Instructions to use rennf93/covenant-v3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use rennf93/covenant-v3 with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("rennf93/covenant-v3", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
-
Create README.md
Verified signature# covenant-v3 A fine-tuned Laya System-1 decision head for SOL-USD trading: **honestly calibrated probabilities, no demonstrated directional alpha.** Built as the System-1 compass of the [Covenant](https://github.com/rennf93/covenant) project (onchain proof-of-performance for trading strategies, Arbitrum Open House Singapore buildathon). ## Model - Base: [convaiinnovations/laya](https://huggingface.co/convaiinnovations/laya) (english checkpoint; ModernBERT-style encoder + typed-decision head). - Fine-tune: full-checkpoint SFT, per-question inverse-sqrt class weighting, 2 epochs, fp16 on a Kaggle T4, then per-(type,size) temperature refit by NLL on a chronological 10% holdout. ## Training data 7 months of SOL-USD 1m "wire-format" engine states (2026-03-01 to 2026-09-26), 299,139 examples, 67,100 tradeable (22.4%). Labels are triple-barrier: 240-bar (~4h) horizon, +3% take / -1.5% stop, net of ~1.2% round-trip taker fees. States are exactly the payloads the live engine serves …9d08f7f
-
Upload folder using huggingface_hub
Verified signature017a486 -
initial commit
Verified signature1316ba4