Conversation
domain-adapters/ holds a domain fine-tuning pipeline on top of the untouched TimesFM 2.5 base model: - fetch_crypto.py: Binance public klines for 36 USDT pairs (1h, 1d) - finetune_domain.py: LoRA training with a price-relative loss and calendar-based train/val/test splits - forecast.py: forecast with base or any adapter (CUDA / MPS / CPU) - adapters/horizon-c1: trained crypto adapter weights (11 MB) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- fetch_crypto_minutes.py: 1m Binance archives (handles the 2025 switch to microsecond timestamps), builds 5m/15m candles - evaluate_layered.py: replays forecast-then-reforecast-every-step on unseen candles and compares against the price-vs-open baseline - adapters/horizon-c2: LoRA trained on 1m/5m/15m/1h Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- ohlcv.py: zero-initialised candle encoder hooked into TimesFM's input layer, flip-invariance aware (swaps wicks on the mirrored pass) - finetune_ohlcv.py: trains it on top of an adapter, with a --no_candle control run - evaluate_layered.py / forecast.py: feed full candles to candle-aware models - README: C3 matched C2 and the control within noise, so it is not promoted Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… C2) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
domain-adapterspipeline that trains small LoRA adapters on the untouched TimesFM 2.5 base for crypto candles, with calendar train/val/test splits so assets cannot leak future prices into each other.Test plan
google/timesfm-2.5-200m-transformerswith--adapter horizon-c1and--adapter horizon-c2.forecast.py --adapter horizon-c2 --symbol BTCUSDT --interval 15m --compareand confirm a point forecast and quantile band are produced.valbeforetest_start) infinetune_domain.py.evaluate_layered.pyon a short candle slice and check it reports the price-vs-open baseline next to the model.