[logging] Fix race condition in LoggerHandler during multi-GPU training (#10156)

Co-authored-by: yurekami <yurekami@users.noreply.github.com>

examples/v1/train_full/train_full_fsdp2.yaml

@@ -18,7 +18,7 @@ init_config:
   name: init_on_meta
 
 ### data
-train_dataset: data/v1_sft_demo.yaml 
+train_dataset: data/v1_sft_demo.yaml
 
 ### training
 output_dir: outputs/test_fsdp2

pyproject.toml

@@ -40,10 +40,10 @@ dependencies = [
     "torch>=2.4.0",
     "torchvision>=0.19.0",
     "torchaudio>=2.4.0",
-    "transformers>=4.51.0,<=4.57.1,!=4.52.0,!=4.57.0",
+    "transformers>=4.51.0,<=5.0.0,!=4.52.0,!=4.57.0",
     "datasets>=2.16.0,<=4.0.0",
     "accelerate>=1.3.0,<=1.11.0",
-    "peft>=0.14.0,<=0.17.1",
+    "peft>=0.18.0,<=0.18.1",
     "trl>=0.18.0,<=0.24.0",
     "torchdata>=0.10.0,<=0.11.0",
     # gui

requirements/deepspeed.txt

@@ -1 +1 @@
-deepspeed>=0.10.0,<=0.16.9
+deepspeed>=0.10.0,<=0.18.4

src/llamafactory/data/mm_plugin.py

@@ -2159,6 +2159,40 @@ class LFMVLPlugin(BasePlugin):
         return messages
 
 
+@dataclass
+class YoutuVLPlugin(BasePlugin):
+    r"""Plugin for Youtu-VL vision-language models."""
+
+    vision_bos_token: str = "<|vision_start|>"
+    vision_eos_token: str = "<|vision_end|>"
+
+    @override
+    def process_messages(
+        self,
+        messages: list[dict[str, str]],
+        images: list["ImageInput"],
+        videos: list["VideoInput"],
+        audios: list["AudioInput"],
+        processor: Optional["MMProcessor"],
+    ) -> list[dict[str, str]]:
+        self._validate_input(processor, images, videos, audios)
+        self._validate_messages(messages, images, videos, audios)
+        messages = deepcopy(messages)
+
+        for message in messages:
+            content = message["content"]
+            content = content.replace(
+                IMAGE_PLACEHOLDER, f"{self.vision_bos_token}{self.image_token}{self.vision_eos_token}"
+            )
+            content = content.replace(
+                VIDEO_PLACEHOLDER, f"{self.vision_bos_token}{self.video_token}{self.vision_eos_token}"
+            )
+
+            message["content"] = content
+
+        return messages
+
+
 PLUGINS = {
     "base": BasePlugin,
     "ernie_vl": ErnieVLPlugin,
@@ -2181,6 +2215,7 @@ PLUGINS = {
     "qwen2_vl": Qwen2VLPlugin,
     "qwen3_vl": Qwen3VLPlugin,
     "video_llava": VideoLlavaPlugin,
+    "youtu_vl": YoutuVLPlugin,
 }
 
 

src/llamafactory/data/template.py

@@ -2146,6 +2146,19 @@ register_template(
 )
 
 
+register_template(
+    name="youtu_vl",
+    format_user=StringFormatter(
+        slots=["<|begin_of_text|>user\n{{content}}<|end_of_text|>\n<|begin_of_text|>assistant\n"]
+    ),
+    format_assistant=StringFormatter(slots=["{{content}}<|end_of_text|>\n"]),
+    format_system=StringFormatter(slots=["<|begin_of_text|>system\n{{content}}<|end_of_text|>\n"]),
+    default_system="You are a helpful assistant.",
+    stop_words=["<|end_of_text|>"],
+    mm_plugin=get_mm_plugin(name="youtu_vl", image_token="<|image_pad|>", video_token="<|video_pad|>"),
+)
+
+
 register_template(
     name="yuan",
     format_user=StringFormatter(slots=["{{content}}", {"token": "<sep>"}]),

src/llamafactory/extras/constants.py

@@ -3375,6 +3375,18 @@ register_model_group(
 )
 
