Upload Kairos checkpoint

README.md

@@ -0,0 +1,52 @@
+---
+license: mit
+tags: [time-series, finance, kronos, kairos, crypto]
+library_name: pytorch
+---
+
+# Shadowell/Kairos-base-crypto
+
+Fine-tuned **Kronos-Tokenizer-base** (crypto (BTC/USDT + ETH/USDT 1-min)), produced by
+**[Kairos](https://github.com/Shadowell/Kairos)**.
+
+The tokenizer encodes a rolling 6-dim OHLCV+amount window into two streams of
+discrete tokens (BSQ — Binary Spherical Quantization) that Kronos predictors
+consume downstream. Only the OHLCV inputs are used for training; the 32-dim
+exogenous channel required by `KronosWithExogenous` is not needed for the
+tokenizer itself.
+
+## Results on test set (304,770 windows, 2 symbols)
+
+| metric | baseline | finetuned | Δ |
+|---|---|---|---|
+| recon_mse_full | 0.005504 | 0.004141 | -24.8% |
+| recon_mae_full | 0.055000 | 0.047185 | -14.2% |
+| bsq_loss_mean | -0.070294 | -0.070731 | +0.6% |
+| s1 util | 0.6904 | 0.7314 | +5.9% |
+| s2 util | 0.4326 | 0.4277 | -1.1% |
+| s1 entropy (bits) | 5.9374 | 6.1164 | +3.0% |
+| s2 entropy (bits) | 4.6063 | 4.7127 | +2.3% |
+
+| channel | baseline | finetuned | Δ |
+|---|---|---|---|
+| open | 0.00528 | 0.00332 | -37.0% |
+| high | 0.00556 | 0.00504 | -9.5% |
+| low | 0.00576 | 0.00502 | -12.9% |
+| close | 0.00437 | 0.00308 | -29.6% |
+| vol | 0.00615 | 0.00420 | -31.8% |
+| amt | 0.00589 | 0.00420 | -28.8% |
+
+## Usage
+
+```python
+from kairos import KronosTokenizer
+tok = KronosTokenizer.from_pretrained("Shadowell/Kairos-base-crypto")
+# Encode a [B, T, 6] OHLCV tensor into (s1_ids, s2_ids)
+s1, s2 = tok.encode(x, half=True)
+```
+
+## Training recipe
+
+See `docs/CRYPTO_TOKENIZER_RUN.md` in the upstream repo for the full
+reproduction checklist (data collection, preparation, 15-epoch + patience 3
+fine-tune, reconstruction evaluation).

config.json

@@ -0,0 +1,18 @@
+{
+  "attn_dropout_p": 0.0,
+  "beta": 0.05,
+  "d_in": 6,
+  "d_model": 256,
+  "ff_dim": 512,
+  "ffn_dropout_p": 0.0,
+  "gamma": 1.1,
+  "gamma0": 1.0,
+  "group_size": 4,
+  "n_dec_layers": 4,
+  "n_enc_layers": 4,
+  "n_heads": 4,
+  "resid_dropout_p": 0.0,
+  "s1_bits": 10,
+  "s2_bits": 10,
+  "zeta": 0.05
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9207551f3a7c1b66e5517602c3b7becbfd915acbb2dad9c96fd13812a1d93779
+size 15842368