423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-17 12:20:37 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix eval in webui

Browse Source
This commit is contained in:
hiyouga
2024-05-04 00:19:19 +08:00
parent 510e64ee70
commit 24cc93ab15
6 changed files with 84 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/llmtuner/webui/common.py
View File

@@ -17,6 +17,7 @@ from ..extras.constants import (
TRAINING_STAGES,
DownloadSource,
)
from ..extras.logging import get_logger
from ..extras.misc import use_modelscope
from ..extras.packages import is_gradio_available
@@ -25,6 +26,9 @@ if is_gradio_available():
import gradio as gr
logger = get_logger(__name__)
ADAPTER_NAMES = {WEIGHTS_NAME, SAFETENSORS_WEIGHTS_NAME}
DEFAULT_CACHE_DIR = "cache"
DEFAULT_CONFIG_DIR = "config"
@@ -128,11 +132,15 @@ def list_adapters(model_name: str, finetuning_type: str) -> "gr.Dropdown":
def load_dataset_info(dataset_dir: str) -> Dict[str, Dict[str, Any]]:
if dataset_dir == "ONLINE":
logger.info("dataset_dir is ONLINE, using online dataset.")
return {}
try:
with open(os.path.join(dataset_dir, DATA_CONFIG), "r", encoding="utf-8") as f:
return json.load(f)
except Exception as err:
print("Cannot open {} due to {}.".format(os.path.join(dataset_dir, DATA_CONFIG), str(err)))
logger.warning("Cannot open {} due to {}.".format(os.path.join(dataset_dir, DATA_CONFIG), str(err)))
return {}

15
src/llmtuner/webui/components/eval.py
View File

@@ -21,16 +21,16 @@ def create_eval_tab(engine: "Engine") -> Dict[str, "Component"]:
with gr.Row():
dataset_dir = gr.Textbox(value=DEFAULT_DATA_DIR, scale=2)
dataset = gr.Dropdown(multiselect=True, scale=4)
dataset = gr.Dropdown(multiselect=True, allow_custom_value=True, scale=4)
preview_elems = create_preview_box(dataset_dir, dataset)
input_elems.update({dataset_dir, dataset})
elem_dict.update(dict(dataset_dir=dataset_dir, dataset=dataset, **preview_elems))
with gr.Row():
cutoff_len = gr.Slider(value=1024, minimum=4, maximum=8192, step=1)
cutoff_len = gr.Slider(value=1024, minimum=4, maximum=65536, step=1)
max_samples = gr.Textbox(value="100000")
batch_size = gr.Slider(value=8, minimum=1, maximum=512, step=1)
batch_size = gr.Slider(value=2, minimum=1, maximum=1024, step=1)
predict = gr.Checkbox(value=True)
input_elems.update({cutoff_len, max_samples, batch_size, predict})
@@ -48,30 +48,27 @@ def create_eval_tab(engine: "Engine") -> Dict[str, "Component"]:
with gr.Row():
cmd_preview_btn = gr.Button()
start_btn = gr.Button(variant="primary")
stop_btn = gr.Button(variant="stop")
with gr.Row():
resume_btn = gr.Checkbox(visible=False, interactive=False)
process_bar = gr.Slider(visible=False, interactive=False)
progress_bar = gr.Slider(visible=False, interactive=False)
with gr.Row():
output_box = gr.Markdown()
output_elems = [output_box, process_bar]
output_elems = [output_box, progress_bar]
elem_dict.update(
dict(
cmd_preview_btn=cmd_preview_btn,
start_btn=start_btn,
stop_btn=stop_btn,
resume_btn=resume_btn,
process_bar=process_bar,
progress_bar=progress_bar,
output_box=output_box,
)
)
cmd_preview_btn.click(engine.runner.preview_eval, input_elems, output_elems, concurrency_limit=None)
start_btn.click(engine.runner.run_eval, input_elems, output_elems)
stop_btn.click(engine.runner.set_abort)
resume_btn.change(engine.runner.monitor, outputs=output_elems, concurrency_limit=None)
dataset_dir.change(list_dataset, [dataset_dir], [dataset], queue=False)

4
src/llmtuner/webui/components/train.py
View File

@@ -27,7 +27,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
choices=list(TRAINING_STAGES.keys()), value=list(TRAINING_STAGES.keys())[0], scale=1
)
dataset_dir = gr.Textbox(value=DEFAULT_DATA_DIR, scale=1)
dataset = gr.Dropdown(multiselect=True, scale=4)
dataset = gr.Dropdown(multiselect=True, allow_custom_value=True, scale=4)
preview_elems = create_preview_box(dataset_dir, dataset)
input_elems.update({training_stage, dataset_dir, dataset})
@@ -52,7 +52,7 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
)
with gr.Row():
cutoff_len = gr.Slider(value=1024, minimum=4, maximum=16384, step=1)
cutoff_len = gr.Slider(value=1024, minimum=4, maximum=65536, step=1)
batch_size = gr.Slider(value=2, minimum=1, maximum=1024, step=1)
gradient_accumulation_steps = gr.Slider(value=8, minimum=1, maximum=1024, step=1)
val_size = gr.Slider(value=0, minimum=0, maximum=1, step=0.001)

4
src/llmtuner/webui/runner.py
View File

@@ -299,12 +299,12 @@ class Runner:
progress_bar: gr.Slider(visible=False),
}
else:
running_log, running_progress, running_loss = get_trainer_info(output_path)
running_log, running_progress, running_loss = get_trainer_info(output_path, self.do_train)
return_dict = {
output_box: running_log,
progress_bar: running_progress,
}
if self.do_train and running_loss is not None:
if running_loss is not None:
return_dict[loss_viewer] = running_loss
yield return_dict

4
src/llmtuner/webui/utils.py
View File

@@ -63,7 +63,7 @@ def get_time() -> str:
return datetime.now().strftime(r"%Y-%m-%d-%H-%M-%S")
def get_trainer_info(output_path: os.PathLike) -> Tuple[str, "gr.Slider", Optional["gr.Plot"]]:
def get_trainer_info(output_path: os.PathLike, do_train: bool) -> Tuple[str, "gr.Slider", Optional["gr.Plot"]]:
running_log = ""
running_progress = gr.Slider(visible=False)
running_loss = None
@@ -91,7 +91,7 @@ def get_trainer_info(output_path: os.PathLike) -> Tuple[str, "gr.Slider", Option
)
running_progress = gr.Slider(label=label, value=percentage, visible=True)
if is_matplotlib_available():
if do_train and is_matplotlib_available():
running_loss = gr.Plot(gen_loss_plot(trainer_log))
return running_log, running_progress, running_loss