release v0.9.0 (real)

Former-commit-id: 90d6df622252c6fad985f68b97771c979357e2fc

.env.local

@@ -24,10 +24,12 @@ WANDB_DISABLED=
 WANDB_PROJECT=huggingface
 WANDB_API_KEY=
 # gradio ui
-GRADIO_SHARE=0
+GRADIO_SHARE=False
 GRADIO_SERVER_NAME=0.0.0.0
 GRADIO_SERVER_PORT=
 GRADIO_ROOT_PATH=
+# setup
+ENABLE_SHORT_CONSOLE=1
 # reserved (do not use)
 LLAMABOARD_ENABLED=
 LLAMABOARD_WORKDIR=

README.md

@@ -275,7 +275,7 @@ You also can add a custom chat template to [template.py](src/llamafactory/data/t
 - [Magpie-Pro-300K-Filtered (en)](https://huggingface.co/datasets/Magpie-Align/Magpie-Pro-300K-Filtered)
 - [Magpie-ultra-v0.1 (en)](https://huggingface.co/datasets/argilla/magpie-ultra-v0.1)
 - [LLaVA mixed (en&zh)](https://huggingface.co/datasets/BUAADreamer/llava-en-zh-300k)
-- [Pokemon-gpt4o-captions](https://huggingface.co/datasets/jugg1024/pokemon-gpt4o-captions)
+- [Pokemon-gpt4o-captions (en&zh)](https://huggingface.co/datasets/jugg1024/pokemon-gpt4o-captions)
 - [Open Assistant (de)](https://huggingface.co/datasets/mayflowergmbh/oasst_de)
 - [Dolly 15k (de)](https://huggingface.co/datasets/mayflowergmbh/dolly-15k_de)
 - [Alpaca GPT4 (de)](https://huggingface.co/datasets/mayflowergmbh/alpaca-gpt4_de)

README_zh.md

@@ -276,7 +276,7 @@ https://github.com/user-attachments/assets/e6ce34b0-52d5-4f3e-a830-592106c4c272
 - [Magpie-Pro-300K-Filtered (en)](https://huggingface.co/datasets/Magpie-Align/Magpie-Pro-300K-Filtered)
 - [Magpie-ultra-v0.1 (en)](https://huggingface.co/datasets/argilla/magpie-ultra-v0.1)
 - [LLaVA mixed (en&zh)](https://huggingface.co/datasets/BUAADreamer/llava-en-zh-300k)
-- [Pokemon-gpt4o-captions](https://huggingface.co/datasets/jugg1024/pokemon-gpt4o-captions)
+- [Pokemon-gpt4o-captions (en&zh)](https://huggingface.co/datasets/jugg1024/pokemon-gpt4o-captions)
 - [Open Assistant (de)](https://huggingface.co/datasets/mayflowergmbh/oasst_de)
 - [Dolly 15k (de)](https://huggingface.co/datasets/mayflowergmbh/dolly-15k_de)
 - [Alpaca GPT4 (de)](https://huggingface.co/datasets/mayflowergmbh/alpaca-gpt4_de)

scripts/cal_mfu.py

@@ -131,6 +131,9 @@ def calculate_mfu(
         "dataset": "c4_demo",
         "cutoff_len": seq_length,
         "output_dir": os.path.join("saves", "test_mfu"),
+        "logging_strategy": "no",
+        "save_strategy": "no",
+        "save_only_model": True,
         "overwrite_output_dir": True,
         "per_device_train_batch_size": batch_size,
         "max_steps": num_steps,

setup.py

@@ -14,11 +14,12 @@
 
 import os
 import re
+from typing import List
 
 from setuptools import find_packages, setup
 
 
-def get_version():
+def get_version() -> str:
     with open(os.path.join("src", "llamafactory", "extras", "env.py"), "r", encoding="utf-8") as f:
         file_content = f.read()
         pattern = r"{}\W*=\W*\"([^\"]+)\"".format("VERSION")
@@ -26,13 +27,21 @@ def get_version():
         return version
 
 
-def get_requires():
+def get_requires() -> List[str]:
     with open("requirements.txt", "r", encoding="utf-8") as f:
         file_content = f.read()
         lines = [line.strip() for line in file_content.strip().split("\n") if not line.startswith("#")]
         return lines
 
 
+def get_console_scripts() -> List[str]:
+    console_scripts = ["llamafactory-cli = llamafactory.cli:main"]
+    if os.environ.get("ENABLE_SHORT_CONSOLE", "1").lower() in ["true", "1"]:
+        console_scripts.append("lmf = llamafactory.cli:main")
+
+    return console_scripts
+
+
 extra_require = {
     "torch": ["torch>=1.13.1"],
     "torch-npu": ["torch==2.1.0", "torch-npu==2.1.0.post3", "decorator"],
@@ -72,7 +81,7 @@ def main():
         python_requires=">=3.8.0",
         install_requires=get_requires(),
         extras_require=extra_require,
-        entry_points={"console_scripts": ["llamafactory-cli = llamafactory.cli:main"]},
+        entry_points={"console_scripts": get_console_scripts()},
         classifiers=[
             "Development Status :: 4 - Beta",
             "Intended Audience :: Developers",

