Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-12-23 21:18:00 +01:00
Code Issues Packages Projects Releases Wiki Activity

use shared rows to make the LoRA Trainer interface a bit more compact / clean

Browse Source
This commit is contained in:
Alex "mcmonkey" Goodwin 2023-03-27 08:31:49 -07:00
parent e439228ed8
commit d911c22af9
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
modules/training.py
View File

@ -14,11 +14,13 @@ def get_json_dataset(path: str):
def create_train_interface():
with gr.Tab('Train LoRA', elem_id='lora-train-tab'):
loraName = gr.Textbox(label="Name", info="The name of your new LoRA file")
# TODO: Implement multi-device support.
microBatchSize = gr.Slider(label='Micro Batch Size', value=4, minimum=1, maximum=128, step=1, info='Per-device batch size (NOTE: multiple devices not yet implemented). Increasing this will increase VRAM usage.')
batchSize = gr.Slider(label='Batch Size', value=128, minimum=1, maximum=1024, step=4, info='Global batch size. The two batch sizes together determine gradient accumulation (gradientAccum = batch / microBatch). Higher gradient accum values lead to better quality training.')
epochs = gr.Slider(label='Epochs', value=1, minimum=1, maximum=1000, info='Number of times every entry in the dataset should be fed into training. So 1 means feed each item in once, 5 means feed it in five times, etc.')
learningRate = gr.Textbox(label='Learning Rate', value='3e-4', info='Learning rate, in scientific notation. 3e-4 is a good starting base point. 1e-2 is extremely high, 1e-6 is extremely low.')
with gr.Row():
# TODO: Implement multi-device support.
microBatchSize = gr.Slider(label='Micro Batch Size', value=4, minimum=1, maximum=128, step=1, info='Per-device batch size (NOTE: multiple devices not yet implemented). Increasing this will increase VRAM usage.')
batchSize = gr.Slider(label='Batch Size', value=128, minimum=1, maximum=1024, step=4, info='Global batch size. The two batch sizes together determine gradient accumulation (gradientAccum = batch / microBatch). Higher gradient accum values lead to better quality training.')
with gr.Row():
epochs = gr.Number(label='Epochs', value=1, minimum=1, maximum=1000, info='Number of times every entry in the dataset should be fed into training. So 1 means feed each item in once, 5 means feed it in five times, etc.')
learningRate = gr.Textbox(label='Learning Rate', value='3e-4', info='Learning rate, in scientific notation. 3e-4 is a good starting base point. 1e-2 is extremely high, 1e-6 is extremely low.')
# TODO: What is the actual maximum rank? Likely distinct per model. This might be better to somehow be on a log scale.
loraRank = gr.Slider(label='LoRA Rank', value=8, minimum=1, maximum=1024, step=4, info='LoRA Rank, or dimension count. Higher values produce a larger file with better control over the model\'s content. Smaller values produce a smaller file with less overall control. Small values like 4 or 8 are great for stylistic guidance, high values like 128 or 256 are good for teaching content upgrades. Higher ranks also require higher VRAM.')
loraAlpha = gr.Slider(label='LoRA Alpha', value=16, minimum=1, maximum=2048, step=4, info='LoRA Alpha. This divided by the rank becomes the scaling of the LoRA. Higher means stronger. A good standard value is twice your Rank.')
@ -29,10 +31,8 @@ def create_train_interface():
datasetFunction = get_json_dataset('training/datasets')
dataset = gr.Dropdown(choices=datasetFunction(), value='None', label='Dataset', info='The dataset file to use for training.')
ui.create_refresh_button(dataset, lambda : None, lambda : {'choices': datasetFunction()}, 'refresh-button')
with gr.Row():
evalDataset = gr.Dropdown(choices=datasetFunction(), value='None', label='Evaluation Dataset', info='The dataset file used to evaluate the model after training.')
ui.create_refresh_button(evalDataset, lambda : None, lambda : {'choices': datasetFunction()}, 'refresh-button')
with gr.Row():
formatsFunction = get_json_dataset('training/formats')
format = gr.Dropdown(choices=formatsFunction(), value='None', label='Data Format', info='The format file used to decide how to format the dataset input.')
ui.create_refresh_button(format, lambda : None, lambda : {'choices': formatsFunction()}, 'refresh-button')