CommunityJev Models & Reimplementations
Bit Jev 2B Distilled
A CPU-first BitNet student: Microsoft's BitNet b1.58 2B quantized to an I2_S GGUF with a float32 pointer head, distilled from Kev 9B logits and run through the bit-jev Python package with no separate open-weight licence.
- Published by
- Community
- Author
- Zeaulo
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- I2_S GGUF of Microsoft's BitNet b1.58 2B with a float32 pointer head, about 1.19 GB downloaded on first use.
- Pointer head scores candidates directly, returning a selection, probabilities, or ordered levels without generating answer tokens.
- Training: 3,144 steps over 2 epochs at BF16 and learning rate 2e-5, distilled from Kev 9B candidate logits at temperature 2.0.
- Card case: 1,954.46 ms average on an EPYC 9654 at 32 threads, 1,625.88 MiB peak RSS, for 703 tokens and 77 candidates.
- No separate open-weight licence; a Yelp data-licence request for derivative weights was unanswered as of 2026-09-28.
Watch out
The checkpoint has no separate open-weight licence: the project code's Apache-2.0, the Microsoft base-model licence, and the Kev source licence apply to their own works, and a Yelp data-licence request covering derivative weights was unanswered as of 2026-09-28. Published results are single-question dev-case latency and memory only, not accuracy.
Questions about this entry
Short answers drawn from the linked project or page.
A structured-decision student model built from Microsoft's BitNet b1.58 2B: an I2_S GGUF backbone of about 1.19 GB plus a float32 pointer head that scores candidate options directly for choice, noul, and score questions, without emitting answer tokens.
Install bit-jev 0.13.13 via pip and call BitJev.from_pretrained(device='cpu', threads=8), or use device=gpu for Vulkan. On Windows x64 with AVX2 the wheel carries native CPU and Vulkan runners; other systems and CUDA build from fixed source with Git, CMake 3.28+, and a C++17 compiler.
python -m bit_jev.demo opens local Choice, Noul, and Score pages where Choice and Score accept 2 to 4 options. The first submission downloads the model, and --source modelscope or --device gpu switch the download source and execution device.
backbone-i2_s.gguf is 1,187,288,192 bytes, about 1.106 GiB, head.f32 is 5,244,948 bytes, pointer.json is 770 bytes, and tokenizer/config files total about 17.2 MB. SHA256SUMS.json lists every size and hash except its own.
On a fixed 703-token, 77-candidate dev question: 1,954.46 ms average on an EPYC 9654 at 32 threads with 1,625.88 MiB peak RSS, and 86.56 ms on another machine's RTX 5090 FP16 with 4,935.53 MiB GPU peak allocation.
A local decision-v7 multi-source decision set that includes Yelp review records, with teacher supervision from Kev 9B candidate logits. The repo includes no Yelp raw records, and a licence request to Yelp covering derivative weights was unanswered as of 2026-09-28.
No verifiable ones. The card says there is no public held-out report of accuracy, Brier, NLL, or ECE, that the Microsoft BitNet base-model benchmarks are separate measurements, and that choice results depend on option wording, order, and training distribution.
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
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 NewsHacker News: A one-token logprob wrapper that also classifies webcam frames.
allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker NewsMore Jev resources like this
AutoTrust's combined System 1 and System 2 bundle on Gemma-4-26B-A4B-it: Decision Index 0.2.1 of 58.05, a 24-slot head with noul, choice and score, and the base chat model left bit-identical.
Jev Models & Reimplementations#community#open-models#local
Community
Reza2kn's llama.cpp Q4_K_M build of Jev-Omni: a 6.87 GiB multimodal decision backbone plus projector and FP32 head, measured at 8.98 GiB VRAM on an RTX 5080 Laptop GPU.
Jev Models & Reimplementations#community#open-models#local
Community
Ruiruiz30's Apple Silicon MLX conversion of Jev-Omni: 4-bit affine language weights on a Mac mini M4 16GB, about 7 GB peak Metal memory and roughly 963 ms text decisions, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
GGUF builds of the JevK5 family for llama.cpp: Q8_0 is effectively lossless (229 of 231 answers match the 4B v0.3 bf16), and every published file clears the card's 95% agreement gate.
Jev Models & Reimplementations#community#open-models#local
Community
Mithril's open, Jev-shaped typed-decision model on DeBERTa-v3-large: choice up to 255 options, score and noul from one forward pass, 0.854 in-domain with ECE 0.022, trained on public gold labels only.
Jev Models & Reimplementations#community#open-models#local
Community
akhilaaa3's merged multimodal decision classifier on Gemma 4 12B IT: probabilities for text, image, audio or video options, 87.57% on DecisionBench Medium and MMAU 63.10%, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
Training recipes, looped architectures, and language ports: what the open side of the typed-decision class shipped this month, and how to pick by runtime and calibration instead of the leaderboard.
Jev Models & Reimplementations#open-models#training#local
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
From the Jev community
Posts from builders shipping with Jev right now.