: 2048
        tokenizer: zai-org/GLM-5.3-Flash-BF16@a6c167b62691b2bac901344b65cb651a70f53e43
        vocab_size: 154880
        reference_id: reference--brandonmusic.glm53-bf16-fp32-logits.final25-clean17
        scope_name: clean17
        covers_full_panel: false
        scope_selection_sha256: 180678c464c8c325a4675a9fd5c54247e99624f0fc911580805e98fd960f79fe
    metrics:
    - type: kl_divergence
      name: Mean tokenwise KLD (reference || candidate), nats
      value: 0.0108294198698829
      args:
        units: nats
        higher_is_better: false
        direction: reference_to_candidate
        estimator: full_vocabulary_fp64
        accumulation_dtype: float64
        logits_dtype: fp32
        head_policy: native_head
        stack_relation: same_stack
        lane: streaming
        run_count: 2
        determinism: bitwise_identical_across_runs
        measurement_id: measurement--glm53.k8-8bpw-stream.brandonmusic-final25.clean17
        comparability_key: cmp--2b9c401d13806d7e
    source:
      name: quant-fidelity-registry
      url: https://huggingface.co/datasets/malaiwah/quant-fidelity-registry/viewer/measurements?q=measurement--glm53.k8-8bpw-stream.brandonmusic-final25.clean17
x_fidelity:
  spec: https://github.com/malaiwah/glm53-flash-fidelity-suite/blob/main/docs/FIDELITY-DATASET-SPEC.md
  spec_version: fidelity-provenance/v1
  role: quant
  reference_model: zai-org/GLM-5.3-Flash-BF16
  reference_revision: a6c167b62691b2bac901344b65cb651a70f53e43
  fidelity_dataset: null
  registry:
    dataset: malaiwah/quant-fidelity-registry
    schema_version: quant-fidelity-registry/v1
    artifact_id: artifact--malaiwah.glm-5.3-flash-tr3-8bpw
    measurement_ids:
    - measurement--glm53.k8-8bpw-stream.brandonmusic-final25
    - measurement--glm53.k8-8bpw-stream.brandonmusic-final25.clean17
    snapshot:
      data_sha256:
        models: 91c950e9708acce30a7210caa3286a40104a54c22e440d5c3a8279fb04b9d55a
        artifacts: 3b4e3f14ee7989e933271efbabb4cde488ac52fc09ad005ac483784620163622
        panels: c9843eeee142167dc8023f6b900159537a44296184b93722f8082e3fa919ede5
        references: 9cca6aa5f1dfc58593f4c3c0a18fa31b6db7e0018281a53c3df1f5c013aed103
        pipelines: 079820a5cf6f3c3ebe7275509a6c893e2e1b1f48d96d6eac30e8d3dd429bfcf4
        measurements: 9e29ee25ca77784f9220d0f50d95106cc4df7d1645ed4a3d693e9f057046f662
  scope_digest: attn.o=quantized:exl3-mcg@8|attn.qkv=quantized:exl3-mcg@8|embed_tokens=native:bf16|lm_head=native:bf16|mlp.down=quantized:exl3-mcg@8|mlp.gate=quantized:exl3-mcg@8|mlp.up=quantized:exl3-mcg@8|moe.experts=quantized:exl3-mcg@8|norm=native:bf16|head=native|kv=bf16
  head:
    policy: native
    quantized: false
    bits: 16
    lm_head_tensor_content_sha256: null
    lm_head_file_sha256: 47eaf729c93346a2394a72a83da2ae4126dadc51155be477d212a3f0fe3085d0
    final_norm_tensor_content_sha256: null
    final_norm_file_sha256: c228a123dee3062c3ad0129094e9d98a264e33087ee88d79c8d6c5a6e60f2fed
    equality_receipt: https://huggingface.co/datasets/malaiwah/GLM-5.3-Flash-fidelity-suite-v1/resolve/main/reports/head-equality-fp8.json
    replay_permitted: false
    note: lm_head_tensor_content_sha256 is null because no head-identity receipt has been published for
      this artifact yet. A comparator MUST refuse to replay this artifact's hidden states through any
      other artifact's head until it is filled in (FIDELITY-DATASET-SPEC HEAD-4). The published receipts
      record only the FILE digest, which is a container digest and never an identity (O-6).
  measurements:
  - id: measurement--glm53.k8-8bpw-stream.brandonmusic-final25
    lane: streaming
    status: published
    comparability_key: cmp--202b717f3219c414
    panel_id: panel--glm53.brandonmusic.final25
    reference_id: reference--brandonmusic.glm53-bf16-fp32-logits.final25
    pipeline_id: pipeline--malaiwah.glm53-stream-packed-kld
    metric_name: mean_of_run_means_tokenwise_kld
    value: 0.012384191023436866
    units: nats
    direction: reference_to_candidate
    run_count: 2
    determinism:
      evidence_kind: tokenwise_kld_sha256
      distinct_evidence_hash_count: 1
      identical_across_runs: true
      evidence_hashes:
      - 763bc4a56a371e11a0f96469885b920deb6acb2c7c576d1268fb0907577f0942
    floor_measurement_id: measurement--glm53.bf16-stream-floor.brandonmusic-final25
    quantization_attributable: 0.0008782684041065674
    quantization_attributable_withheld: null
    measured_by: self-measured
    disclosures:
    - reduced_run_count
    - non_sealed_lane
  - id: measurement--glm53.k8-8bpw-stream.brandonmusic-final25.clean17
    lane: streaming
    status: published
    comparability_key: cmp--2b9c401d13806d7e
    panel_id: panel--glm53.brandonmusic.final25-clean17
    reference_id: reference--brandonmusic.glm53-bf16-fp32-logits.final25-clean17
    pipeline_id: pipeline--malaiwah.glm53-stream-packed-kld
    metric_name: mean_of_run_means_tokenwise_kld
    value: 0.0108294198698829
    units: nats
    direction: reference_to_candidate
    run_count: 2
    determinism:
      evidence_kind: tokenwise_kld_sha256
      distinct_evidence_hash_count: 1
      identical_across_runs: true
      evidence_hashes:
      - 763bc4a56a371e11a0f96469885b920deb6acb2c7c576d1268fb0907577f0942
    floor_measurement_id: null
    quantization_attributable: null
    quantization_attributable_withheld: null
    measured_by: self-measured
    disclosures:
    - reduced_run_count
    - non_sealed_lane
    - subset_of_panel
    - calibration_panel_overlap
---