 
+register_model_group(
+    models={
+        "Youtu-VL-4B-Instruct": {
+            DownloadSource.DEFAULT: "tencent/Youtu-VL-4B-Instruct",
+            DownloadSource.MODELSCOPE: "Tencent-YouTu-Research/Youtu-VL-4B-Instruct",
+        },
+    },
+    template="youtu_vl",
+    multimodal=True,
+)
+
+
 register_model_group(
     models={
         "Yuan2-2B-Chat": {

src/llamafactory/extras/logging.py

@@ -41,12 +41,13 @@ class LoggerHandler(logging.Handler):
             datefmt="%Y-%m-%d %H:%M:%S",
         )
         self.setLevel(logging.INFO)
+        self.thread_pool = ThreadPoolExecutor(max_workers=1)
         os.makedirs(output_dir, exist_ok=True)
         self.running_log = os.path.join(output_dir, RUNNING_LOG)
-        if os.path.exists(self.running_log):
+        try:
             os.remove(self.running_log)
-
+        except OSError:
-        self.thread_pool = ThreadPoolExecutor(max_workers=1)
+            pass
 
     def _write_log(self, log_entry: str) -> None:
         with open(self.running_log, "a", encoding="utf-8") as f:

src/llamafactory/extras/misc.py

@@ -94,10 +94,10 @@ def check_version(requirement: str, mandatory: bool = False) -> None:
 
 def check_dependencies() -> None:
     r"""Check the version of the required packages."""
-    check_version("transformers>=4.51.0,<=4.57.1")
+    check_version("transformers>=4.51.0,<=5.0.0")
     check_version("datasets>=2.16.0,<=4.0.0")
     check_version("accelerate>=1.3.0,<=1.11.0")
-    check_version("peft>=0.14.0,<=0.17.1")
+    check_version("peft>=0.18.0,<=0.18.1")
     check_version("trl>=0.18.0,<=0.24.0")
 
 
@@ -157,6 +157,33 @@ def get_current_device() -> "torch.device":
     return torch.device(device)
 
 
+def get_device_name() -> str:
+    r"""Get the name of available devices."""
+    if is_torch_xpu_available():
+        device = "xpu"
+    elif is_torch_npu_available():
+        device = "npu"
+    elif is_torch_mps_available():
+        device = "mps"
+    elif is_torch_cuda_available():
+        device = "gpu"
+    else:
+        device = "cpu"
+
+    return device
+
+
+def get_torch_device():
+    r"""Get the torch device namespace for the available devices."""
+    device_name = get_device_name()
+    device_name = "cuda" if device_name == "gpu" else device_name
+    try:
+        return getattr(torch, device_name)
+    except AttributeError:
+        logger.warning_rank0(f"Device namespace '{device_name}' not found in torch, try to load torch.cuda.")
+        return torch.cuda
+
+
 def get_device_count() -> int:
     r"""Get the number of available devices."""
     if is_torch_xpu_available():

src/llamafactory/hparams/data_args.py

@@ -65,7 +65,9 @@ class DataArguments:
     )
     mix_strategy: Literal["concat", "interleave_under", "interleave_over", "interleave_once"] = field(
         default="concat",
-        metadata={"help": "Strategy to use in dataset mixing (concat/interleave) (undersampling/oversampling/sampling w.o. replacement)."},
+        metadata={
+            "help": "Strategy to use in dataset mixing (concat/interleave) (undersampling/oversampling/sampling w.o. replacement)."
+        },
     )
     interleave_probs: str | None = field(
         default=None,

src/llamafactory/hparams/model_args.py

@@ -206,9 +206,6 @@ class BaseModelArguments:
         if self.model_name_or_path is None:
             raise ValueError("Please provide `model_name_or_path`.")
 
