AX-Muse-Glimmer-30B-MLX-AXQ-6bit

Research development artifact. Base affine6 packing with higher-precision protection, not MXFP6 or uniformly six-bit storage. No quality certificate is issued.

Property	Value
Source	meta-models/Muse-Glimmer-30B
Immutable source revision	a4e59da52a7bc87ae7251dd5545c0dd437c44b68
Complete weight storage	30,074,430,738 bytes
Complete measured BPW	8.080010
Manifest budget class	8bit
MTP	No MTP payload is advertised.

The precision distribution and exact per-module settings are recorded in axquant_manifest.json, axquant_plan.json, and config.json. Complete BPW includes protected tensors and storage overhead. File size does not establish peak runtime memory requirements. Converted from the pinned original source with existing precision protection floors.

Observed runtime evidence

The exact runtime payload passed the following scoped check on Metal. The receipt binds its runtime input files. This is a functional smoke, not a quality benchmark, MTEB evaluation, coding evaluation, long-context test or MTP acceleration result.

{
  "task": "text",
  "runtime": "mlx-lm",
  "version": "0.32.2",
  "mlx_version": "0.32.4",
  "cases": [
    {
      "expected": "42",
      "final_answer": "42",
      "output": "to=self<|message|>What is 19 plus 23? Reply with only the number.\n\nReply with only the number.\n\n19+23=42\n\nOutput only number.<|eom|><|start|>assistant to=user<|message|>42",
      "passed": true,
      "prompt": "What is 19 plus 23? Reply with only the number."
    },
    {
      "expected": "READY",
      "final_answer": "READY",
      "output": "to=self<|message|>Reply with exactly the word READY.\n\nWe must reply with exactly the word READY. Probably just READY. No punctuation? Exactly the word READY. Probably uppercase. Output exactly READY.\n\nEnsure no extra whitespace? Probably newline okay. Provide exactly word.\n\n<|eom|><|start|>assistant to=user<|message|>READY",
      "passed": true,
      "prompt": "Reply with exactly the word READY."
    }
  ]
}

Runtime	Evidence for this artifact
MLX-LM 0.32.2	Passed the task-specific smoke above
MLX-VLM	Not separately tested
AX Engine	Not tested; architecture metadata is not execution evidence
omlx	unsupported (static only)
mtplx	unsupported (static only)

Unexercised modalities and MTP behavior remain unverified. Peer runtime-specific exports require distinct variants and their own testing. See the manifest-bound static report.

Use and provenance

python -m pip install -U huggingface_hub mlx-vlm
hf download AutomatosX/AX-Muse-Glimmer-30B-MLX-AXQ-6bit --local-dir ./AX-Muse-Glimmer-30B-MLX-AXQ-6bit

Pin the immutable Hub revision. Use the architecture-aware runtime above and preserve the source processor/template. The tested MLX-LM port is text-only and discards vision weights; image inference is unsupported by that path. Preserve ATEM channel framing and read only the explicit to=user final message. The smoke checks that message and retains the raw response; reasoning-only text does not pass. The source license and notices are retained where supplied. The upstream repository and its license govern downstream use; no additional commercial grant or upstream benchmark/certification transfer is implied.