# GLM-5.3-Flash-TR3-8bpw (K8)

**The first 8-bit (K8) TR3/MCG trellis quantization of
[zai-org/GLM-5.3-Flash](https://huggingface.co/zai-org/GLM-5.3-Flash)** —
321B-total / A18B MoE, `glm5_next` hybrid architecture. Routed experts and the
MTP layer quantized at K8 (128-word trellis, MCG `0xCBAC1FED`); everything else
(KDA linear-attention layers, DSA indexer, hyper-connections, routers, norms,
embeddings, lm_head) **bit-exact native BF16**. 331.4 GB — within 1% of the
official FP8 release's footprint, at 1.66× lower divergence.

## Quality — SEALED, full panel, two bitwise-identical cold runs

> ### ⚠ Scope disclosure — this number is a **panel25** number
>
> Added 2026-08-29. **Nothing here is a correction: 0.012384 is and remains the
> correct mean over the full 25-window panel.** What changed is that the panel is
> now known to contain calibration-adjacent windows, so the scope has to travel
> with the number.
>
> [brandonmusic](https://huggingface.co/brandonmusic/GLM-5.3-Flash-tr3-4bpw)
> ran a **13-gram overlap scan** of his sealed panel against its own
> calibration-role windows and found that the whole `axis4_reasoning` domain
> shares **37–39 %** of its 13-grams with calibration material — despite the
> panel being clean at the *document-hash* level. Document-hash dedup is not
> enough. He excluded that domain and scored his primary numbers on the **17
> windows that survive**. The finding, the scan and the 0.05 threshold are his.
>
> Every malaiwah number on this panel used all 25 windows, so every one of them
> carries the same contamination. Recomputed on his clean scope, from our own
> published per-window arrays (no GPU, no re-measurement — this is arithmetic on
> data already published):
>
> | | panel25 (published) | clean17 (his scope) | move |
> |---|---:|---:|---:|
> | **K8** | 0.012384 | **0.010829** | −12.55 % |
> | K6 sealed | 0.013723 | 0.011677 | −14.91 % |
> | K6 streaming | 0.013715 | 0.011676 | −14.87 % |
> | official FP8 | 0.020615 | 0.018665 | −9.46 % |
> | BF16 floor (cross-stack) | 0.012712 | 0.010648 | −16.24 % |
> | brandonmusic 4bpw | 0.024555 | 0.024949 | **+1.61 %** |
>
> **The comparisons hold, and one of them weakens.** K8 beats the official FP8
> on **17 of 17** clean windows, and the margin *widens*: 1.66× on panel25
> becomes **1.72×** on clean17. The K8-over-K6 result survives but weakens — the
> paired BCa interval still excludes zero, but its lower bound falls from
> +0.000695 to +0.000153 and the sign test goes from p = 0.0041 to p = 0.049.
> **We will not restate "K8 is better than K6" without naming the scope.**
>
> **Do not difference a panel25 number against a clean17 one.** They are answers
> to different questions. Our registry enforces this structurally: `clean17` is
> its own derived panel with its own comparability key.
>
> The **quantization-attributable** table below cannot be recomputed on the clean
> scope — its floor is the *streaming* BF16 floor, whose receipt is scalar-only
