Open-weight model
patchtst-wavelet-sp500-research
by Alex Goldenstein goldale/patchtst-wavelet-sp500-research
patchtst-wavelet-sp500-research is an open-weight model from Alex Goldenstein. It has 602,497 parameters. At 16-bit it needs about 0 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
OHLCV + MA5 + MA23 + MA53 + RSI + MACD + VIX + MAVOL = 12 channels Multi-scale moving averages, RSI, MACD, VIX, and volume smoothing — none break 50% DA.
Runs On
What it takes to serve patchtst-wavelet-sp500-research (602,497 parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 0.0 GB | 0.0 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 0.0 GB | 0.0 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 0.0 GB | 0.0 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 20, 2026.
patchtst-wavelet-sp500-research on every accelerator the SAVRN Index prices, at every precision
Model Card
OHLCV + MA5 + MA23 + MA53 + RSI + MACD + VIX + MAVOL = 12 channels Multi-scale moving averages, RSI, MACD, VIX, and volume smoothing — none break 50% DA. Cross-sectional ranking (not direction prediction): - Signal is in slow factors (60-day volatility, momentum), not daily direction 1. DA(diff) ≈ 70% is fake — always use DA(ctx) 2. Global wavelet = look-ahead bias — v3's 53.3% was 100% from leakage 3. More features = overfitting — v4 (25ch) complex — LightGBM (2 sec) > PatchTST (hours) > Kronos (102M params) - notebooks/ — v3-v6 Colab training notebooks - results/ — v3-v6 results JSON files - PROJECTCONCLUSION.md — Full findings + future work - CLAUDE.md — Agent reference (lessons…
Excerpt from the card by Alex Goldenstein.
Configuration
- Architecture
- PatchTSTForPrediction
- Layers
- 3
- Attention heads
- 16
- Model type
- patchtst
Identity and Version
- Repository
- goldale/patchtst-wavelet-sp500-research
- Publisher
- Alex Goldenstein
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- 602,497 parameters
- Languages
- Not stated by the source
- Revision
- 48678db79f4714d2a855dd4ef920e02718f34f0a
- First published
- 2026-09-20
- Last updated
- 2026-09-20
Files and Weights
79 files, 5.7 MB in total. The weights are 2 files totalling 4.9 MB in pt, safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| best_model.pt | Weights | 2.4 MB | 271d1cf144a7 |
| model.safetensors | Weights | 2.4 MB | cf4c652ad0a0 |
| config.json | Configuration | 1.2 KB | — |
| diagnose_v2.py | Configuration | 10.2 KB | — |
| fast_diag.py | Configuration | 8.7 KB | — |
| fast_diagnostic.json | Configuration | 872 B | — |
| metrics.json | Configuration | 879 B | — |
| mini_e2e_test.py | Configuration | 5.7 KB | — |
| patchtst_config_reference.py | Configuration | 5.2 KB | — |
| results/D1a_final_results.json | Configuration | 10.1 KB | — |
| results/D1a_live_checkpoint.json | Configuration | 9.8 KB | — |
| results/D2_final_results.json | Configuration | 5.1 KB | — |
| results/D2a_live_checkpoint.json | Configuration | 4.8 KB | — |
| results/kronos_daily_results.json | Configuration | 1.6 KB | — |
| results/kronos_findings.json | Configuration | 3.6 KB | — |
| results/kronos_sweep_results.json | Configuration | 5.4 KB | — |
| results/lgbm_sp500_full_results.json | Configuration | 429 B | — |
| results/lgbm_sp500_results.json | Configuration | 524 B | — |
| results/model_comparison_sp100.json | Configuration | 2.4 KB | — |
| results/v2_fast_diagnostic.json | Configuration | 872 B | — |
| results/v2_phase1_metrics.json | Configuration | 1.9 KB | — |
| results/v3_checkpoint.json | Configuration | 2.1 KB | — |
| results/v3_interim_results.json | Configuration | 4.0 KB | — |
| results/v3_live_checkpoint.json | Configuration | 46.6 KB | — |
