Text Classification
jev-style
Safetensors
Transformers
qwen3_5_text
text-generation
decision-model
system-one
calibration
classification
long-context
multilingual
qwen3.5
on-device
llm-routing
guardrails
Instructions to use chaoliangUNSW/Jev-Style-0.8B-Decision-v3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- jev-style
How to use chaoliangUNSW/Jev-Style-0.8B-Decision-v3 with jev-style:
pip install "jev-style[torch]"
from jev_style import JevStyle, noul, choice js = JevStyle.from_pretrained("chaoliangUNSW/Jev-Style-0.8B-Decision-v3") out = js.decide("I was charged twice for one order.", { "billing": noul("This message is about billing."), "team": choice("Which team should handle it?", ["billing", "shipping", "tech"]), }) print(out["answers"]["team"]["choice"]) - Transformers
How to use chaoliangUNSW/Jev-Style-0.8B-Decision-v3 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="chaoliangUNSW/Jev-Style-0.8B-Decision-v3")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("chaoliangUNSW/Jev-Style-0.8B-Decision-v3") model = AutoModelForCausalLM.from_pretrained("chaoliangUNSW/Jev-Style-0.8B-Decision-v3", device_map="auto") - Notebooks
- Google Colab
- Kaggle
File size: 4,357 Bytes
656ca59 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 | {
"chart": "beyond_laya",
"unit": "percent (value x 100)",
"source": "runs/macjev/report_r2/scoreboard/scoreboard.json via hf_staging/v3_card/chart_data.json",
"delta_ci_method": "case-cluster bootstrap within suites, 2000 resamples (scoreboard n_boot)",
"rows": [
{
"task": "Banking77",
"metric": "accuracy, 77 intents",
"v3": 0.6825,
"best_laya": 0.4925,
"best_laya_checkpoint": "typed",
"delta_pts": 19.0,
"delta_ci95_pts": [
13.99,
24.0
],
"n": 400,
"v3_entry": "banking77.v3",
"laya_entry": "banking77.laya_best",
"scoreboard_field": "metrics['banking77']",
"protocol_label": "v3: held-out (banking77 never trained; CLINC150/HWU64 intents are) / Laya: held-out. Laya numbers: official checkpoints re-run by us on identical rows, shipped temperatures."
},
{
"task": "Jailbreak",
"metric": "balanced accuracy",
"v3": 0.9036238842064085,
"best_laya": 0.8311462000782389,
"best_laya_checkpoint": "multilingual",
"delta_pts": 7.25,
"delta_ci95_pts": [
2.99,
11.43
],
"n": 400,
"v3_entry": "theme.guardrails_jailbreak.balanced_accuracy.v3",
"laya_entry": "theme.guardrails_jailbreak.balanced_accuracy.laya_best",
"scoreboard_field": "metrics['theme.guardrails_jailbreak.balanced_accuracy']",
"protocol_label": "v3: held-out source (other permissive jailbreak sets + teacher) / Laya: held-out. Laya numbers: official checkpoints re-run by us on identical rows, shipped temperatures."
},
{
"task": "Toxicity",
"metric": "macro-F1",
"v3": 0.7089254055198327,
"best_laya": 0.41360068097985436,
"best_laya_checkpoint": "multilingual",
"delta_pts": 29.53,
"delta_ci95_pts": [
24.65,
34.37
],
"n": 400,
"v3_entry": "theme.moderation_toxicity.macro_f1.v3",
"laya_entry": "theme.moderation_toxicity.macro_f1.laya_best",
"scoreboard_field": "metrics['theme.moderation_toxicity.macro_f1']",
"protocol_label": "v3: held-out source (civil_comments + teacher; toxic-chat eval-only) / Laya: held-out. Laya numbers: official checkpoints re-run by us on identical rows, shipped temperatures."
},
{
"task": "Model routing",
"metric": "accuracy",
"v3": 0.9624060150375939,
"best_laya": 0.6591478696741855,
"best_laya_checkpoint": "typed",
"delta_pts": 30.33,
"delta_ci95_pts": [
26.06,
35.09
],
"n": 399,
"v3_entry": "theme.model_routing_domain.v3",
"laya_entry": "theme.model_routing_domain.laya_best",
"scoreboard_field": "metrics['theme.model_routing_domain']",
"protocol_label": "v3: near-domain (teacher-written; gsm8k/mbpp/AG rows eval-only) / Laya: held-out. Laya numbers: official checkpoints re-run by us on identical rows, shipped temperatures."
},
{
"task": "MASSIVE intent",
"metric": "37 held-out locales",
"v3": 0.6551351351351351,
"best_laya": 0.36108108108108106,
"best_laya_checkpoint": "multilingual",
"delta_pts": 29.41,
"delta_ci95_pts": [
27.54,
31.3
],
"n": 3700,
"v3_entry": "massive51.unseen37.macro_accuracy.v3",
"laya_entry": "massive51.unseen37.macro_accuracy.laya_best",
"scoreboard_field": "metrics['massive51.unseen37.macro_accuracy']",
"protocol_label": "v3: held-out (37 languages never trained) / Laya: held-out. Laya numbers: official checkpoints re-run by us on identical rows, shipped temperatures."
}
],
"footnote": "Laya numbers: official checkpoints (English, typed-decisions, multilingual) re-run by us on identical rows with their shipped temperatures and\ndefault token budgets; the best of the three is shown per task. v3 trained on same-kind tasks from other datasets, never on these eval rows:\nintent (CLINC150/HWU64; Banking77 never trained), jailbreak (other permissive sets + teacher), toxicity (civil_comments + teacher; toxic-chat\neval-only), routing (teacher-written; gsm8k/mbpp/AG rows eval-only), MASSIVE in 14 other locales (no MASSIVE rows in these 37).\nn = 400 / 400 / 400 / 399 / 3,700 (37 x 100). Every gap's paired 95% bootstrap CI excludes zero. Plotted values: figures/beyond_laya.json."
} |