-        if self.split_special_tokens and self.use_fast_tokenizer:
-            raise ValueError("`split_special_tokens` is only supported for slow tokenizers.")
-
         if self.adapter_name_or_path is not None:  # support merging multiple lora weights
             self.adapter_name_or_path = [path.strip() for path in self.adapter_name_or_path.split(",")]
 

src/llamafactory/hparams/parser.py

@@ -139,10 +139,6 @@ def _verify_model_args(
         if model_args.adapter_name_or_path is not None and len(model_args.adapter_name_or_path) != 1:
             raise ValueError("Quantized model only accepts a single adapter. Merge them first.")
 
-    if data_args.template == "yi" and model_args.use_fast_tokenizer:
-        logger.warning_rank0("We should use slow tokenizer for the Yi models. Change `use_fast_tokenizer` to False.")
-        model_args.use_fast_tokenizer = False
-
 
 def _check_extra_dependencies(
     model_args: "ModelArguments",
@@ -188,9 +184,7 @@ def _check_extra_dependencies(
 
     if training_args is not None:
         if training_args.deepspeed:
-            # pin deepspeed version < 0.17 because of https://github.com/deepspeedai/DeepSpeed/issues/7347
             check_version("deepspeed", mandatory=True)
-            check_version("deepspeed>=0.10.0,<=0.16.9")
 
         if training_args.predict_with_generate:
             check_version("jieba", mandatory=True)

src/llamafactory/hparams/training_args.py

@@ -14,7 +14,6 @@
 
 import json
 from dataclasses import dataclass, field
-from typing import Literal
 
 from transformers import Seq2SeqTrainingArguments
 from transformers.training_args import _convert_str_dict
@@ -40,56 +39,29 @@ else:
 class RayArguments:
     r"""Arguments pertaining to the Ray training."""
 
-    ray_run_name: str | None = field(
-        default=None,
-        metadata={"help": "The training results will be saved at `<ray_storage_path>/ray_run_name`."},
-    )
-    ray_storage_path: str = field(
-        default="./saves",
-        metadata={"help": "The storage path to save training results to"},
-    )
-    ray_storage_filesystem: Literal["s3", "gs", "gcs"] | None = field(
-        default=None,
-        metadata={"help": "The storage filesystem to use. If None specified, local filesystem will be used."},
-    )
     ray_num_workers: int = field(
         default=1,
         metadata={"help": "The number of workers for Ray training. Default is 1 worker."},
     )
-    resources_per_worker: dict | str = field(
-        default_factory=lambda: {"GPU": 1},
-        metadata={"help": "The resources per worker for Ray training. Default is to use 1 GPU per worker."},
-    )
-    placement_strategy: Literal["SPREAD", "PACK", "STRICT_SPREAD", "STRICT_PACK"] = field(
-        default="PACK",
-        metadata={"help": "The placement strategy for Ray training. Default is PACK."},
-    )
     ray_init_kwargs: dict | str | None = field(
         default=None,
         metadata={"help": "The arguments to pass to ray.init for Ray training. Default is None."},
     )
+    master_addr: str | None = field(
+        default=None,
+        metadata={"help": "The master address for init_process_group"},
+    )
+    master_port: str | None = field(
+        default=None,
+        metadata={"help": "The master port for init_process_group"},
+    )
 
     def __post_init__(self):
         self.use_ray = use_ray()
-        if isinstance(self.resources_per_worker, str) and self.resources_per_worker.startswith("{"):
-            self.resources_per_worker = _convert_str_dict(json.loads(self.resources_per_worker))
 
         if isinstance(self.ray_init_kwargs, str) and self.ray_init_kwargs.startswith("{"):
             self.ray_init_kwargs = _convert_str_dict(json.loads(self.ray_init_kwargs))
 