> (run means and a tokenwise digest, no per-window array), and substituting the
> cross-stack floor would be the cross-lane subtraction our registry refuses. It
> stands as a panel25 number.
>
> Full recompute, with per-domain tables, paired intervals and provenance:
> [`reports/clean-scope-recompute.json`](https://huggingface.co/datasets/malaiwah/GLM-5.3-Flash-fidelity-suite-v1/blob/main/reports/clean-scope-recompute.json).
> Working: [PROTOCOL-ALIGNMENT.md](https://github.com/malaiwah/glm53-flash-fidelity-suite/blob/main/docs/PROTOCOL-ALIGNMENT.md) §4.
>
> **One protocol note, not a correction.** His protocol masks the 24 padded
> `lm_head` columns before the log-softmax; ours never has. Measured on his real
> teacher window, the padded columns hold ~1.6e-8 of the probability mass, and
> because this quant shares the teacher's native BF16 head the effect collapses
> to `KLD × mass` — **1.0e-10 nats**, moving the value above at its *9th*
> significant figure. For scale, our own sealed-vs-streaming bridge is 8.5e-6 and
> the window-clustered SE on this panel is 3.19e-3. No correction and no bias
> disclosure is warranted; we are adopting masking anyway. Script and receipts:
> [`bin/padded_column_study.py`](https://github.com/malaiwah/glm53-flash-fidelity-suite/blob/main/bin/padded_column_study.py).


**Mean KLD(teacher ‖ K8) = 0.012384191023436866** over the full sealed panel
(25 windows / 51,175 positions), **two cold runs producing identical means to
the last digit** (`bitwise_deterministic: true`). Quality gate passed.
Receipt: [`receipts/stream-k8-kld.json`](receipts/stream-k8-kld.json).

| Model | Mean KLD (nats) | Size | Lane |
|---|---:|---:|---|
| **This K8** | **0.012384** | 331 GB | streaming, 2 runs |
| [K6](https://huggingface.co/malaiwah/GLM-5.3-Flash-TR3-6bpw) | 0.013715 | 254 GB | streaming, 2 runs |
| K6 (sealed 8×H200, 5 runs) | 0.013723 | 254 GB | sealed EP8 |
| Official FP8 | 0.020615 | 328 GB | cross-stack |
| brandonmusic 4bpw | 0.024555 | 176 GB | his stack |
| 0xSero Dione Q4 | 0.027263 | 188 GB | [our measurement](https://huggingface.co/datasets/malaiwah/GLM-5.3-Flash-fidelity-suite-v1/blob/main/reports/dione-q4-packed-kld.json) |
| NVFP4 | 0.060535 | ~180 GB | his stack, 1 window |

**At the same footprint as the official FP8 release (331 vs 328 GB), K8 is
1.66× closer to the BF16 teacher** — and 1.11× closer than K6 at 30 % more
bytes. Weight-space corroboration: with the intermediate-channel permutation
undone, K8's shipped store is **13.2× tighter in NMSE** than K6's (3.505e-5 vs
4.624e-4, better in 30 of 30 sampled matrices).

**Lane disclosure.** Measured on the single-GPU *streaming* lane (~$6/model),
not the 8×H200 sealed lane. The lanes were bridged on this exact panel: K6
reads 0.013714889 streaming vs 0.013723385 sealed — **−8.5e-6 (0.06 %)**, with
the worst single window differing by 2.9e-4. The streaming receipt sets
`publishable_as_reproduction: false` because a different expert-combine order
is an independent measurement that agrees closely, not a bitwise reproduction.

