alt.freeweights.vision / Decision-2.0-Kai-0.6B-ONNX

From:

onnx-community <huggingface.co/onnx-community>Newsgroups: alt.freeweights.visionSubject:

Decision-2.0-Kai-0.6B-ONNX

Date: 2026-11-15Commit: 13ab550Signed: 2DC28CC7B26A283B ✓

Text classification model. Onnx weights, 623 MB in all. Published by onnx-community under Apache-2.0.

vision

623MBDecision-2.0-Kai-0.6B-ONNXonnxpermissive

FactsParamsnot statedSize623 MBFormatsonnxTasktext classificationLicenseApache-2.0From sourcepermissive

$ fw get onnx-community/Decision-2.0-Kai-0.6B-ONNX
# manifest sha256 5124e7e34d35… · 623 MB · 17 files
$ fw verify ./Decision-2.0-Kai-0.6B-ONNX onnx-community/Decision-2.0-Kai-0.6B-ONNX

From sourceHosted (signed)onnx-community/Decision-2.0-Kai-0.6B-ONNXLast check: 2026-11-15 05:20 UTC

Run at home Get help

Get help: custom fine-tune or private deploy. We review your setup before we quote.

Quantized of Decision-2.0-Kai-0.6B.

Works with

Run2 Convert3 Scan1

Runs with

Ranked by fit to this model (onnx), then by GitHub stars.

Ultralytics OpenVINO

Convert with

ONNX

Optimum

Olive

Scan with

ModelScan

Overview Model card Files Use it License Provenance Model tree

Overview

Size: 623 MB
Weights: 612 MB (ONNX)
Parameters: not stated
Formats: onnx
Task: text classification
License: Apache-2.0 (permissive)
Files: 17, each with a sha256
Publisher: onnx-community
Added: 2026-11-15
Section: alt.freeweights.vision

Lineage

Original release from the publisher.

Base model: vllm-sr/Decision-2.0-Kai-0.6B

Where these bytes come from

Publisher at the pinned commit→You

Sources are tried in order. A file counts only when its size and sha256 match the manifest. Sources are checked when you ask for them, not on a timer.

Sources

Badge	Host	Last check	Result
From source	huggingface.co	2026-11-15 05:20 UTC	passed

Model card

From the publisher at commit 13ab550. Shown as plain text with links; images open at the publisher.

Decision-2.0-Kai-0.6B (ONNX)#

ONNX conversion of vllm-sr/Decision-2.0-Kai-0.6B, a Decision 2.0 model from vLLM Semantic Router: give it a state and typed questions (choice, noul, score) and it returns a probability for every option. For the model, its training and evaluation, see the original card.

It runs in the browser on WebGPU with open-jev (model: "decision2-kai-0.6b"), or with ONNX Runtime anywhere.

How it was made#

Decision 2.0 is a full fine-tune of Qwen3-0.6B-Base with a candidate head on the final hidden states. Qwen3 is a plain transformer, so the backbone and the head were exported together with torch.onnx (fp32, eager attention), then weight-quantized with ONNX Runtime's MatMulNBits quantizer (embeddings 4-bit, matmuls 8-bit).

input_ids [B, L], attention_mask [B, L], answer_pos [B], option_pos [B, K]  ->  logits [B, K]

Build one prompt per question as the package's decision_model.segments / encode do (Context: … Task type: … Question: … Options:, one <option> segment per option holding {"description": …, "key": …} as sorted, compact JSON, then the Decision: suffix), tokenizing each segment separately; option_pos is the last token of each option segment and answer_pos the last token. Add decision2.score_bias[L] from config.json to the logits of an L-level score question (Kai ships offsets for 5 levels), then softmax (temperature 1).

Scripts and reference cases are in conversion/.

Files#

dtype (Transformers.js)	File	Weights	Size
`q8` (default)	`onnx/model_quantized.onnx`	8-bit block-32 weights, 4-bit embeddings, fp32 activations	0.61 GB

q8 here is 8-bit block quantization (MatMulNBits, bits=8), not the usual dynamic int8 export; it uses the q8 slot because Transformers.js has no q8f16.

