Up to date
caiovicentino1 MIT out 32 days ago
27.8B parameters, text in, labels out, 256K tokens of context.
Verified 2026-10-25 Rank #0 in Decision models · score 0 method v1 Sources 3 Downloads 30d 401 · #610 of 619 tracked
Picked by rule from 1 tracked piece: the lab's own post, the most shared, a field report, the latest.
Eikos is an open family of calibrated, single-pass typed-decision models released under MIT for finance and trading.
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: 4 Spaces.
Top Spaces running it: multimodalart/jev-decision-index ♥ 802
Weights only, an estimate: leave room for the KV cache. Method
About 17 GB at INT4, 33 GB at FP8, 67 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-09-23), 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 AgentRewardBench | 78.9 | Precision | model author | 2026-10-07 |
| author-reported AgentRewardBench | 73.6 | Recall | model author | 2026-10-07 |
| author-reported AgentRewardBench | 76.1 | F1 | model author | 2026-10-07 |
| author-reported Compositional rules — same type / new domain / rulebooks | 95.6 / 94.7 / 95.0 | Validation against bf16 | model author | 2026-10-07 |
| author-reported Decision Index (DI) · Classify and make typed decisions | 55.46 | Index | model author | 2026-10-07 |
| author-reported Decision Index (DI) · Classify and make typed decisions | 39.9 | Knowledge & Reasoning | model author | 2026-10-07 |
| author-reported Decision Index (DI) · Classify and make typed decisions | 63.3 | Language | model author | 2026-10-07 |
| author-reported Decision Index (DI) · Classify and make typed decisions | 55.9 | Retrieval & Classification | model author | 2026-10-07 |
| author-reported Decision Index (DI) · Classify and make typed decisions | 74.4 | Tools & Automation | model author | 2026-10-07 |
| author-reported Decision Index (DI) · Classify and make typed decisions | 39.8 | Arts & Human Taste | model author | 2026-10-07 |
| author-reported DecisionBench — medium / hard | 88.7 / 78.8 | Validation against bf16 | model author | 2026-10-07 |
| author-reported ECE (lower is better) | 0.04 | Validation against bf16 | model author | 2026-10-07 |
| author-reported Finance (CUAD, sentiment, FinQA-judge) | 85.3 | Validation against bf16 | model author | 2026-10-07 |
| author-reported General battery (9 tasks) | 82.8 | Validation against bf16 | model author | 2026-10-07 |
| author-reported JevBench public — original / hard | 100.0 / 83.8 | Validation against bf16 | model author | 2026-10-07 |
| author-reported MME (yes/no) | 90.0% percent | Zero-shot | model author | 2026-10-07 |
| author-reported MMStar · Understand images and video | 70.7% percent | Zero-shot | model author | 2026-10-07 |
| author-reported SEED-Bench-2-Plus | 74.3% percent | Zero-shot | model author | 2026-10-07 |
| author-reported Same answer as bf16 (all / confident ≥0.9) | 98.8% / 100.0% percent | Validation against bf16 | model author | 2026-10-07 |
| author-reported Same answer as bf16: all / confident (≥0.9) | 97.6% / 100.0% percent | Zero-shot | model author | 2026-10-07 |
| author-reported ScreenSpot-Pro, no marks | 65.7% percent | Zero-shot | model author | 2026-10-07 |
| author-reported ScreenSpot-v2, 4×4 grid drawn | 60.0% percent | Zero-shot | model author | 2026-10-07 |
| author-reported ScreenSpot-v2, no marks | 71.3% percent | Zero-shot | model author | 2026-10-07 |
| author-reported Trade rules — seen / unseen | 85.6 / 86.9 | Validation against bf16 | model author | 2026-10-07 |
| author-reported ≥0.90 confidence: decides / error | 44.2% / 2.6% percent | Validation against bf16 | model author | 2026-10-07 |
This quantized build provides calibrated probabilities for typed decisions and supports image-text-to-text modalities.[1][2] The model is released under an MIT license and requires vLLM version 0.30.0 or higher.[3][4]
Generated 2026-10-07 15:19 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
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
| Pipeline tag | text-classification |
|---|---|
| Library | transformers |
| Architectures | Qwen3_5ForConditionalGeneration |
| Model type | qwen3_5 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 3,430,871,792 · F8_E4M3: 24,350,556,160 |
| Size band | L |
| MoE | no |
| Context length | 262,144 (config.json max_position_embeddings) |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Qwen3_5ForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3_5ForConditionalGeneration · 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 Eikos-27B-FP8: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 67 GB · Single high-end GPU (<80 GB) |
|---|---|
| FP8 (1 byte/param) | 33 GB · Single high-end GPU (<80 GB) |
| INT4 (0.5 bytes/param) | 17 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
| How it was made | LoRA/adapter fine-tune |
|---|---|
| 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/Qwen3.8-27B |
| DI 0.2.1 score | 53.13 (independent (DI 0.2.1 board)) |
| ECE (board) | 0.0574 |
| Median latency | 129.7 ms (DI board, 1x RTX PRO 6000, single process) |
| Hosted price (input) | self-hosted/M |
| Contamination notes | The whole FinEntity benchmark counts as wrong: its full test set is in the declared training data |
| Source | https://huggingface.co/caiovicentino1/Eikos-27B-FP8 |
Base model: caiovicentino1/Eikos-27B
| Date | Commit | SHA |
|---|---|---|
| 2026-10-01 | serve v1.3: images | 8b053df |
| 2026-09-27 | Card: Decision Index 0.2.1 result (55.46), with areas and links | bbdb023 |
| 2026-09-27 | Card: third-party benchmarks (AgentRewardBench, Jevals, JevBench setup check) | e40a92f |
| 2026-09-26 | serve_vllm.sh: --max-num-seqs 64 by default (MAX_NUM_SEQS), so the 27B builds start on one 80-96 GB GPU | bd884f6 |
| 2026-09-26 | serve v1.2: up to 588 options in one pass, per-model limit (max_one_pass in decision_config.json) | 90e43e8 |
| 2026-09-24 | serve.py: HTTP/1.1 keep-alive, so clients reuse the connection between calls (no change to any decision) | c58dd05 |
| 2026-09-24 | Add launch figures; final release tables | d2b146a |
| 2026-09-23 | Add files using upload-large-folder tool | da5f9a0 |
| 2026-09-23 | initial commit | 879473a |
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 21:44 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:19 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.
