zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-08-14 05:36:38 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

udpate LORA.md

Browse Source
This commit is contained in:
Goekdeniz-Guelmez 2025-03-11 10:56:37 +01:00
parent d9d77f952d
commit 7300b3d871
1 changed files with 4 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
llms/mlx_lm/LORA.md
View File

@ -135,6 +135,8 @@ mlx_lm.lora \
- `--reference-model-path`: Path to reference model weights. If not specified, uses the same model - `--reference-model-path`: Path to reference model weights. If not specified, uses the same model
- `--temperature`: Sampling temperature for generations. Higher values increase randomness (default: 1.0) - `--temperature`: Sampling temperature for generations. Higher values increase randomness (default: 1.0)
- `--reward-weights`: Optional list of weights for multiple reward functions. Must match number of reward functions. If not specified, all rewards weighted equally with 1.0 - `--reward-weights`: Optional list of weights for multiple reward functions. Must match number of reward functions. If not specified, all rewards weighted equally with 1.0
- `--use-chat-template`: Whether to use the model's chat template for formatting prompts (default: False)
- `--use-prompt`: Whether to use the prompt as part of the input for generation (default: False)
## Training Process ## Training Process
@ -170,6 +172,8 @@ mlx_lm.lora \
--reference-model-path <optional_path_to_reference_model> \ --reference-model-path <optional_path_to_reference_model> \
--temperature 1.0 \ --temperature 1.0 \
--reward-weights 1.0 1.0 \ --reward-weights 1.0 1.0 \
--use-chat-template False \
--use-prompt False \
--batch-size 4 \ --batch-size 4 \
--learning-rate 1e-5 \ --learning-rate 1e-5 \
--num-epochs 3 --num-epochs 3