**Methodology note worth stealing.** A single-window comparison of these two
rates is *statistically meaningless*: per-window KLD scatter has sd 1.73e-3
against a K6-vs-K8 effect of 1.22e-3. On one unlucky window (`window-0000`)
K8 appeared *worse* than K6; over the full panel it wins decisively. Never
quote a single-window KLD as a rate comparison —
[full write-up](https://github.com/malaiwah/glm53-flash-fidelity-suite/blob/main/k6/K8-ANOMALY.md).

### Quantization-attributable error (the floor removed)

Scoring the **unquantized BF16 weights** against this teacher on this panel
already costs **0.011506 nats** — the price of the comparison itself (teacher
captured on a different runtime; bf16 addition is not associative across
differing expert-combine orders). Two cold runs, identical means. Removing it:

| | panel KLD | attributable to quantization |
|---|---:|---:|
| BF16 (floor) | 0.011506 | — |
| K8 (331 GB) | 0.012384 | **0.000878** |
| K6 (254 GB) | 0.013715 | **0.002209** |

**K8's quantization error is 2.52x smaller than K6's**, against a raw ratio of
only 1.11x — K8 removes ~60% of the divergence K6 leaves behind. Raw KLD
understates differences between good quants because the floor is common to
both. Method, receipts and the ways this subtraction can be misused:
[BF16-FLOOR.md](https://github.com/malaiwah/glm53-flash-fidelity-suite/blob/main/k6/BF16-FLOOR.md).

## What this is (and is not)

- **Codec:** EXL3-format TR3/MCG trellis (turboderp's
  [exllamav3](https://github.com/turboderp-org/exllamav3) kernels @ `c5d9c657`),
  through [brandonmusic's GLM-5.3 pipeline](https://github.com/brandonmmusic-max/glm-5.3-flash-exl3-4bpw)
  with a disclosed patch series. His published core admits K3/K4/K5, so **K8 is
  a declared rate extension** — our encoder was verified **byte-identical** to
  his sealed core across 120 encodes / 624 MiB / 0 differing bytes
  ([evidence](https://github.com/malaiwah/glm53-flash-fidelity-suite/blob/main/k6/fallback/closure-comparison.json),
  [issue #1](https://github.com/brandonmmusic-max/glm-5.3-flash-exl3-4bpw/issues/1)).
- **Serving runtime:** use
  [`malaiwah/glm52-exl3-vast`](https://github.com/malaiwah/glm52-exl3-vast)
  with `MODEL_PROFILE=glm53-k8`. The image pins the qualified Glm5Next vLLM,
  B12X sparse-attention stack, native EXL3 K8 extension, CUDA, and fail-closed
  runtime overlays as one contract.
- **Not stock exllamav3/TabbyAPI or stock upstream vLLM:** those stacks do not
  carry this complete `glm5_next` + TR3/MCG K8 serving path.
- **Topology-neutral checkpoint:** canonical unsharded tensors; TP layout is a
  load-time decision. The qualified deployment is exactly TP4/DCP4 on four
  96 GiB RTX PRO 6000 Blackwell GPUs.
- **Measured memory:** the packaged runtime loads 76.31 GiB of model tensors per
  rank. At GMU 0.93, profiling reported 78.94–78.97 GiB weights + non-torch,
  2.25 GiB peak activations, zero CUDA-graph memory, and 7.10–7.14 GiB KV per
  GPU.
- **Parts-bin sibling:** encoded with the **same transform seed and calibration
  as [K6](https://huggingface.co/malaiwah/GLM-5.3-Flash-TR3-6bpw)**, so the two
  per-choice payload stores are mix-and-matchable — a K6K8 multi-precision build
  (e.g. K8 on `down_proj`, K6 on gate/up) is **offline CPU assembly, no GPU
  re-encode**. The parts bin is published: [GLM-5.3-Flash-TR3-partsbin-v1](https://huggingface.co/datasets/malaiwah/GLM-5.3-Flash-TR3-partsbin-v1).

## Provenance & disclosed deviations

Pins: BF16 source `zai-org/GLM-5.3-Flash-BF16` (weights == `a6c167b6`);
calibration = brandonmusic's published EP4 captures (sealed inventory
`f56e9d62…` adopted verbatim). Deviations, all receipted: encoded on 4×H200
SM90 (his campaign attests 4×B200 SM100; fat `9.0;10.0` extension build), K4-KL
gate satisfied via a disclosed bridge document carrying his real published K4
receipt hashes, measurement on the streaming lane at EP8 emulation with fp32
combine order. Materialization receipt: bits 8, `complete`,
`main_and_mtp_complete`, `nonrouted_native_exact`, 331,449,761,784 logical
bytes, 37,152 routed choices, 1,618 native tensors.