Checks#

Against the packaged runtime (decision2.api.Decision2, exact path, CPU fp32), on 11 questions over 6 states: largest difference in any option's probability and how many top answers changed.

Runtime	Max difference	Top answer changed
`q8`, ONNX Runtime CPU (fixtures)	0.017	0 of 11
`q8`, WebGPU (Chrome, M3 Pro)	0.017	0 of 11
`q8` on 30 typed-decisions questions	0.013	2 of 30 (near-ties)
fp32 export	< 0.0001	0 of 11

open-jev in Node reproduces the Python run of the same graph exactly (same token ids, option positions and score offsets).

In the browser#

On a MacBook Pro (M3 Pro, Chrome, WebGPU), on the 400 states × 5 questions of typed-decisions (test split; top answer = annotator label, as in the open-jev demo's benchmark):

dtype	choice	score	noul	overall	median per state
`q8`	258/600	330/800	383/600	48.6%	1.5 s

License#

Apache-2.0, as the original model.

Files

17 files · 623 MBWeights 612 MB (ONNX). Sizes and sha256 from the signed manifest.

Hide all 17 files with their sha256

File	Size	sha256
.gitattributes	1.64 KB	9c896fa7185b550d0a0a3c1e87569428c643155ce7ee6d232273c6f06e9fa88e
README.md	3.51 KB	e8233a3c7012cd2a6b910bd6bc3dfca631841adb7952a454b5886ce15675f2b8
config.json	885 B	7fdd02f449566a4f726ddbc8e3b4623322f948a172d9867f0353ad1dbcdae8c0
conversion/bench_check.py	2.60 KB	f2bd4976c8aabb833334de88cdacb1191d8c136c3150805a343e6893105478ed
conversion/build.py	12.4 KB	b22a884567cd7d037fbd9375919ceea3bccfa00ffbbc353229232bb3398cb0ce
conversion/export_qwen3.py	3.60 KB	7336b9708f44fa235487bb9d3d814ad943b356676053c7f6d44abb91dd2cc9c6
conversion/fixtures/reference-Kai-0.6B.json	20.5 KB	9000ab86103ed4b045fa2a93190ae361c1c1973acad16a229cf28800bc60ce8a
conversion/fixtures/requests.json	2.31 KB	b337132b79990533d1e7d6f02a64d117f90eb0e51a5d9e3653d3bcf8cf535dd8
conversion/make_repo.py	2.59 KB	e1babb3ac57e16f566b4b2e46825ab2436568c1a38fb9900889eec8564289baf
conversion/parity.py	1.99 KB	2777e07df9c4c8a71cfcd1a86b7d777e1bd8d33eecf23830374da52753988172
conversion/quantize_qwen3.py	1.31 KB	ce9791562b58fc49fe7380f781559032965aa0a5b74557951d008c75b079d4bc
conversion/qwen3_5_opt_mapping.json	78.9 KB	67fae39e0923e3e097c43b5871f8127986581d1f13d8454ff788a21a44ef86a3
conversion/reference.py	2.15 KB	6938846e8ea2cf852d8191d739833f8b73b499096cd6dedcbf314aa875df983d
onnx/model_quantized.onnx	1.52 MB	a509e10307478ca829e2aa90142066a8ed9a5a0dd867add9ed9a2039e4770fde
onnx/model_quantized.onnx_data	610 MB	5e4cae16253eb2ad94bdb5e82296c9dd55c3dd961d42d5b18542ff765b67e4df
tokenizer.json	11.4 MB	be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
tokenizer_config.json	695 B	cbea7bca9904d56693d2226b00fd564e3be053609d1e50f8f1885510f0f6e790

VersionsEach time the pinned upstream commit or file list changed in this repo. Newest first.

Upstream commit	Date	Contents
13ab55013a pinned now	2026-11-15 05:20 UTC	17 files · 623 MB

Use it

$ fw get onnx-community/Decision-2.0-Kai-0.6B-ONNX
# manifest sha256 5124e7e34d35… · 623 MB · 17 files
$ fw verify ./Decision-2.0-Kai-0.6B-ONNX onnx-community/Decision-2.0-Kai-0.6B-ONNX

fw.pyChecks the signature and every sha256, tries each source in order, resumes.