-        if self.ray_storage_filesystem is not None:
-            if self.ray_storage_filesystem not in ["s3", "gs", "gcs"]:
-                raise ValueError(
-                    f"ray_storage_filesystem must be one of ['s3', 'gs', 'gcs'], got {self.ray_storage_filesystem}."
-                )
-
-            import pyarrow.fs as fs
-
-            if self.ray_storage_filesystem == "s3":
-                self.ray_storage_filesystem = fs.S3FileSystem()
-            elif self.ray_storage_filesystem == "gs" or self.ray_storage_filesystem == "gcs":
-                self.ray_storage_filesystem = fs.GcsFileSystem()
-
 
 @dataclass
 class Fp8Arguments:

src/llamafactory/model/loader.py

@@ -22,7 +22,6 @@ from transformers import (
     AutoModelForImageTextToText,
     AutoModelForSeq2SeqLM,
     AutoModelForTextToWaveform,
-    AutoModelForVision2Seq,
     AutoProcessor,
     AutoTokenizer,
 )
@@ -166,11 +165,9 @@ def load_model(
         else:
             if type(config) in AutoModelForImageTextToText._model_mapping.keys():  # image-text
                 load_class = AutoModelForImageTextToText
-            elif type(config) in AutoModelForVision2Seq._model_mapping.keys():  # image-text
-                load_class = AutoModelForVision2Seq
             elif type(config) in AutoModelForSeq2SeqLM._model_mapping.keys():  # audio-text
                 load_class = AutoModelForSeq2SeqLM
-            elif type(config) in AutoModelForTextToWaveform._model_mapping.keys():  # audio hack for qwen omni
+            elif type(config) in AutoModelForTextToWaveform._model_mapping.keys():  # audio-text for qwen omni
                 load_class = AutoModelForTextToWaveform
             else:
                 load_class = AutoModelForCausalLM

src/llamafactory/model/model_utils/attention.py

@@ -57,6 +57,11 @@ def configure_attn_implementation(config: "PretrainedConfig", model_args: "Model
                 "Gemma-2 should use soft-capping attention, while the SDPA attention does not support it."
             )
 
+    if getattr(config, "model_type", None) in ["youtu", "youtu_vl"]:
+        if model_args.flash_attn in (AttentionFunction.AUTO, AttentionFunction.SDPA):
+            logger.warning_rank0("Youtu-VL does not support SDPA, forcing eager attention.")
+            model_args.flash_attn = AttentionFunction.DISABLED
+
     if model_args.flash_attn == AttentionFunction.AUTO:
         return
 
@@ -85,6 +90,13 @@ def configure_attn_implementation(config: "PretrainedConfig", model_args: "Model
     elif getattr(config, "model_type", None) == "kimi_vl":
         setattr(config.vision_config, "_attn_implementation", requested_attn_implementation)
         setattr(config.text_config, "_attn_implementation", requested_attn_implementation)
+    elif getattr(config, "model_type", None) == "youtu_vl":
+        setattr(config, "attn_implementation", requested_attn_implementation)
+        setattr(config, "_attn_implementation", requested_attn_implementation)
+        if hasattr(config, "vision_config"):
+            setattr(config.vision_config, "_attn_implementation", requested_attn_implementation)
+        if hasattr(config, "text_config"):
+            setattr(config.text_config, "_attn_implementation", requested_attn_implementation)
     else:
         setattr(config, "_attn_implementation", requested_attn_implementation)
 

src/llamafactory/model/model_utils/visual.py

@@ -374,7 +374,13 @@ _register_composite_model(
 _register_composite_model(
     model_type="qwen3_omni_moe_thinker",
     projector_key="visual.merger",
-    vision_model_keys=["visual.pos_embed", "visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list", "audio_tower"],
+    vision_model_keys=[
+        "visual.pos_embed",
+        "visual.patch_embed",
+        "visual.blocks",
+        "visual.deepstack_merger_list",
+        "audio_tower",
+    ],
     language_model_keys=["model", "lm_head"],
     lora_conflict_keys=["patch_embed"],
 )

src/llamafactory/model/patcher.py

@@ -61,6 +61,26 @@ def patch_qwen3_omni_moe_thinker_text_sparse_moe_block():
         modeling_qwen3_omni_moe.Qwen3OmniMoeThinkerTextSparseMoeBlock = Qwen3OmniMoeThinkerTextSparseMoeBlock
 
