mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-12-16 03:40:34 +08:00
@@ -41,13 +41,12 @@ def create_top() -> Dict[str, "Component"]:
|
||||
finetuning_type = gr.Dropdown(choices=METHODS, value="lora", scale=1)
|
||||
checkpoint_path = gr.Dropdown(multiselect=True, allow_custom_value=True, scale=6)
|
||||
|
||||
with gr.Accordion(open=False) as advanced_tab:
|
||||
with gr.Row():
|
||||
quantization_bit = gr.Dropdown(choices=["none", "8", "4"], value="none", allow_custom_value=True, scale=2)
|
||||
quantization_method = gr.Dropdown(choices=["bitsandbytes", "hqq", "eetq"], value="bitsandbytes", scale=2)
|
||||
template = gr.Dropdown(choices=list(TEMPLATES.keys()), value="default", scale=2)
|
||||
rope_scaling = gr.Radio(choices=["none", "linear", "dynamic"], value="none", scale=3)
|
||||
booster = gr.Radio(choices=["auto", "flashattn2", "unsloth", "liger_kernel"], value="auto", scale=5)
|
||||
with gr.Row():
|
||||
quantization_bit = gr.Dropdown(choices=["none", "8", "4"], value="none", allow_custom_value=True, scale=2)
|
||||
quantization_method = gr.Dropdown(choices=["bitsandbytes", "hqq", "eetq"], value="bitsandbytes", scale=2)
|
||||
template = gr.Dropdown(choices=list(TEMPLATES.keys()), value="default", scale=2)
|
||||
rope_scaling = gr.Radio(choices=["none", "linear", "dynamic"], value="none", scale=3)
|
||||
booster = gr.Radio(choices=["auto", "flashattn2", "unsloth", "liger_kernel"], value="auto", scale=5)
|
||||
|
||||
model_name.change(get_model_info, [model_name], [model_path, template], queue=False).then(
|
||||
list_checkpoints, [model_name, finetuning_type], [checkpoint_path], queue=False
|
||||
@@ -66,7 +65,6 @@ def create_top() -> Dict[str, "Component"]:
|
||||
model_path=model_path,
|
||||
finetuning_type=finetuning_type,
|
||||
checkpoint_path=checkpoint_path,
|
||||
advanced_tab=advanced_tab,
|
||||
quantization_bit=quantization_bit,
|
||||
quantization_method=quantization_method,
|
||||
template=template,
|
||||
|
||||
@@ -91,7 +91,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
|
||||
save_steps = gr.Slider(minimum=10, maximum=5000, value=100, step=10)
|
||||
warmup_steps = gr.Slider(minimum=0, maximum=5000, value=0, step=1)
|
||||
neftune_alpha = gr.Slider(minimum=0, maximum=10, value=0, step=0.1)
|
||||
optim = gr.Textbox(value="adamw_torch")
|
||||
extra_args = gr.Textbox(value='{"optim": "adamw_torch"}')
|
||||
|
||||
with gr.Row():
|
||||
with gr.Column():
|
||||
@@ -116,7 +116,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
|
||||
save_steps,
|
||||
warmup_steps,
|
||||
neftune_alpha,
|
||||
optim,
|
||||
extra_args,
|
||||
packing,
|
||||
neat_packing,
|
||||
train_on_prompt,
|
||||
@@ -134,7 +134,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
|
||||
save_steps=save_steps,
|
||||
warmup_steps=warmup_steps,
|
||||
neftune_alpha=neftune_alpha,
|
||||
optim=optim,
|
||||
extra_args=extra_args,
|
||||
packing=packing,
|
||||
neat_packing=neat_packing,
|
||||
train_on_prompt=train_on_prompt,
|
||||
|
||||
Reference in New Issue
Block a user