Runtime: Osaurus 0.25.26 or newer.

OsaurusAI/d3-flash-JANG_6M

d3-flash in 7.8 GB: the same decision as the full-precision model on 99.40 % of 2,000 requests of tasksource/tasksource-jev-typed-decisions test.

A JANG bundle of vllm-sr/d3-flash, the 9B multimodal foundation decision model of vLLM Semantic Router's Decision 3.0. Give it a state (text or JSON, optionally with images and videos) and typed questions: pick one of several options, yes or no, or a score on an ordered scale. It answers every question with a probability per option in one forward pass and generates no text.

• Layout: MLP (gate / up / down) affine 6-bit; attention, linear attention, embeddings and vision tower affine 8-bit. The 255-way answer head (decision/readout.safetensors) is the original FP32 tensor, byte for byte; the linear-attention gate inputs stay BF16.
• Size: 7.84 GB of model shards, 7.48 bits per weight over the 8.39B source parameters (BF16: 16.8 GB).

Agreement with the full-precision model

Each question is answered by this bundle and by the BF16 model; decisions counts the same top option (or the same side of 0.5 for yes / no), KL is measured on each question's option probabilities. None of these questions were used for calibration.

test set	questions	decisions	mean KL ↓	gold accuracy, BF16 → this bundle
Decision Index 0.3 public suite, 466 of its questions across the 41 benchmarks	466	99.14 %	0.0004	71.7 → 71.9 %
LocalLLaMA/typed-decisions test (800 questions, soft gold)	800	99.62 %	0.0003	69.6 → 70.0 %
held-out decisions over our own texts, images and videos (487 questions)	487	98.77 %	0.0011	—
2,000 requests of tasksource/tasksource-jev-typed-decisions test (the split the vendor scored its quantized packages on)	2,000	99.40 %	0.0003	81.2 → 81.1 %

All three bundles

bundle	size	decisions (tsj, 2,000)	decisions (suite)	mean KL (suite)
d3-flash-JANG_6M	7.8 GB	99.40 %	99.14 %	0.0004
d3-flash-JANG_4M	6.6 GB	97.65 %	97.00 %	0.0038
d3-flash-JANGH2	5.3 GB	92.65 %	89.27 %	0.0608

How it was made

Calibration ran d3-flash's own decision forward (bidirectional full attention, the decision prompt format, images and videos) over 2,778 decision prompts and 1.81M tokens: routing questions over real requests, tool selection over real tool lists with up to 255 options, agent and terminal traces, code, security, long documents, typed decisions, images and screen videos. Activation-aware scales (AWQ) are folded into the producing norms and projections, and every quantized MLP is rounded with GPTQ on Hessians from that forward, each module kept only where it beats plain rounding.

License

Apache-2.0, as the base model vllm-sr/d3-flash (built on Qwen/Qwen3.5-9B). NOTICE and LICENSE are carried over. Quantized by Osaurus (eric@osaurus.ai).