ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

support HQQ/EETQ #4113

Browse Source 
Former-commit-id: b7cb51ddb394f04fe4646b2c297fc8d918c9979e
This commit is contained in:
hiyouga
2024-06-27 00:29:42 +08:00
parent 08fa707085
commit 8aaf1185a5
16 changed files with 134 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
src/llamafactory/webui/components/top.py
View File

@@ -18,7 +18,7 @@ from ...data import TEMPLATES
from ...extras.constants import METHODS, SUPPORTED_MODELS
from ...extras.packages import is_gradio_available
from ..common import get_model_info, list_checkpoints, save_config
from ..utils import can_quantize
from ..utils import can_quantize, can_quantize_to
if is_gradio_available():
@@ -43,10 +43,11 @@ def create_top() -> Dict[str, "Component"]:
with gr.Accordion(open=False) as advanced_tab:
with gr.Row():
quantization_bit = gr.Dropdown(choices=["none", "8", "4"], value="none", 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=["none", "flashattn2", "unsloth"], value="none", scale=3)
quantization_bit = gr.Dropdown(choices=["none", "8", "4"], value="none", scale=1)
quantization_method = gr.Dropdown(choices=["bitsandbytes", "hqq", "eetq"], value="bitsandbytes", scale=1)
template = gr.Dropdown(choices=list(TEMPLATES.keys()), value="default", scale=1)
rope_scaling = gr.Radio(choices=["none", "linear", "dynamic"], value="none", scale=2)
booster = gr.Radio(choices=["auto", "flashattn2", "unsloth"], value="auto", scale=2)
visual_inputs = gr.Checkbox(scale=1)
model_name.change(get_model_info, [model_name], [model_path, template, visual_inputs], queue=False).then(
@@ -58,6 +59,7 @@ def create_top() -> Dict[str, "Component"]:
list_checkpoints, [model_name, finetuning_type], [checkpoint_path], queue=False
)
checkpoint_path.focus(list_checkpoints, [model_name, finetuning_type], [checkpoint_path], queue=False)
quantization_method.change(can_quantize_to, [quantization_method], [quantization_bit], queue=False)
return dict(
lang=lang,
@@ -67,6 +69,7 @@ def create_top() -> Dict[str, "Component"]:
checkpoint_path=checkpoint_path,
advanced_tab=advanced_tab,
quantization_bit=quantization_bit,
quantization_method=quantization_method,
template=template,
rope_scaling=rope_scaling,
booster=booster,