423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-16 03:40:34 +08:00
Code Issues Packages Projects Releases Wiki Activity

support HQQ/EETQ #4113

Browse Source 
Former-commit-id: ad144c2265
This commit is contained in:
hiyouga
2024-06-27 00:29:42 +08:00
parent 6b2733ce12
commit d2d9fa4abb
16 changed files with 134 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llamafactory/webui/components/export.py
View File

@@ -18,7 +18,7 @@ from ...extras.constants import PEFT_METHODS
from ...extras.misc import torch_gc
from ...extras.packages import is_gradio_available
from ...train.tuner import export_model
from ..common import get_save_dir
from ..common import GPTQ_BITS, get_save_dir
from ..locales import ALERTS
@@ -32,9 +32,6 @@ if TYPE_CHECKING:
from ..engine import Engine
GPTQ_BITS = ["8", "4", "3", "2"]
def can_quantize(checkpoint_path: Union[str, List[str]]) -> "gr.Dropdown":
if isinstance(checkpoint_path, list) and len(checkpoint_path) != 0:
return gr.Dropdown(value="none", interactive=False)

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,