Text Classification
Transformers
Safetensors
English
Chinese
modernbert
reranker
cross-encoder
agent
decision-making
zero-shot-classification
fast-decider
text-embeddings-inference
Instructions to use mkzero/FastDecider-149M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use mkzero/FastDecider-149M with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="mkzero/FastDecider-149M")# Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("mkzero/FastDecider-149M") model = AutoModelForSequenceClassification.from_pretrained("mkzero/FastDecider-149M", device_map="auto") - Notebooks
- Google Colab
- Kaggle
| { | |
| "model_name": "FastDecider-149M", | |
| "version": "1.3.0", | |
| "archive_date": "2026-09-22", | |
| "base_model": "Alibaba-DAMO/gte-reranker-modernbert-base", | |
| "base_model_architecture": "ModernBERT-base Cross-Encoder (SequenceClassification)", | |
| "parameter_count": 149079553, | |
| "parameter_size_mb": "149M", | |
| "num_layers": 22, | |
| "hidden_size": 768, | |
| "max_position_embeddings": 8192, | |
| "training_details": { | |
| "dataset": "/mnt/workspace/data/grand_unified_train_35k.jsonl", | |
| "total_samples": 30456, | |
| "trainer_script": "/mnt/workspace/arena/train_grand_unified_35k.py", | |
| "training_steps": 600, | |
| "optimizer": "AdamW (lr=2e-5, warmup=50, cosine schedule)", | |
| "loss_function": "Cross-Entropy over candidate option logits", | |
| "precision": "bfloat16" | |
| }, | |
| "benchmarks": { | |
| "jevbench_public_easy": { | |
| "accuracy": "48/48 (100.00%)", | |
| "extraction": "12/12 (100.0%)", | |
| "fact": "12/12 (100.0%)", | |
| "intent": "12/12 (100.0%)", | |
| "tool_selection": "12/12 (100.0%)" | |
| }, | |
| "jevbench_public_hard": { | |
| "accuracy": "49/111 (44.14%)", | |
| "routing_hard": "5/5 (100.0%)", | |
| "tradeoff": "6/6 (100.0%)", | |
| "trap": "5/8 (62.5%)", | |
| "ambiguous": "4/7 (57.1%)", | |
| "adversarial": "3/6 (50.0%)", | |
| "probability": "5/10 (50.0%)", | |
| "judge_hard": "7/17 (41.2%)", | |
| "temporal_numeric": "5/15 (33.3%)", | |
| "long_policy": "6/19 (31.6%)", | |
| "multi_hop": "3/18 (16.7%)" | |
| }, | |
| "jevbench_public_original": { | |
| "accuracy": "49/72 (68.06%)", | |
| "intelligence_score": 65.35 | |
| }, | |
| "six_business_pillars": { | |
| "browser_dom_control": "50/50 (100.0%)", | |
| "api_dispatch_routing": "49/50 (98.0%)", | |
| "ecom_brand_extraction": "50/50 (100.0%)", | |
| "ecom_specs_matching": "50/50 (100.0%)", | |
| "ecom_category_prediction": "47/50 (94.0%)", | |
| "legal_contract_matching": "48/50 (96.0%)" | |
| }, | |
| "latency_and_cost": { | |
| "p50_latency_ms": 14.93, | |
| "p99_latency_ms": 17.8, | |
| "serving_cost": "$0.0012 per 1k decisions", | |
| "throughput_qps": 84.2 | |
| } | |
| }, | |
| "reproducibility": "All scores evaluated via standard argmax over cross-encoder logits without heuristic overrides." | |
| } |