Runtime: Osaurus 0.25.26 or newer.

OsaurusAI/d3-JANG_6M

d3 in 23.6 GB: the same decision as the full-precision model on 99.35 % of 2,000 requests of tasksource/tasksource-jev-typed-decisions test.

A JANG bundle of vllm-sr/d3, the 27B 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: 23.59 GB of model shards, 7.23 bits per weight over the 26.09B source parameters (BF16: 52.2 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 (41 benchmarks, 1,610 questions)	1,610	99.38 %	0.0003	78.9 → 78.8 %
LocalLLaMA/typed-decisions test (800 questions, soft gold)	800	99.50 %	0.0001	73.4 → 73.1 %
held-out decisions over our own texts, images and videos (487 questions)	487	98.77 %	0.0016	—
2,000 requests of tasksource/tasksource-jev-typed-decisions test (the split the vendor scored its quantized packages on)	2,000	99.35 %	0.0002	84.2 → 84.0 %

All three bundles

bundle	size	decisions (tsj, 2,000)	decisions (suite)	mean KL (suite)
d3-JANG_6M	23.6 GB	99.35 %	99.38 %	0.0003
d3-JANG_4M	19.3 GB	98.45 %	98.07 %	0.0024
d3-JANGH2	14.5 GB	93.45 %	93.17 %	0.0345

For scale, the vendor's own quantized packages (NVIDIA / AMD formats), as they report them on all 14,031 test requests of the same dataset (their request mapping, their hardware):

vendor package	decisions	memory
FP8	99.08 %	31.9 GB
W4A16 GPTQ (group 128)	97.98 %	16.4 GB
W4A16 AWQ (group 128)	97.76 %	16.5 GB
NVFP4	96.64 %	17.2 GB

How it was made

Calibration ran d3'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 (built on Qwen/Qwen3.8-27B). NOTICE and LICENSE are carried over. Quantized by Osaurus (eric@osaurus.ai).