Jev directory
CommunityJev Models & Reimplementations
StartLux-Decision 35B-A3B GGUF
mradermacher's static GGUF quants of startlux-models/StartLux-Decision-v1-35B-A3B, the typed-decision mixture-of-experts: thirteen files from Q2_K at 13.3 GB to Q8_0 at 37.9 GB, plus mmproj supplements for the multimodal base.
- Reading time
- 2 min
- Published by
- Community
- Author
- mradermacher
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Capabilities
- Self-hostedRuns locally
Highlights
- Static GGUF quants of startlux-models/StartLux-Decision-v1-35B-A3B by mradermacher, from x-f16 down to Q2_K.
- Thirteen files: Q2_K 13.3 GB, IQ4_XS 19.5 GB, Q4_K_S 20.5 GB and Q4_K_M 21.8 GB (both marked fast, recommended), up to Q8_0 37.9 GB.
- Two mmproj supplements cover image input for the multimodal base: mmproj-Q8_0 at 0.7 GB and mmproj-f16 at 1.0 GB.
- No weighted or imatrix quants are provided; the card says it followed static quantization only and invites requests via Community Discussion.
- cc-by-nc-4.0 licence, inherited from the StartLux-Decision-v1-35B-A3B base checkpoints.
Watch out
Card reviewed 2026-10-17; this quantization repository documents file sizes, quant types and the absence of imatrix quants, but reports no accuracy, calibration or latency measurements, so the decision-quality cost of the smaller files relative to startlux-models/StartLux-Decision-v1-35B-A3B is undocumented. The cc-by-nc-4.0 licence inherited from the base excludes commercial use.
Questions about this entry
Short answers drawn from the linked project or page.
mradermacher's collection of static GGUF quants of startlux-models/StartLux-Decision-v1-35B-A3B, a typed-decision mixture-of-experts model. The repository carries thirteen quant files plus two multi-modal supplements for the vision tower.
From smallest: Q2_K 13.3 GB, Q3_K_S 15.6 GB, Q3_K_M 17.3 GB (lower quality), Q3_K_L 18.7 GB, IQ4_XS 19.5 GB, Q4_K_S 20.5 GB, Q4_K_M 21.8 GB, Q5_K_S 24.7 GB, Q5_K_M 25.4 GB, Q6_K 29.3 GB and Q8_0 37.9 GB, with Q4_K_S and Q4_K_M marked fast and recommended.
It quantizes StartLux-Decision-v1-35B-A3B, which answers typed decision questions with a probability per option rather than generated text — the System One pattern Jev popularized. This card itself only documents the quantization artifacts, not the model's benchmarks.
The base model's state can include images, and the two mmproj files (Q8_0 at 0.7 GB, f16 at 1.0 GB) are the multi-modal supplement that covers image input. Text-only usage does not need them.
The card says weighted/imatrix quants seem not to be available from this quantizer at this time, and that requests can be made by opening a Community Discussion. Only static quants were produced.
cc-by-nc-4.0, inherited from the startlux-models base checkpoints, so the quants are restricted to non-commercial use. The card gives no other licensing detail or grants beyond that.
The card refers to TheBloke's README format for using GGUF files, including how to concatenate multi-part files, and notes the graph from ikawrakow and Artefact2's notes as quality references. GGUF files are run by llama.cpp-family clients rather than the transformers stack.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
X: Standard One SH (StandardThinking)
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
prithivMLmods' GGUF builds of autotrust/JEV-35B, the Qwen3.5-35B-A3B MoE decision model that returns calibrated choice, noul and score answers over text and images; files run from BF16 to Q3_K_L for llama.cpp.
Jev Models & Reimplementations#community#open-models#local
Community
A 4-bit W4A8 AWQ plus GPTQ build of AutoTrust's JEV-27B-VL from AI Outfitters that keeps the rank-32 LoRA decision head and a bf16 vision tower, served by vLLM on a single 24 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
GGUF quantizations of StartLux Labs' 2B Instant decision model for llama.cpp: Q8_0 (2.01 GB) and Q4_K_M (1.27 GB), plus a 668 MB F16 vision encoder, with Q8_0 scoring 23.7 DecisionPivot against 23.5 for BF16 on vLLM.
Jev Models & Reimplementations#community#open-models#local
Community
GGUF quantizations of StartLux Labs' 0.8B Instant decision model for llama.cpp: Q8_0 (0.81 GB, DecisionPivot 9.7) and Q4_K_M (0.53 GB, 8.7), plus a 205 MB F16 vision encoder, against 9.5 for BF16 on vLLM.
Jev Models & Reimplementations#community#open-models#local
Community
A small distilled sibling of Perplexity's pplx-decider-v1.2: a Qwen3.5-4B decider that reads a state plus typed choice, noul and score questions in one pass, keeping most of the 27B model's Decision Index.
Jev Models & Reimplementations#community#open-models#multimodal
Community
Perplexity's 27B decision model rebuilt on Qwen3.8-27B with a joint request format that answers a whole request's questions in one sequence, reporting 68.09 on Decision Index 0.3 and 77.43 on vision.
Jev Models & Reimplementations#community#open-models#multimodal
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.