opentslm 0.1.0

Rust implementation of OpenTSLM using Burn, WGPU, and llama.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29

//! Dataset loaders, batch utilities, and the optional auto-downloader.
//!
//! Each sub-module corresponds to one benchmark used during curriculum
//! training and mirrors the Python `time_series_datasets/` package.
//!
//! | Module | Dataset | Curriculum stage |
//! |--------|---------|------------------|
//! | [`tsqa`] | TSQA MCQ (WISDM-W accel) | Stage 1 |
//! | [`m4`] | M4 captioning (WISDM-W accel) | Stage 2 |
//! | [`har`] | HAR CoT (WISDM-W 3-axis) | Stage 3 |
//! | [`sleep`] | SleepEDF CoT (Fpz-Cz EEG) | Stage 4 |
//! | [`ecg`] | ECG-QA CoT (synthetic PTB-XL) | Stage 5 |
//!
//! The [`batch`] module provides the [`batch::Sample`] and
//! [`batch::TokenizedBatch`] types shared by all loaders, along with
//! helper functions for padding and normalisation.
//!
//! The `download` feature gate controls whether [`downloader`] is compiled.
//! In release builds the feature is enabled by default so that `train` and
//! `eval` can fetch missing datasets automatically.

pub mod batch;
#[cfg(feature = "download")]
pub mod downloader;
pub mod ecg;
pub mod har;
pub mod m4;
pub mod sleep;
pub mod tsqa;