| license: other | |
| tags: | |
| - ltx-2 | |
| - ltx-video | |
| - lora | |
| - text-to-video | |
| - audio-video | |
| base_model: Lightricks/LTX-2.3-22B | |
| library_name: ltx-trainer | |
| # LTX-2.3 DJ LoRA — lr5e4 / rank128 (step 1800) | |
| LoRA adapter for the **LTX-2.3 (22B)** audio-video generation model, fine-tuned on a DJ / music-performance video dataset. | |
| ## Checkpoint | |
| - **File:** `lora_weights_step_01800.safetensors` | |
| - **Training step:** 1800 | |
| ## Training configuration | |
| | Setting | Value | | |
| |---|---| | |
| | Base model | LTX-2.3 (22B) — `ltx-2.3-22b-dev` | | |
| | Training mode | LoRA | | |
| | LoRA rank | 128 | | |
| | LoRA alpha | 128 | | |
| | LoRA dropout | 0.0 | | |
| | Target modules | `to_k`, `to_q`, `to_v`, `to_out.0` | | |
| | Learning rate | 5e-4 | | |
| | Optimizer | AdamW | | |
| | LR scheduler | linear | | |
| | Batch size | 1 | | |
| | Grad accumulation | 8 | | |
| | Mixed precision | bf16 | | |
| | Gradient checkpointing | enabled | | |
| | Timestep sampling | shifted_logit_normal | | |
| | Strategy | text_to_video (audio disabled in training) | | |
| | Seed | 42 | | |
| Trained with [LTX Trainer](../ltx-trainer). | |