0.0.0

• First release since migration/refactor from ECG-Bench to ECG-Language-Models.
• We now support pretraining, supervised finetuning (sft), and post-training (SAPO).
• For post-training RL, we support both explicit and non explicit thinking.
• We also have all datasets migrated to an organization on HuggingFace.
• We have many new models, such as Base, Patch, and Conv. ELF. We also include a Llava based model using a more complicated MLP connector, instead of a simple linear projection.
• We have a Chat CLI. Feel free to check it out!
• Major bug fixes.
  • Spacing between signal token placeholder.
  • RL stability issues.
  • Gradient checkpointing issues.