Downloads · 30 days
3
8% of all-time downloads
crypto-olmo2-13b-r5-v9
majid2230/crypto-olmo2-13b-r5-v9
crypto-olmo2-13b-r5-v9 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. It is set up for peft. The card lists the license as apache-2.0.
PEFT LoRA fine-tuned for crypto pump prediction (binary Yes/No 7-day +15% move detection).
All-time downloads
39
Public
Repo size
1 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors1 GB · 99%
At a glance
- Library
- peft
- License
- apache-2.0
- Access
- Public
- Created
- May 15, 2026
- Updated
- May 15, 2026
- SHA
- 5878cdbd
Base models
Model card
From the Hugging Face model README
crypto-olmo2-13b-r5-v9
PEFT LoRA fine-tuned for crypto pump prediction (binary Yes/No 7-day +15% move detection).
V9 results — PASSED gate
| Metric | v9 | v8 |
|---|---|---|
| Raw MCC | +0.1987 (CI [+0.1729, +0.2278]) | +0.1410 |
| Threshold-tuned MCC | +0.1975 | — |
| Platt-calibrated MCC | +0.1372 | — |
| AUC-ROC | 0.683 | (saturation collapsed v8) |
| AUC-PR | 0.276 | — |
| Accuracy | 81.4% | — |
| F1 | 0.306 | — |
| Saturation@95 | 0.12% | ~100% in v8 |
| ECE | 0.173 (raw) / 0.020 (Platt) | — |
| Brier | 0.149 | — |
| n_test | 8000 (5898 coin-holdout unseen) | — |
Improvement: +40.9% over v8 (+0.1410 -> +0.1987).
Why v9 beats v8
- Calibrated CE loss (label_smoothing=0.05, pos_weight=6, conf_penalty=0.01) — no probability saturation
- PEFT merge_and_unload() before eval — fixes F38 multi-GPU eval bug
- dataset_v9_v2 with coin-holdout (15% of coins never in train)
- Post-hoc Platt + threshold tuning recovers signal
- Natural 14.3% Yes balance (+15% threshold) vs v8 oversampled 35%
Usage
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
base = AutoModelForCausalLM.from_pretrained("allenai/OLMo-2-1124-13B", torch_dtype=torch.bfloat16, device_map="auto")
model = PeftModel.from_pretrained(base, "majid2230/crypto-olmo2-13b-r5-v9")
model = model.merge_and_unload()
tok = AutoTokenizer.from_pretrained("allenai/OLMo-2-1124-13B")
Apply Platt scaling (a=0.8692625781817079, b=-1.0850393771368787) + threshold tune for best results.
Recipe (locked v9)
epochs=3 lora_r=64 LR=1.5e-5 warmup=0.05 max_length=768
label_smoothing=0.05 pos_weight=6.0 conf_penalty=0.01 patience=2
Part of R5 v9 cohort — https://huggingface.co/majid2230
- Library
- peft
- License
- apache-2.0
- Created
- May 15, 2026
- Updated
- May 15, 2026