## Lineage on the Hub

Z.ai published two sibling roots for this model and neither declares the other:
[`zai-org/GLM-5.3-Flash`](https://huggingface.co/zai-org/GLM-5.3-Flash) (the
**FP8** release, where most traffic lands) and
[`zai-org/GLM-5.3-Flash-BF16`](https://huggingface.co/zai-org/GLM-5.3-Flash-BF16)
(the **BF16** weights). This quant declares BF16 as its `base_model` because
that is what it was actually quantized from — the FP8 release is a *sibling*
quantization of the same model, not our source, and it is the baseline we
measure against rather than build on. Quants that list FP8 as their base were
genuinely made from the FP8 weights; the trees differ for real reasons.

Related work on the same model, all measured on one panel in the
[quant-fidelity registry](https://huggingface.co/datasets/malaiwah/quant-fidelity-registry):
[brandonmusic 4bpw](https://huggingface.co/brandonmusic/GLM-5.3-Flash-tr3-4bpw),
[0xSero Dione Q4](https://huggingface.co/0xSero/GLM-5.3-Flash-EXL3-Q4),
[orcarouter MLX](https://huggingface.co/orcarouter/GLM-5.3-Flash-MLX).
Collection: [GLM-5.3-Flash — measured quants & fidelity](https://huggingface.co/collections/malaiwah/glm-53-flash-measured-quants-and-fidelity-6a91f253e7107818359f37c8).

## Credits

Base model by [Z.ai](https://huggingface.co/zai-org). Quantization pipeline,
calibration captures, and teacher panel by
[brandonmusic](https://huggingface.co/brandonmusic) — co-credited, see the
[collaboration thread](https://huggingface.co/brandonmusic/GLM-5.3-Flash-tr3-4bpw/discussions/1).
Trellis codec and kernels by [turboderp](https://github.com/turboderp-org/exllamav3).
Every tool, patch, receipt and the full campaign log:
[malaiwah/glm53-flash-fidelity-suite](https://github.com/malaiwah/glm53-flash-fidelity-suite).
Comparable measurements across quants:
[quant-fidelity-registry](https://huggingface.co/datasets/malaiwah/quant-fidelity-registry).

## Serving — live-qualified turnkey profile

### Qualification result

The shipped profile is **`glm53-k8`**. It was booted on 4× RTX PRO 6000
Blackwell 96 GiB and passed arithmetic, factual, instruction-following, strict
structured-output, and tokenizer-exact 32K retrieval startup gates. The final
published image was then pulled by digest and smoke-tested without source or
runtime-overlay mounts.

- Appliance source commit: [`a0d05f76994cf44f3667c0d2910d3b0e4d305d23`](https://github.com/malaiwah/glm52-exl3-vast/commit/a0d05f76994cf44f3667c0d2910d3b0e4d305d23)
- Checkpoint revision: `b5ef443adce36ba5a10f2d5aa682fc9f2f0d0fae`
- Qualified parent: `verdictai/glm53-flash-exl3-k4@sha256:0f1cdcc8891f1cc3a444121eb61d366289a1cbba285f0892dcbb24bc94961692`
- Published appliance: `ghcr.io/malaiwah/glm52-exl3-vast@sha256:5a0d4b370e9f6a2ef85fa8b8c213492122554b34ba18d630a3a78130758914cf`
- Shape: TP4 / DCP4 A2A, B12X sparse MLA, Triton MoE, native EXL3 K8,
  calibrated NVFP4-DS MLA KV, MTP off, eager mode, batch 512, C8, GMU 0.93
- Request limit: **458,752 tokens**; text-only qualification scope

### Why K8 uses the native eager path

The eight overlapping 16-bit MCG windows for K8 span 72 bits, while B12X's
fused EXL3 decoder represents that state in two 64-bit words. The runtime
therefore fails closed instead of merely widening the fused decoder's bitrate
guard. This profile uses ExLlamaV3's compiled native K8 extension and passes the
actual uniform layer bitrate into the K8 dispatch. `VLLM_EXL3_PREFILL_CAPACITY`
is bounded to the scheduler's 512-token batch and eager mode avoids an
unqualified graph path. B12X sparse MLA and the rest of the qualified GLM-5.3
stack remain enabled.

