# All dtypes at once
bash scripts/convert_all.sh

# Single dtype
./target/release/tirex-rs convert --dtype f32 --output gguf/tirex-f32.gguf
./target/release/tirex-rs convert --dtype f16 --output gguf/tirex-f16.gguf
./target/release/tirex-rs convert --dtype q8  --output gguf/tirex-q8.gguf

Supported dtypes: f32, f16, q8.

Infer

bash

./target/release/tirex-rs infer \
  --gguf gguf/tirex-f32.gguf \
  --data "1.0,2.1,3.3,2.8,1.9,3.1,4.0,3.5" \
  --horizon 32

Add --all-outputs to include all 9 quantile forecasts (q0.1–q0.9) in the JSON response.

The output is OpenAI-compatible JSON:

json

{
  "choices": [{
    "forecast": {
      "point": [...],
      "quantiles": { "0.10": [...], "0.50": [...], "0.90": [...] }
    }
  }]
}

Architecture

TiRex is a 35M parameter sLSTM-based time-series foundation model:

Input: Patch context of fixed length 2048 (left-padded with NaN if shorter), patch size 32 → 64 patches
Normalization: Per-series StandardScaler (non-causal mean/std over full context)
Patch embedding: ResidualBlock(64→2048→512) over concatenated [values | mask]
Backbone: 12 × sLSTM blocks (sequential recurrence over 64 tokens)
- Pre-RMSNorm → 4 headwise-linear gate projections (NH=4, DH=128) → sLSTM cell → MultiHeadLayerNorm → residual
- Pre-RMSNorm → SiLU gated FFN (512→1408→512) → residual
Output: ResidualBlock(512→2048→288) → 9 quantiles × 32 patch offsets per token
Decoding: AR loop: take last token's prediction, extend context with NaN, repeat

Python binding

bash

uv add --dev maturin
uv run maturin develop --release

python

import tirex_rs
model = tirex_rs.TiRex("gguf/tirex-f32.gguf")
result = model.forecast([1.0, 2.1, 3.3, 2.8, 1.9], horizon=32, all_outputs=True)
print(result["choices"][0]["forecast"]["point"])