mirror of
https://github.com/ml-explore/mlx-examples.git
synced 2025-06-26 18:51:18 +08:00
update lora_config.yaml
This commit is contained in:
parent
1b4e19675d
commit
06a9f5d106
@ -7,6 +7,19 @@ train: true
|
|||||||
# The fine-tuning method: "lora", "dora", or "full".
|
# The fine-tuning method: "lora", "dora", or "full".
|
||||||
fine_tune_type: lora
|
fine_tune_type: lora
|
||||||
|
|
||||||
|
# The training-mode: "normal", or "dpo"
|
||||||
|
training_mode: normal
|
||||||
|
|
||||||
|
# If you set training_mode to "dpo"
|
||||||
|
# beta: 0.1
|
||||||
|
# The dpo-lodd-type: "sigmoid", "hinge", "ipo", or "dpop"
|
||||||
|
# dpo_loss_type: "sigmoid"
|
||||||
|
# is_reference_free: False
|
||||||
|
# delta: 50.0
|
||||||
|
# If reference_model_path is not given it will just use the same model
|
||||||
|
# reference_model_path: "mlx_model"
|
||||||
|
# train_bias_only: False
|
||||||
|
|
||||||
# Directory with {train, valid, test}.jsonl files
|
# Directory with {train, valid, test}.jsonl files
|
||||||
data: "/path/to/training/data"
|
data: "/path/to/training/data"
|
||||||
|
|
||||||
|
Loading…
Reference in New Issue
Block a user