 
+def patch_youtu_vl_model(model: "PreTrainedModel") -> None:
+    original_forward = model.forward
+
+    def forward(self, *args, **kwargs):
+        outputs = original_forward(*args, **kwargs)
+        if "loss" not in outputs and "labels" in kwargs:
+            logits = outputs.get("logits")
+            labels = kwargs.get("labels")
+            if logits is not None and labels is not None:
+                shift_logits = logits[..., :-1, :].contiguous()
+                shift_labels = labels[..., 1:].contiguous()
+                loss_fct = torch.nn.CrossEntropyLoss()
+                loss = loss_fct(shift_logits.view(-1, self.config.vocab_size), shift_labels.view(-1))
+                outputs["loss"] = loss
+
+        return outputs
+
+    model.forward = MethodType(forward, model)
+
+
 def patch_tokenizer(tokenizer: "PreTrainedTokenizer", model_args: "ModelArguments") -> None:
     if "PreTrainedTokenizerBase" not in str(tokenizer._pad.__func__):
         tokenizer._pad = MethodType(PreTrainedTokenizerBase._pad, tokenizer)
@@ -207,6 +227,9 @@ def patch_model(
         if getattr(model.config, "model_type", None) == "gemma3n":
             setattr(model_args, "disable_gradient_checkpointing", True)
 
+        if getattr(model.config, "model_type", None) == "youtu_vl":
+            patch_youtu_vl_model(model)
+
         prepare_model_for_training(model, model_args)
         autocast_projector_dtype(model, model_args)
         add_z3_leaf_module(model)

src/llamafactory/train/callbacks.py

@@ -103,7 +103,9 @@ class FixValueHeadModelCallback(TrainerCallback):
         if args.should_save:
             output_dir = os.path.join(args.output_dir, f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}")
             fix_valuehead_checkpoint(
-                model=kwargs.pop("model"), output_dir=output_dir, safe_serialization=args.save_safetensors
+                model=kwargs.pop("model"),
+                output_dir=output_dir,
+                safe_serialization=getattr(args, "save_safetensors", True),
             )
 
 
@@ -137,7 +139,7 @@ class PissaConvertCallback(TrainerCallback):
             if isinstance(model, PeftModel):
                 init_lora_weights = getattr(model.peft_config["default"], "init_lora_weights")
                 setattr(model.peft_config["default"], "init_lora_weights", True)
-                model.save_pretrained(pissa_init_dir, safe_serialization=args.save_safetensors)
+                model.save_pretrained(pissa_init_dir, safe_serialization=getattr(args, "save_safetensors", True))
                 setattr(model.peft_config["default"], "init_lora_weights", init_lora_weights)
 
     @override
@@ -155,11 +157,11 @@ class PissaConvertCallback(TrainerCallback):
             if isinstance(model, PeftModel):
                 init_lora_weights = getattr(model.peft_config["default"], "init_lora_weights")
                 setattr(model.peft_config["default"], "init_lora_weights", True)
-                model.save_pretrained(pissa_backup_dir, safe_serialization=args.save_safetensors)
+                model.save_pretrained(pissa_backup_dir, safe_serialization=getattr(args, "save_safetensors", True))
                 setattr(model.peft_config["default"], "init_lora_weights", init_lora_weights)
                 model.save_pretrained(
                     pissa_convert_dir,
-                    safe_serialization=args.save_safetensors,
+                    safe_serialization=getattr(args, "save_safetensors", True),
                     path_initial_model_for_weight_conversion=pissa_init_dir,
                 )
                 model.load_adapter(pissa_backup_dir, "default", is_trainable=True)

