Up to date
juspay Apache 2.0 out 34 days ago
35.1B parameters (8 of 256 experts per token), text in, text out, 256K tokens of context.
Verified 2026-10-25 Rank #9075 in Decision models · score 41 method v1 Sources 3 Downloads 30d 1.9k · #579 of 619 tracked
Picked by rule from 1 tracked piece: the lab's own post, the most shared, a field report, the latest.
Tiny-TPU is a custom accelerator that re-invents Google's TPU logic by scaling down the original 256x256 systolic array to a 2x2 grid.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 7 Spaces.
Top Spaces running it: multimodalart/jev-decision-index ♥ 802
Weights only, an estimate: leave room for the KV cache. Method
About 21 GB at INT4, 42 GB at FP8, 84 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-09-21), amber triangles are 1 dated article (each links to it), grey ticks are repo commits.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported Brier mean | 0.18 | All public | model author | 2026-10-07 |
| author-reported ECE | 0.07 | All public | model author | 2026-10-07 |
| author-reported JEVBench | 1 | Easy | model author | 2026-10-07 |
| author-reported JEVBench | 0.96 | Original | model author | 2026-10-07 |
| author-reported JEVBench | 0.82 | Hard public | model author | 2026-10-07 |
| author-reported JEVBench | 0.9 | All public | model author | 2026-10-07 |
| author-reported Operational success | 1 | All public | model author | 2026-10-07 |
| author-reported coverage | 1 | All public | model author | 2026-10-07 |
| author-reported macro accuracy | 0.91 | All public | model author | 2026-10-07 |
| author-reported schema validity | 1 | All public | model author | 2026-10-07 |
| author-reported strict schema validity | 1 | All public | model author | 2026-10-07 |
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in Decision models · LLMs
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | Qwen3_5MoeForConditionalGeneration |
| Model type | qwen3_5_moe |
| Dtype | bfloat16 |
| Params by dtype | BF16: 35,107,181,936 |
| Size band | XL |
| MoE | yes |
| Experts | 256 |
| Experts per token | 8 |
| Context length | 262,144 (config.json max_position_embeddings) |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Qwen3_5MoeForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3_5MoeForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for xor: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 84 GB · Multi-GPU node (<640 GB) |
|---|---|
| FP8 (1 byte/param) | 42 GB · Single high-end GPU (<80 GB) |
| INT4 (0.5 bytes/param) | 21 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.
| How it was made | full fine-tune (whole weights published) |
|---|---|
| Lineage bucket | F - fine-tune of an open model (public/synthetic data) |
| Lineage evidence | board file inspection (adapter/head/weight files) + HF base_model field |
| Base model | Qwen/Qwen3.6-35B-A3B |
| DI 0.2.1 score | 41.48 (independent (DI 0.2.1 board)) |
| ECE (board) | 0.0149 |
| Median latency | 139.9 ms (DI board, 1x RTX PRO 6000, single process) |
| Hosted price (input) | self-hosted/M |
| Source | https://huggingface.co/juspay/xor |
Base model: Qwen/Qwen3.6-35B-A3B
| Date | Commit | SHA |
|---|---|---|
| 2026-09-28 | Release Xor 1.2 | a286a46 |
| 2026-09-27 | Update README.md | 8d2d708 |
| 2026-09-26 | Clarify media limits in model card | ace46ec |
| 2026-09-26 | Release Xor 1.1 | a21c92d |
| 2026-09-23 | Rename JevOne to Xor | 679decd |
| 2026-09-23 | Document multimodal serving support | 8a03d38 |
| 2026-09-23 | Update multimodal serving wrapper | be42456 |
| 2026-09-22 | Update model card with post-training and benchmark details | 362f60b |
| 2026-09-21 | Update serving bundle checksum | 4de0db7 |
| 2026-09-21 | Fix host port publishing | 6f1839d |
| 2026-09-21 | Remove superseded jev-trained serving artifacts | fc03bb6 |
| 2026-09-21 | Rename model to JevOne | bdac774 |
| 2026-09-21 | Add JevOne serving checksum | a247e96 |
| 2026-09-21 | Add JevOne serving bundle | 4210cec |
| 2026-09-21 | Upload serving/jev-trained-serving.tar.gz.sha256 with huggingface_hub | 93f1a30 |
| 2026-09-21 | Upload THIRD_PARTY_NOTICES.md with huggingface_hub | 0d4be71 |
| 2026-09-21 | Upload LICENSE with huggingface_hub | 7ff7442 |
| 2026-09-21 | Update README.md | 7cd68c6 |
| 2026-09-21 | Upload README.md with huggingface_hub | 3130c43 |
| 2026-09-21 | Upload checksums.sha256 with huggingface_hub | aab97aa |
| 2026-09-21 | Upload serving/jev-trained-serving.tar.gz with huggingface_hub | 1f5fc56 |
| 2026-09-21 | Upload folder using huggingface_hub | afa42a3 |
| 2026-09-21 | initial commit | def428d |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:43 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:17 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
