What's Changed
- Add input_embeddings input to generate_step, Gemma 3, Qwen 2 by @mattjcly in #179
- Wandb fix by @Goekdeniz-Guelmez in #177
- fix upload to hub usage by @IANTHEREAL in #180
- Pixtral text support, pipe input_embeddings through llama arch by @mattjcly in #181
- Fix server cache by @awni in #183
- fix WandB setting in lora_congig.yaml by @Goekdeniz-Guelmez in #184
- fix server.py null tool content (#185) by @jpangburn in #186
- Fix URL for LoRA Data in Error Message by @rgeorgi in #188
- Faster top-p and min-p sampling by @awni in #187
- Add Evaluation Progress by @rgeorgi in #158
- DWQ updates by @awni in #196
- nits for qwen3_moe file by @Goekdeniz-Guelmez in #199
- Update LoRA training logic to reset lora_parameters when fine-tuning … by @Goekdeniz-Guelmez in #200
- Fix chat templates by @awni in #205
- Add total params to metadata + cleanup by @awni in #207
- Dynamic quants by @awni in #202
- some nits in lora by @awni in #208
- Fix quant package by @awni in #214
New Contributors
- @mattjcly made their first contribution in #179
- @IANTHEREAL made their first contribution in #180
- @jpangburn made their first contribution in #186
- @rgeorgi made their first contribution in #188
Full Changelog: v0.24.1...v0.25.1