Zero-Shot Classification
Laya
Safetensors
GGUF
Hebrew
hebrew
decision-model
calibrated
scam-detection
routing
Instructions to use BrainboxAI/nitzotz with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Laya
How to use BrainboxAI/nitzotz with Laya:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Download rl_agent_config.json from BrainboxAI/nitzotz: direct link, hf CLI and curl.
- Browser
- Download file 2.89 kB
-
https://huggingface.co/BrainboxAI/nitzotz/resolve/main/rl_agent_config.json
- Command line
-
hf download hf://BrainboxAI/nitzotz/rl_agent_config.json
-
curl -L -o rl_agent_config.json https://huggingface.co/BrainboxAI/nitzotz/resolve/main/rl_agent_config.json
2.89 kB
| { | |
| "encoder": "HalleluBERT/HalleluBERT_large", | |
| "head_layers": 2, | |
| "max_len": 512, | |
| "head_max_len": 256, | |
| "act_costs": { | |
| "escalate": 0.5 | |
| }, | |
| "amp_dtype": "bf16", | |
| "model_name": "nitzotz", | |
| "temperature": [ | |
| 1.0970542430877686, | |
| 1.1389989852905273, | |
| 1.1619657278060913 | |
| ], | |
| "training": { | |
| "encoder_init": "HalleluBERT-large after extractive QA on HeQ v1.1 train (test-overlapping passages removed)", | |
| "data": "116,854 items: synthetic Hebrew messages, HeQ v1.1 train, MASSIVE he-IL train, reading and topic items on FineWeb-2 Hebrew passages, warnings about scams, short scams and look-alike messages, 'none of the options' items, reworded copies (see README)", | |
| "items_total": 116854, | |
| "train_items": 112029, | |
| "calib_items": 4825, | |
| "epochs_planned": 2, | |
| "epochs_run": 2, | |
| "kept_epoch": 2, | |
| "heldout_by_epoch": [ | |
| { | |
| "acc": 0.9389, | |
| "soft_nll": 0.4909, | |
| "acc_base_families": 0.9359, | |
| "by_family": { | |
| "aug_heq": 0.9399, | |
| "aug_massive": 0.9464, | |
| "aug_scam": 0.9624, | |
| "aug_type": 0.9299, | |
| "aug_urgency": 0.8488, | |
| "claim": 0.9498, | |
| "heq_read": 0.964, | |
| "heq_unans": 0.892, | |
| "heq_verify": 0.9733, | |
| "massive_intent": 0.9286, | |
| "rel6_scam": 0.9574, | |
| "rel6_type": 0.931, | |
| "rel_scam": 0.9767, | |
| "rel_type": 0.9216, | |
| "routing": 0.9224, | |
| "spam_type": 0.9329, | |
| "urgency": 0.92 | |
| }, | |
| "epoch": 1, | |
| "step": 3501, | |
| "train_seconds": 1746.6 | |
| }, | |
| { | |
| "acc": 0.9505, | |
| "soft_nll": 0.4741, | |
| "acc_base_families": 0.9452, | |
| "by_family": { | |
| "aug_heq": 0.9563, | |
| "aug_massive": 0.9375, | |
| "aug_scam": 0.9731, | |
| "aug_type": 0.9482, | |
| "aug_urgency": 0.8721, | |
| "claim": 0.9658, | |
| "heq_read": 0.976, | |
| "heq_unans": 0.896, | |
| "heq_verify": 0.98, | |
| "massive_intent": 0.9314, | |
| "rel6_scam": 0.9787, | |
| "rel6_type": 0.9425, | |
| "rel_scam": 0.9833, | |
| "rel_type": 0.9515, | |
| "routing": 0.9361, | |
| "spam_type": 0.9486, | |
| "urgency": 0.8971 | |
| }, | |
| "epoch": 2, | |
| "step": 7002, | |
| "train_seconds": 1757.1 | |
| } | |
| ], | |
| "steps": 7002, | |
| "batch": 32, | |
| "micro_batch": 16, | |
| "lr_encoder": 1e-05, | |
| "lr_head": 0.0001, | |
| "warmup": 0.1, | |
| "schedule": "cosine", | |
| "seed": 1, | |
| "split_seed": 1234, | |
| "loss": "soft cross-entropy", | |
| "amp": "bf16", | |
| "device": "cuda", | |
| "train_seconds": 3520.6, | |
| "head_init": "fresh (no laya checkpoint weights)", | |
| "class_weights": "none", | |
| "option_shuffle_families": [ | |
| "read2_multi", | |
| "read2_neg", | |
| "read2_para", | |
| "read2_pos" | |
| ], | |
| "holdout_by_parent": true, | |
| "parent_copies_dropped": 0 | |
| } | |
| } |