model_name stringclasses 1
value | model_repo stringclasses 1
value | public_suite_edition stringclasses 1
value | vision dict | text dict | latency_scope stringclasses 1
value | private_scores null | leaderboard_acceptance stringclasses 1
value |
|---|---|---|---|---|---|---|---|
VisionJEV 1.0.0 (4B) | https://huggingface.co/tinnel123/VisionJEV-4B | 0.3 | {
"index": 73.03369467702923,
"complete": true,
"requests": 12210,
"runtime_errors": 0,
"warm_mean_ms": 247.27594685617964,
"warm_sum_ms": 3019239.3111139536,
"mean_executed_layers": 32,
"official_score_sha256": "dd2f0cd170f110903f2af01271d1e8f2095fbed7dc9ca436e47ee4a7a949af2c"
} | {
"index": 57.78,
"raw_index": 67.71,
"complete": true,
"requests": 140178,
"runtime_errors": 0,
"warm_mean_ms": 86.70113558053053,
"warm_sum_ms": 12153591.78340761,
"mean_executed_layers": 32,
"official_score_sha256": "82f0c53a2d34293e22d1ff58e35692ff3193dbc4e9e78a2bde34cc39f704c150"
} | submitter MLU measurements; not maintainer RTX PRO 6000 certification | null | pending |
VisionJEV 1.0.0 (4B): public evaluation results
Complete public Vision 0.3 and Text 0.3 results for VisionJEV 1.0.0 (4B).
| Track | Public index | Completed requests | Warm mean (ms) | Mean executed layers |
|---|---|---|---|---|
| Vision 0.3 | 73.03369467702923 | 12,210 | 247.27594685617964 | 32 |
| Text 0.3 | 57.78 | 140,178 | 86.70113558053053 | 32 |
Both score files declare complete: true, with zero runtime errors or missing requests. Text raw index: 67.71. Private scores and leaderboard acceptance are pending.
| Track | Official score file | Prediction records | Runtime settings |
|---|---|---|---|
| Vision | vision-scores.json | results.jsonl | environment.json |
| Text | scores.json | results.jsonl | environment.json |
Warm mean is the sum of second, device-synchronized forward times divided by request count, excluding model loading. Runs use eight independent MLU590-M9DG workers, BF16 inference and all 32 layers. Original score timing fields include the two-call wrapper; warm timing is reported separately. Standardized RTX PRO 6000 latency is pending.
Reproduction instructions and exact settings · Source revisions · Capacity and reproduction limits
Vision benchmark results
Accuracy and chance-corrected skill are percentages. The public index is the weighted mean of clipped skills, using the official weights.
| Benchmark | Requests | Accuracy (%) | Skill (%) | Weight |
|---|---|---|---|---|
| CV-Bench | 2,038 | 86.5064 | 77.7397 | 1 |
| BLINK | 961 | 74.4017 | 60.3705 | 1 |
| RealWorldQA | 589 | 78.2683 | 64.4855 | 1 |
| CharXiv | 532 | 80.6391 | 74.1855 | 1 |
| InfographicVQA | 244 | 95.9016 | 94.5355 | 1 |
| Mind2Web | 973 | 94.2446 | 92.3261 | 1 |
| Winoground | 800 | 90.8750 | 81.7500 | 1 |
| KIE (CORD+FUNSD) | 2,465 | 95.6187 | 92.7868 | 1 |
| Moderation (Hateful Memes) | 1,213 | 77.7411 | 55.4823 | 1 |
| R-Bench-M | 665 | 34.7368 | 21.6724 | 0.5 |
| MMMU-Pro vision | 1,730 | 39.2486 | 30.3215 | 0.25 |
Text benchmark results
Native metric scores and index skills below are shown on a 0–100 scale; the native metric is not always accuracy. Official JSON files retain the source values and detailed tracks.