The engine exposed **6,610,733 logical KV tokens**, or 14.41 maximum-length
requests. Per-GPU profiling reported 78.94–78.97 GiB weights plus non-torch,
2.25 GiB peak activations, zero graph memory, and 7.10–7.14 GiB KV.

### Correctness and context

Two independent 448K trials each built a tokenizer-exact 449,461-token document
and retrieved all three facts, in 170.775 and 175.086 seconds. Short and strict
structured-output checks still passed after the long-context stress. The common
458,752-token K6/K8 cap is a correctness boundary, not an extrapolation from KV
capacity.

### Measured throughput

Unique-prefix prefill, one request, no prefix reuse:

| prompt | tok/s |
|---:|---:|
| 8K | 2,684 |
| 32K | 2,825 |
| 64K | 2,938 |
| 128K | 2,986 |

Aggregate target-only decode (`MTP_TOKENS=0`), eight requests per level:

| input context | C1 tok/s | C4 tok/s | C8 tok/s |
|---:|---:|---:|---:|
| 256 | 10.29 | 38.79 | 75.66 |
| 32K | 8.42 | 23.43 | 28.46 |
| 128K | 5.42 | 12.05 | 13.25 |

All 72 requests completed without failure, preemption, or prefix reuse. These
are measurements from one four-GPU PCIe host, not guarantees for other topology,
clock, thermal, driver, storage, or request mixes.

### K8 versus the K6 production default

K8 lowers panel KLD from K6's 0.013723 to 0.012384. After subtracting the common
BF16/runtime floor, its attributable quantization error is 2.52× smaller. The
cost is about 77 GB / 30% more checkpoint bytes, about 15.2 GiB more non-KV
memory per GPU, and about 14.4 GiB less KV per GPU. K6 is 5.3–7.3× faster at
short context and 8.3–24.4× faster when the measured 32K/128K prefill cost is
included. K8 is the fidelity-first option; K6 remains the production default.

### Docker Compose

Prerequisites: Linux x86-64, four visible RTX PRO 6000 Blackwell GPUs, NVIDIA
driver ≥ 590.48.01 / CUDA 13.2 compatibility, the NVIDIA Container Toolkit,
and roughly 400 GiB free persistent storage for the checkpoint plus caches.
PCIe P2P on this card family requires NVIDIA's open kernel modules; see the
[RTX 6000 Pro multi-GPU notes](https://github.com/brandonmmusic-max/rtx6kpro).

```yaml
name: glm53-k8
services:
  api:
    image: ghcr.io/malaiwah/glm52-exl3-vast@sha256:5a0d4b370e9f6a2ef85fa8b8c213492122554b34ba18d630a3a78130758914cf
    pull_policy: always
    restart: unless-stopped
    network_mode: host
    ipc: host
    shm_size: 32gb
    stop_grace_period: 2m
    ulimits:
      memlock:
        soft: -1
        hard: -1
    environment:
      MODEL_PROFILE: glm53-k8
      AUTH: key
      VLLM_API_KEY: ${VLLM_API_KEY:?set VLLM_API_KEY to a long random secret}
      HF_TOKEN: ${HF_TOKEN:-}
      SSH_ENABLED: "0"
      SOUL_ENABLED: "0"
      VERIFY_HEALTH_TIMEOUT_S: "3600"
    volumes:
      - /srv/glm53-turnkey:/workspace
      - /srv/glm53-cache:/cache
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 4
              capabilities: [gpu]
```

```bash
sudo mkdir -p /srv/glm53-turnkey /srv/glm53-cache
export VLLM_API_KEY="$(openssl rand -hex 32)"
docker compose up -d
docker compose logs -f
```

First boot downloads about 309 GiB and can take substantial time. The container
is ready only after the log reports `>>> Verified: serving; long-context
retrieval verified`. API: `http://HOST:8000/v1`; dashboard:
`http://HOST:1111`. The served model name is `GLM-5.3-Flash-K8`.

```bash
curl http://127.0.0.1:8000/v1/chat/completions \
  -H "Authorization: Bearer $VLLM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"GLM-5.3-Flash-K8","messages":[{"role":"user","content":"Reply with exactly READY"}],"max_tokens":256}'
```

Do not replace only the checkpoint path in another vLLM command. The profile,
parent digest, runtime overlays, quantization, attention backend, DCP topology,
KV calibration, scheduler, and eager execution mode are one qualified contract.