Qwen3.8-27B-Fable-Distill-mlx-uniform-4bit

Model-card update:2026-09-14.

Status

Historical published candidate, not a current B/C pick. Its recorded recipe and tests are retained below; it has not received the current two-pick C84 runtime screen or a new predictor certification.

Conversion and attribution

MLX uniform 4-bit conversion of TeichAI/Qwen3.8-27B-Fable-Distill. Historical metadata reports 27B VLM parameters, 498 four-bit quantized modules, 4.0 effective bits/weight and 16.05 GB weights. This is a separate conversion from Qwen3.8-27B-Fable-Distill-OptiQ-4.5bpw-mixed; results and predictor compatibility must not be transferred merely because the fine-tuned parent is shared.

Packed low-bit tensor counts can understate true parameter counts in hub badges. Weight-file sizes, effective bits/weight and total MLX runtime peak are different measurements.

Historical comparison, 2026-09-03 (M21b)

At this recipe's t0.6 versus Qwen3.8-27B-Fable-Distill-OptiQ-4.5bpw-mixed at t0.5, both predictor OFF with thinking enabled, 50 tasks×3 draws per dataset: HumanEvalPlus strict 90.0% here versus 89.3% mixed; MBPPPlus 79.3% here versus 81.3% mixed. The paired mixed/uniform token ratios were 0.66[0.39,0.98] and 0.64[0.40,0.97]. The historical pooled comparison supported carrying the mixed recipe; this is a comparison of complete recipe+tune pairs, not an isolated precision effect. Tail variability matters.

No external MTP predictor is certified for this exact fine-tuned uniform conversion. The current C84 results for Qwen3.8-27B-Fable-Distill-OptiQ-4.5bpw-mixed or Qwen3.8-27B-mlx-uniform-4bit cannot be transferred here. Vision packaging does not itself establish image quality; no new vision gate or runtime smoke is claimed in this refresh.

Predictor

Predictor OFF: no external MTP recommendation is configured or certified for this exact recipe. A bundled head or MTP tag alone is not evidence to enable speculation.

Recorded historical-candidate configuration, not a new recommendation

Save this as models.yaml in an environment with the compatible fork revisions described below. The IDs name the target and, where configured, its exact companion; weight quantization and attention-KV precision are separate settings.

manager_port: 8000
mlx_port: 8091
startup_timeout_seconds: 300
models:
- name: Qwen3.8-27B-Fable-Distill-mlx-uniform-4bit
  type: vision
  hf_path: caslca/Qwen3.8-27B-Fable-Distill-mlx-uniform-4bit
  max_kv_cache_size: 262144
  kv_prealloc_tokens: 262144
  kv_quant_scheme: turboquant
  kv_bits: 4
  quantized_kv_start: 0
  prefill_step_size: 512
  generation_defaults:
    temperature: 0.6
    top_p: 0.95
    top_k: 20
    min_p: 0.0
    presence_penalty: 0.0
    max_tokens: 102400
    thinking_budget: 81920
    enable_thinking: true
  on_demand: true

Start from the compatible installed fork environment without additional worker overrides. This command clears cache/predictor environment overrides, disables APC and selects two retained sessions with one resident model:

env -u APC_ENABLED -u KV_BITS -u KV_QUANT_SCHEME -u KV_GROUP_SIZE \
  -u KV_KEY_BITS -u KV_VALUE_BITS -u KV_KEY_SCHEME -u KV_VALUE_SCHEME \
  -u QUANTIZED_KV_START -u MLX_VLM_SESSION_SHRINK_ON_RETIRE \
  -u MLX_VLM_DRAFT_KIND -u MLX_VLM_DRAFT_MODEL -u MLX_VLM_DRAFT_BLOCK_SIZE \
  MLX_VLM_CACHE_SESSION_MAX=2 MLX_SERVE_CONFIG=models.yaml mlx-serve start

The model loads on demand. An OpenAI-compatible client sends requests to http://localhost:8000/v1 with model ID Qwen3.8-27B-Fable-Distill-mlx-uniform-4bit. Omitted sampling fields inherit generation_defaults; thinking is enabled. The output cap is 102400 and thinking budget 81920. Near the context limit the server resolves a smaller budget from remaining space; a configured 262144 cap does not imply 262144 input tokens plus another 102400 output tokens.

The full active KV floor equals the context cap; keep prefill 512. kv_bits: 0 means unquantized native16 attention KV, not unquantized model weights or a newly measured IEEE-FP16 dtype. TurboQuant is inactive when bits are 0 even if its scheme name remains configured. Idle retirement is explicitly true only on Qwen3.8-27B-Fable-Distill-OptiQ-4.5bpw-mixed; omitted cache_session_shrink preserves the worker default OFF only when MLX_VLM_SESSION_SHRINK_ON_RETIRE is absent, as in the command above. Suffix decoding stays OFF. These structural settings target the 64 GB evaluation machine, not arbitrary hardware.

Runtime and evidence scope

The final C84 environment uses MLX-VLM 522671c4bebc5ff492d465a1d0e6a14251f18260, MLX-Serve b632280709f771972bffbaf3231e996e8a89f4e8, and MLX/Metal 0.32.2. Those fork commits are published. Live validation of this exact final bundle covers the two current picks only; other cards retain historical model evidence and source-level compatible configuration, not a newly run model smoke. Older experiments used their recorded earlier runtime and cache mode.

Dates, original source revisions, per-axis observations and evidence-file hashes are recorded in the accompanying evaluation evidence alongside this card. The tables here are self-contained: they do not rely on unpublished stack commits or links to future GitHub main content. Speculative decoding can change BF16 outputs; favorable task-specific intervals do not imply byte-lossless inference. Historical and current measurements are not pooled across cache modes, predictor states, budgets, corpora or runtimes.

Artifact integrity audit, 2026-09-14

All 13 audited model and support files match the corresponding local artifact by full-byte cryptographic hashing, including weights and applicable configuration/tokenizer/vision files. README, license and repository-attribute files are tracked separately from inference artifacts. This publication changes the card and its dated evidence only; model artifacts remain unchanged. Per-file hashes, audited revision and scope.