423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-23 15:20:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

[model] support yarn (#6693)

Browse Source 
Former-commit-id: 8c412abc44a4c61b683465e36c6288580d980250
This commit is contained in:
hoshi-hiyouga
2025-01-18 13:56:09 +08:00
committed by GitHub
parent e4046bdd1f
commit 87d685b59f
11 changed files with 84 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
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 GPTQ_BITS, get_save_dir
from ..common import get_save_dir, load_config
from ..locales import ALERTS
@@ -32,6 +32,9 @@ 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)
@@ -54,6 +57,7 @@ def save_model(
export_dir: str,
export_hub_model_id: str,
) -> Generator[str, None, None]:
user_config = load_config()
error = ""
if not model_name:
error = ALERTS["err_no_model"][lang]
@@ -75,6 +79,7 @@ def save_model(
args = dict(
model_name_or_path=model_path,
cache_dir=user_config.get("cache_dir", None),
finetuning_type=finetuning_type,
template=template,
export_dir=export_dir,

10
src/llamafactory/webui/components/top.py
View File

@@ -41,11 +41,11 @@ def create_top() -> Dict[str, "Component"]:
checkpoint_path = gr.Dropdown(multiselect=True, allow_custom_value=True, scale=6)
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)
quantization_bit = gr.Dropdown(choices=["none", "8", "4"], value="none", allow_custom_value=True)
quantization_method = gr.Dropdown(choices=["bitsandbytes", "hqq", "eetq"], value="bitsandbytes")
template = gr.Dropdown(choices=list(TEMPLATES.keys()), value="default")
rope_scaling = gr.Dropdown(choices=["none", "linear", "dynamic", "yarn", "llama3"], value="none")
booster = gr.Dropdown(choices=["auto", "flashattn2", "unsloth", "liger_kernel"], value="auto")
model_name.change(get_model_info, [model_name], [model_path, template], queue=False).then(
list_checkpoints, [model_name, finetuning_type], [checkpoint_path], queue=False