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

better llamaboard

Browse Source 
* easily resume from checkpoint
* support full and freeze checkpoints
* faster ui


Former-commit-id: 84cfb2452cc86b037ccddee6e833f8eb7c129fa4
This commit is contained in:
hiyouga
2024-05-29 23:55:38 +08:00
parent f90c4ca672
commit 87aa332583
14 changed files with 303 additions and 193 deletions
Download Patch File Download Diff File Expand all files Collapse all files

109
src/llamafactory/webui/common.py
View File

@@ -1,12 +1,12 @@
import json
import os
from collections import defaultdict
from typing import Any, Dict, Optional
from typing import Any, Dict, Optional, Tuple
from peft.utils import SAFETENSORS_WEIGHTS_NAME, WEIGHTS_NAME
from yaml import safe_dump, safe_load
from ..extras.constants import (
CHECKPOINT_NAMES,
DATA_CONFIG,
DEFAULT_MODULE,
DEFAULT_TEMPLATE,
@@ -29,7 +29,6 @@ if is_gradio_available():
logger = get_logger(__name__)
ADAPTER_NAMES = {WEIGHTS_NAME, SAFETENSORS_WEIGHTS_NAME}
DEFAULT_CACHE_DIR = "cache"
DEFAULT_CONFIG_DIR = "config"
DEFAULT_DATA_DIR = "data"
@@ -38,19 +37,31 @@ USER_CONFIG = "user_config.yaml"
def get_save_dir(*paths: str) -> os.PathLike:
r"""
Gets the path to saved model checkpoints.
"""
paths = (path.replace(os.path.sep, "").replace(" ", "").strip() for path in paths)
return os.path.join(DEFAULT_SAVE_DIR, *paths)
def get_config_path() -> os.PathLike:
r"""
Gets the path to user config.
"""
return os.path.join(DEFAULT_CACHE_DIR, USER_CONFIG)
def get_save_path(config_path: str) -> os.PathLike:
def get_arg_save_path(config_path: str) -> os.PathLike:
r"""
Gets the path to saved arguments.
"""
return os.path.join(DEFAULT_CONFIG_DIR, config_path)
def load_config() -> Dict[str, Any]:
r"""
Loads user config if exists.
"""
try:
with open(get_config_path(), "r", encoding="utf-8") as f:
return safe_load(f)
@@ -59,6 +70,9 @@ def load_config() -> Dict[str, Any]:
def save_config(lang: str, model_name: Optional[str] = None, model_path: Optional[str] = None) -> None:
r"""
Saves user config.
"""
os.makedirs(DEFAULT_CACHE_DIR, exist_ok=True)
user_config = load_config()
user_config["lang"] = lang or user_config["lang"]
@@ -69,23 +83,10 @@ def save_config(lang: str, model_name: Optional[str] = None, model_path: Optiona
safe_dump(user_config, f)
def load_args(config_path: str) -> Optional[Dict[str, Any]]:
try:
with open(get_save_path(config_path), "r", encoding="utf-8") as f:
return safe_load(f)
except Exception:
return None
def save_args(config_path: str, config_dict: Dict[str, Any]) -> str:
os.makedirs(DEFAULT_CONFIG_DIR, exist_ok=True)
with open(get_save_path(config_path), "w", encoding="utf-8") as f:
safe_dump(config_dict, f)
return str(get_save_path(config_path))
def get_model_path(model_name: str) -> str:
def get_model_path(model_name: str) -> Optional[str]:
r"""
Gets the model path according to the model name.
"""
user_config = load_config()
path_dict: Dict[DownloadSource, str] = SUPPORTED_MODELS.get(model_name, defaultdict(str))
model_path = user_config["path_dict"].get(model_name, None) or path_dict.get(DownloadSource.DEFAULT, None)
@@ -99,40 +100,71 @@ def get_model_path(model_name: str) -> str:
def get_prefix(model_name: str) -> str:
r"""
Gets the prefix of the model name to obtain the model family.
"""
return model_name.split("-")[0]
def get_model_info(model_name: str) -> Tuple[str, str, bool]:
r"""
Gets the necessary information of this model.
Returns:
model_path (str)
template (str)
visual (bool)
"""
return get_model_path(model_name), get_template(model_name), get_visual(model_name)
def get_module(model_name: str) -> str:
return DEFAULT_MODULE.get(get_prefix(model_name), "q_proj,v_proj")
r"""
Gets the LoRA modules of this model.
"""
return DEFAULT_MODULE.get(get_prefix(model_name), "all")
def get_template(model_name: str) -> str:
r"""
Gets the template name if the model is a chat model.
"""
if model_name and model_name.endswith("Chat") and get_prefix(model_name) in DEFAULT_TEMPLATE:
return DEFAULT_TEMPLATE[get_prefix(model_name)]
return "default"
def get_visual(model_name: str) -> bool:
r"""
Judges if the model is a vision language model.
"""
return get_prefix(model_name) in VISION_MODELS
def list_adapters(model_name: str, finetuning_type: str) -> "gr.Dropdown":
if finetuning_type not in PEFT_METHODS:
return gr.Dropdown(value=[], choices=[], interactive=False)
adapters = []
if model_name and finetuning_type == "lora":
def list_checkpoints(model_name: str, finetuning_type: str) -> "gr.Dropdown":
r"""
Lists all available checkpoints.
"""
checkpoints = []
if model_name:
save_dir = get_save_dir(model_name, finetuning_type)
if save_dir and os.path.isdir(save_dir):
for adapter in os.listdir(save_dir):
if os.path.isdir(os.path.join(save_dir, adapter)) and any(
os.path.isfile(os.path.join(save_dir, adapter, name)) for name in ADAPTER_NAMES
for checkpoint in os.listdir(save_dir):
if os.path.isdir(os.path.join(save_dir, checkpoint)) and any(
os.path.isfile(os.path.join(save_dir, checkpoint, name)) for name in CHECKPOINT_NAMES
):
adapters.append(adapter)
return gr.Dropdown(value=[], choices=adapters, interactive=True)
checkpoints.append(checkpoint)
if finetuning_type in PEFT_METHODS:
return gr.Dropdown(value=[], choices=checkpoints, multiselect=True)
else:
return gr.Dropdown(value=None, choices=checkpoints, multiselect=False)
def load_dataset_info(dataset_dir: str) -> Dict[str, Dict[str, Any]]:
r"""
Loads dataset_info.json.
"""
if dataset_dir == "ONLINE":
logger.info("dataset_dir is ONLINE, using online dataset.")
return {}
@@ -145,12 +177,11 @@ def load_dataset_info(dataset_dir: str) -> Dict[str, Dict[str, Any]]:
return {}
def list_dataset(dataset_dir: str = None, training_stage: str = list(TRAINING_STAGES.keys())[0]) -> "gr.Dropdown":
def list_datasets(dataset_dir: str = None, training_stage: str = list(TRAINING_STAGES.keys())[0]) -> "gr.Dropdown":
r"""
Lists all available datasets in the dataset dir for the training stage.
"""
dataset_info = load_dataset_info(dataset_dir if dataset_dir is not None else DEFAULT_DATA_DIR)
ranking = TRAINING_STAGES[training_stage] in STAGES_USE_PAIR_DATA
datasets = [k for k, v in dataset_info.items() if v.get("ranking", False) == ranking]
return gr.Dropdown(value=[], choices=datasets)
def autoset_packing(training_stage: str = list(TRAINING_STAGES.keys())[0]) -> "gr.Button":
return gr.Button(value=(TRAINING_STAGES[training_stage] == "pt"))
return gr.Dropdown(choices=datasets)