| results/v3_results.json | Configuration | 50.2 KB | — |
| results/v3_single_split_metrics.json | Configuration | 495 B | — |
| results/v4_live_checkpoint.json | Configuration | 11.8 KB | — |
| results/v5_live_checkpoint.json | Configuration | 152 B | — |
| results/v6_live_checkpoint.json | Configuration | 12.9 KB | — |
| results/v6_results.json | Configuration | 1.7 KB | — |
| smoke_test.py | Configuration | 5.1 KB | — |
| train_patchtst.py | Configuration | 15.3 KB | — |
| train_stock_model.py | Configuration | 23.7 KB | — |
| train_v2_phase1.py | Configuration | 34.0 KB | — |
| train_v3.py | Configuration | 27.0 KB | — |
| train_v6.py | Configuration | 16.2 KB | — |
| train_v7.py | Configuration | 25.4 KB | — |
| train_volatility.py | Configuration | 15.8 KB | — |
| wavelet_denoise.py | Configuration | 3.9 KB | — |
| AGENT_HANDOFF.md | Documentation | 10.3 KB | — |
| CHECKPOINT.md | Documentation | 4.0 KB | — |
| CLAUDE.md | Documentation | 5.7 KB | — |
| ENHANCEMENT_RESEARCH.md | Documentation | 39.6 KB | — |
| GAPS_AND_MISSING_ANALYSIS.md | Documentation | 21.1 KB | — |
| GAP_FIX_PLAN.md | Documentation | 10 B | — |
| KRONOS_STUDY.md | Documentation | 7.9 KB | — |
| PROJECT_CONCLUSION.md | Documentation | 13.0 KB | — |
| README.md | Documentation | 1.8 KB | — |
| RESEARCH_FINDINGS.md | Documentation | 8.8 KB | — |
| ROADMAP.md | Documentation | 17.8 KB | — |
| SOTA_STUDY.md | Documentation | 4.2 KB | — |
| TASK_HANDOFF.md | Documentation | 6.3 KB | — |
| TSFM_STUDY.md | Documentation | 16.3 KB | — |
| V1_V5_DATA_REPORT.md | Documentation | 17.0 KB | — |
| V4_DESIGN.md | Documentation | 4.0 KB | — |
| XLSTM_TS_PAPER_BREAKDOWN.md | Documentation | 21.0 KB | — |
| debug/yfinance_output.txt | Other | 3.0 KB | — |
| logs/diagnose.log | Other | 309 B | — |
| logs/fast_diag.log | Other | 2.4 KB | — |
| logs/training.log | Other | 15.1 KB | — |
| logs/v3_training.log | Other | 5.2 KB | — |
| notebooks/D1a_paper_exact.ipynb | Other | 29.0 KB | — |
| notebooks/D2a_returns_baselines.ipynb | Other | 30.0 KB | — |
| notebooks/debug_yfinance.ipynb | Other | 4.3 KB | — |
| notebooks/gru_sp100_ranking.ipynb | Other | 12.8 KB | — |
| notebooks/kronos_demo.ipynb | Other | 8.8 KB | — |
| notebooks/kronos_optimized.ipynb | Other | 11.8 KB | — |
| notebooks/lgbm_sp500_ranking.ipynb | Other | 12.4 KB | — |
| notebooks/multi_market_test.ipynb | Other | 12.4 KB | — |
| notebooks/qlib_csi300_ranking.ipynb | Other | 11.3 KB | — |
| notebooks/qlib_quick_demo.ipynb | Other | 11.0 KB | — |
| notebooks/v3_inference_colab.ipynb | Other | 13.2 KB | — |
| notebooks/v3_train_colab.ipynb | Other | 26.4 KB | — |
| notebooks/v4_train_colab.ipynb | Other | 21.4 KB | — |
| notebooks/v5_train_colab.ipynb | Other | 18.8 KB | — |
| notebooks/v6_train_colab.ipynb | Other | 22.7 KB | — |
| requirements.txt | Other | 117 B | — |
| scaler.pkl | Other | 1.0 KB | 291badccf00d |
| .gitattributes | Repository | 1.5 KB | — |
License and Download
- License
- Not stated by the source
- Access
- Open weights, no gate
- Download size
- 4.9 MB
Released by Alex Goldenstein through its official repository on Hugging Face.
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 4.9 MB |
| 16-bit | 0.0 GB |
| 8-bit | 0.0 GB |
| 4-bit | 0.0 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About patchtst-wavelet-sp500-research
How much GPU memory does patchtst-wavelet-sp500-research need?
About 0 GB at 16-bit and 0 GB at 4-bit: the weights (602,497 parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run patchtst-wavelet-sp500-research on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.