src/llamafactory/extras/constants.py

@@ -829,7 +829,7 @@ register_model_group(
 
 register_model_group(
     models={
-        "MiniCPM3-4B": {
+        "MiniCPM3-4B-Chat": {
             DownloadSource.DEFAULT: "openbmb/MiniCPM3-4B",
             DownloadSource.MODELSCOPE: "OpenBMB/MiniCPM3-4B",
         },

src/llamafactory/train/callbacks.py

@@ -96,38 +96,45 @@ def fix_valuehead_checkpoint(
 
 
 class FixValueHeadModelCallback(TrainerCallback):
+    r"""
+    A callback for fixing the checkpoint for valuehead models.
+    """
+
     @override
     def on_save(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
         r"""
         Event called after a checkpoint save.
         """
         if args.should_save:
+            output_dir = os.path.join(args.output_dir, "{}-{}".format(PREFIX_CHECKPOINT_DIR, state.global_step))
             fix_valuehead_checkpoint(
-                model=kwargs.pop("model"),
+                model=kwargs.pop("model"), output_dir=output_dir, safe_serialization=args.save_safetensors
-                output_dir=os.path.join(args.output_dir, "{}-{}".format(PREFIX_CHECKPOINT_DIR, state.global_step)),
-                safe_serialization=args.save_safetensors,
             )
 
 
 class SaveProcessorCallback(TrainerCallback):
+    r"""
+    A callback for saving the processor.
+    """
+
     def __init__(self, processor: "ProcessorMixin") -> None:
-        r"""
-        Initializes a callback for saving the processor.
-        """
         self.processor = processor
 
+    @override
+    def on_save(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
+        if args.should_save:
+            output_dir = os.path.join(args.output_dir, "{}-{}".format(PREFIX_CHECKPOINT_DIR, state.global_step))
+            getattr(self.processor, "image_processor").save_pretrained(output_dir)
+
     @override
     def on_train_end(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called at the end of training.
-        """
         if args.should_save:
             getattr(self.processor, "image_processor").save_pretrained(args.output_dir)
 
 
 class PissaConvertCallback(TrainerCallback):
     r"""
-    Initializes a callback for converting the PiSSA adapter to a normal one.
+    A callback for converting the PiSSA adapter to a normal one.
     """
 
     @override
@@ -147,9 +154,6 @@ class PissaConvertCallback(TrainerCallback):
 
     @override
     def on_train_end(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called at the end of training.
-        """
         if args.should_save:
             model = kwargs.pop("model")
             pissa_init_dir = os.path.join(args.output_dir, "pissa_init")
@@ -177,21 +181,22 @@ class PissaConvertCallback(TrainerCallback):
 
 
 class LogCallback(TrainerCallback):
+    r"""
+    A callback for logging training and evaluation status.
+    """
+
     def __init__(self) -> None:
-        r"""
+        # Progress
-        Initializes a callback for logging training and evaluation status.
-        """
-        """ Progress """
         self.start_time = 0
         self.cur_steps = 0
         self.max_steps = 0
         self.elapsed_time = ""
         self.remaining_time = ""
         self.thread_pool: Optional["ThreadPoolExecutor"] = None
-        """ Status """
+        # Status
         self.aborted = False
         self.do_train = False
-        """ Web UI """
+        # Web UI
         self.webui_mode = os.environ.get("LLAMABOARD_ENABLED", "0").lower() in ["true", "1"]
         if self.webui_mode:
             signal.signal(signal.SIGABRT, self._set_abort)
@@ -233,9 +238,6 @@ class LogCallback(TrainerCallback):
 
     @override
     def on_init_end(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called at the end of the initialization of the `Trainer`.
-        """
         if (
             args.should_save
             and os.path.exists(os.path.join(args.output_dir, TRAINER_LOG))
@@ -246,9 +248,6 @@ class LogCallback(TrainerCallback):
 
     @override
     def on_train_begin(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called at the beginning of training.
-        """
         if args.should_save:
             self.do_train = True
             self._reset(max_steps=state.max_steps)
@@ -256,50 +255,32 @@ class LogCallback(TrainerCallback):
 
     @override
     def on_train_end(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called at the end of training.
-        """
         self._close_thread_pool()
 
     @override
     def on_substep_end(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called at the end of an substep during gradient accumulation.
-        """
         if self.aborted:
             control.should_epoch_stop = True
             control.should_training_stop = True
 
     @override
     def on_step_end(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called at the end of a training step.
-        """
         if self.aborted:
             control.should_epoch_stop = True
             control.should_training_stop = True
 
     @override
     def on_evaluate(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called after an evaluation phase.
-        """
         if not self.do_train:
             self._close_thread_pool()
 
     @override
     def on_predict(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called after a successful prediction.
-        """
         if not self.do_train:
             self._close_thread_pool()
 
     @override
     def on_log(self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs):
-        r"""
-        Event called after logging the last logs.
-        """
         if not args.should_save:
             return
 
@@ -342,9 +323,6 @@ class LogCallback(TrainerCallback):
     def on_prediction_step(
         self, args: "TrainingArguments", state: "TrainerState", control: "TrainerControl", **kwargs
     ):
-        r"""
-        Event called after a prediction step.
-        """
         if self.do_train:
             return