| Benchmark | Requests | Native metric | Metric × 100 | Index skill × 100 | Included in index |
|---|---|---|---|---|---|
| BFCL | 1,694 | case exact accuracy | 79.75 | 72.67 | Yes |
| ToolRet | 685 | nDCG@10 | 59.84 | 53.62 | Yes |
| API-Bank | 508 | accuracy | 86.22 | 85.95 | Yes |
| BANKING77 | 3,080 | macro-F1 | 90.08 | 89.95 | Yes |
| CLINC150+OOS | 5,500 | macro-F1 | 92.95 | 92.91 | Yes |
| RouterBench | 10,000 | selected quality (quality objective) | 79.18 | 51.18 | Display only |
| Home appliance simulator | 88 | case exact accuracy | 72.73 | 72.73 | Yes |
| SGD/SGD-X | 2,500 | macro-F1 | 35.58 | 0.00 | Display only |
| ContractNLI | 123 | macro-F1 | 82.86 | 75.21 | Yes |
| ANLI | 3,200 | macro-F1 | 68.50 | 52.82 | Yes |
| BPoMP | 5,000 | accuracy | 91.28 | 83.19 | Yes |
| Humicroedit | 2,628 | accuracy | 61.64 | 23.29 | Yes |
| POP909-CL | 2,000 | accuracy | 27.50 | 25.91 | Yes |
| cfcolor | 5,000 | accuracy | 66.36 | 38.48 | Yes |
| MMLU | 14,033 | accuracy | 79.80 | 73.07 | Display only |
| GPQA Diamond | 196 | accuracy | 46.43 | 28.57 | Yes |
| ARC-Easy | 2,376 | accuracy | 99.12 | 98.83 | Display only |
| ARC-Challenge | 1,172 | accuracy | 95.73 | 94.31 | Display only |
| WinoGrande | 1,267 | accuracy | 88.32 | 76.64 | Yes |
| HellaSwag | 10,042 | accuracy | 93.71 | 91.61 | Yes |
| ChessBench | 5,000 | accuracy | 21.22 | 14.20 | Yes |
| MuSR | 752 | accuracy | 60.24 | 36.79 | Yes |
| SATA-Bench | 1,650 | case exact accuracy | 22.55 | 21.52 | Yes |
| SimpleBench | 10 | accuracy | 30.00 | 16.00 | Display only |
| BRIGHT | 220 | nDCG@10 | 33.61 | 24.90 | Yes |
| Amazon ESCI | 5,000 | macro-F1 | 53.82 | 42.07 | Yes |
| ACOS | 1,565 | per-review F1 | 11.42 | 8.59 | Yes |
| FinEntity | 979 | macro-F1 | 91.87 | 88.04 | Yes |
| iSarcasmEval | 4,600 | Sarcasm F1 · track A, English | 55.70 | 43.01 | Yes |
| VAST | 3,006 | macro-F1 | 77.39 | 66.08 | Yes |
| NLI4CT | 5,500 | macro-F1 | 82.18 | 65.33 | Yes |
| CRUXEval | 570 | accuracy | 61.93 | 39.60 | Yes |
| CLadder | 5,000 | accuracy | 84.70 | 69.40 | Yes |
| HLE | 501 | accuracy | 9.98 | 0.00 | Yes |
| Habermas Machine | 1,676 | accuracy | 60.62 | 42.85 | Yes |
| GSM8K | 2,638 | accuracy | 67.40 | 61.02 | Yes |
| PhishNChips phishing decisions | 2,000 | accuracy | 73.85 | 47.70 | Yes |
| MMLU-Pro | 12,032 | accuracy | 59.44 | 54.38 | Yes |
| BBH fixed-option tasks | 5,507 | accuracy | 72.87 | 60.68 | Yes |
| RAGTruth response-level hallucination | 2,700 | F1 on hallucinated class | 82.28 | 63.26 | Yes |
| HoVer claim verification | 4,000 | accuracy | 86.78 | 73.56 | Yes |
| When2Call MCQ | 3,652 | accuracy | 89.21 | 85.61 | Yes |
| New Yorker caption matching | 528 | accuracy | 74.62 | 68.27 | Yes |
Benchmark questions and images are not included. See the official benchmark terms and bundled license notices.
- Downloads last month
- 228