$ python3 fw.py get https://freeweights.allternit.com/manifests/onnx-community/Decision-2.0-Kai-0.6B-ONNX.json --dir ./Decision-2.0-Kai-0.6B-ONNX
$ python3 fw.py verify ./Decision-2.0-Kai-0.6B-ONNX ./onnx-community__Decision-2.0-Kai-0.6B-ONNX.json

hfThe Hugging Face CLI with one setting. Same repo id.

$ HF_ENDPOINT=https://freeweights.allternit.com hf download onnx-community/Decision-2.0-Kai-0.6B-ONNX --revision 13ab55013ab1828d5dea3b716ee4d6229faf3fde --local-dir ./Decision-2.0-Kai-0.6B-ONNX

Run at home

License

Name: Apache License 2.0
Id: Apache-2.0
Class: Permissive
Redistribution: yes
Listing: Hosted (signed)
Decided: License as declared by the publisher. No person read it. 2026-11-15
Text in: README.md
Text sha256: e8233a3c7012cd2a6b910bd6bc3dfca631841adb7952a454b5886ce15675f2b8
Terms gate: No extra terms beyond the license.

Read the license text

Licenses are checked against the license table. A custom license passes only after a person reads it.

Provenance

From:

onnx-community <huggingface.co/onnx-community>Newsgroups: alt.freeweights.visionSubject: Decision-2.0-Kai-0.6B-ONNXDate: 2026-11-15Commit: 13ab550Signed: 2DC28CC7B26A283B ✓

Upstream: onnx-community/Decision-2.0-Kai-0.6B-ONNX
Pinned commit: 13ab55013ab1828d5dea3b716ee4d6229faf3fde
Manifest: Decision-2.0-Kai-0.6B-ONNX.json · .minisig
Manifest sha256: 5124e7e34d3580542d847111c7e3e203ecb201ca9157d8d15e161c5afdd2fb41
Signature: minisign, Ed25519, key 2DC28CC7B26A283B
Created: 2026-11-15 05:20 UTC
Generator: freeweights-manifest/0.1.0
Takedowns: None.

How to verify

$ curl -O https://freeweights.allternit.com/manifests/onnx-community/Decision-2.0-Kai-0.6B-ONNX.json
$ curl -O https://freeweights.allternit.com/manifests/onnx-community/Decision-2.0-Kai-0.6B-ONNX.json.minisig
$ minisign -Vm Decision-2.0-Kai-0.6B-ONNX.json -P RWQ7KGqyx4zCLXtUFOqvfv8L3g+8fpKJ1FsuxPyOYMF+4KufbUxKq1Oq
$ shasum -a 256 <file>   # compare with the Files tab

A manifest whose signature fails must be rejected. A file whose size or sha256 does not match must be discarded. Verify a download

Model tree

Parent

vllm-sr/Decision-2.0-Kai-0.6B · this model is a quantization of it

No model in the catalog names this one as its parent.

Get it

Download

$ HF_ENDPOINT=https://freeweights.allternit.com hf download onnx-community/Decision-2.0-Kai-0.6B-ONNX

Open in Freeweights Studio

Pull, verify and run it in the Allternit app on Desktop, web or phone.

Signed manifest

minisign, key 2DC28CC7B26A283B. Decision-2.0-Kai-0.6B-ONNX.json. Provenance. Takedown history: none.

Source

huggingface.co/onnx-community/Decision-2.0-Kai-0.6B-ONNX

Related models

All of alt.freeweights.vision →

From: dima806 <huggingface.co/dima806>Subject: ai_vs_real_image_detection

From: legekka <huggingface.co/legekka>Subject: AI-Anime-Image-Detector-ViT

From:

onnx-community <huggingface.co/onnx-community>Subject: AI-Anime-Image-Detector-ViT-ONNX

From: AICodexLab <huggingface.co/AICodexLab>Subject: answerdotai-ModernBERT-base-ai-detector

From:

onnx-community <huggingface.co/onnx-community>Subject: answerdotai-ModernBERT-base-ai-detector-ONNX

From:

onnx-community <huggingface.co/onnx-community>Subject: bart-large-cnn-ONNX