Downloads · 30 days
0
crypto-supreme-v13-distill
majid2230/crypto-supreme-v13-distill
crypto-supreme-v13-distill is a machine learning model from majid2230. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Qwen3.6-27B + LoRA r64 all-linear, distilled from GBM-OOF soft labels + hard labels on a 134,730-sample purged/embargoed pool (539 coins, 65 features incl. order-flow, funding, 52w-high anchoring, BTC-residualized mom…
Access
Public
Updated Jul 22, 2026
Repo size
1.9 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors1.9 GB · 100%
At a glance
- License
- apache-2.0
- Access
- Public
- Created
- Jul 22, 2026
- Updated
- Jul 22, 2026
- SHA
- 1fb54ab1
Base models
Model card
From the Hugging Face model README
crypto-supreme-v13 (LoRA adapter, THE deployed v13 student)
Qwen3.6-27B + LoRA r64 all-linear, distilled from GBM-OOF soft labels + hard labels on a 134,730-sample purged/embargoed pool (539 coins, 65 features incl. order-flow, funding, 52w-high anchoring, BTC-residualized momentum, sector momentum, altseason breadth). Selected checkpoint: distill step 16000 (val win@10 35.1%).
Results (v13 test, 17,457 events after event de-dup)
| slice | base | win@top-10% |
|---|---|---|
| ALL | 14.0% | 41.5% |
| unseen holdout coins | 14.9% | 43.2% |
| recent 60d (forward-looking, honest) | 9.5% | 32.4% (3.4x lift) |
| v13 GBM baseline (same set) | 14.0% | 26.3% |
OWCE (outcome-weighted RL-style stage) was trained and REJECTED by the val gate (35.1% -> 24.2%); the plain distilled checkpoint is the deployed model.
Score by Yes/No token logprobs on prompts ending Answer:. Never parse generated text.
- License
- apache-2.0
- Created
- Jul 22, 2026
- Updated
- Jul 22, 2026