src/llamafactory/train/ppo/workflow.py

@@ -72,7 +72,7 @@ def run_ppo(
         ppo_trainer.ppo_train(resume_from_checkpoint=training_args.resume_from_checkpoint)
         ppo_trainer.save_model()
         if training_args.should_save:
-            fix_valuehead_checkpoint(model, training_args.output_dir, training_args.save_safetensors)
+            fix_valuehead_checkpoint(model, training_args.output_dir, getattr(training_args, "save_safetensors", True))
 
         ppo_trainer.save_state()  # must be called after save_model to have a folder
         if ppo_trainer.is_world_process_zero() and finetuning_args.plot_loss:

src/llamafactory/train/rm/trainer.py

@@ -114,7 +114,7 @@ class PairwiseTrainer(Trainer):
         if state_dict is None:
             state_dict = self.model.state_dict()
 
-        if self.args.save_safetensors:
+        if getattr(self.args, "save_safetensors", True):
             from collections import defaultdict
 
             ptrs = defaultdict(list)

src/llamafactory/train/rm/workflow.py

@@ -65,7 +65,7 @@ def run_rm(
         train_result = trainer.train(resume_from_checkpoint=training_args.resume_from_checkpoint)
         trainer.save_model()
         if training_args.should_save:
-            fix_valuehead_checkpoint(model, training_args.output_dir, training_args.save_safetensors)
+            fix_valuehead_checkpoint(model, training_args.output_dir, getattr(training_args, "save_safetensors", True))
 
         trainer.log_metrics("train", train_result.metrics)
         trainer.save_metrics("train", train_result.metrics)

src/llamafactory/train/trainer_utils.py

@@ -20,7 +20,6 @@
 import json
 import os
 from collections.abc import Callable, Mapping
-from pathlib import Path
 from typing import TYPE_CHECKING, Any, Optional, Union
 
 import torch
@@ -34,6 +33,7 @@ from typing_extensions import override
 
 from ..extras import logging
 from ..extras.constants import IGNORE_INDEX, SWANLAB_CONFIG
+from ..extras.misc import get_device_name
 from ..extras.packages import is_apollo_available, is_galore_available, is_ray_available
 from ..hparams import FinetuningArguments, ModelArguments
 from ..model import find_all_linear_modules, load_model, load_tokenizer, load_valuehead_params
@@ -49,15 +49,15 @@ if is_apollo_available():
 
 if is_ray_available():
     import ray
-    from ray.train import RunConfig, ScalingConfig
+    from ray.util.placement_group import PlacementGroup, placement_group
-    from ray.train.torch import TorchTrainer
+    from ray.util.scheduling_strategies import PlacementGroupSchedulingStrategy
 
 
 if TYPE_CHECKING:
     from transformers import PreTrainedModel, TrainerCallback, TrainerState
     from trl import AutoModelForCausalLMWithValueHead
 
-    from ..hparams import DataArguments, RayArguments, TrainingArguments
+    from ..hparams import DataArguments, TrainingArguments
 
 
 logger = logging.get_logger(__name__)
@@ -807,36 +807,88 @@ def get_swanlab_callback(finetuning_args: "FinetuningArguments") -> "TrainerCall
     return swanlab_callback
 
 
-def get_ray_trainer(
+def get_placement_group(num_workers: int) -> tuple["PlacementGroup", dict[str, int]]:
-    training_function: Callable,
+    r"""Get the Ray placement group for distributed training."""
-    train_loop_config: dict[str, Any],
+    bundle = {"CPU": 10}
-    ray_args: "RayArguments",
+    device_name = get_device_name().upper()
-) -> "TorchTrainer":
+    if device_name != "CPU":
-    if not ray_args.use_ray:
+        bundle[device_name] = 1
-        raise ValueError("Ray was not enabled. Please set `USE_RAY=1` to enable ray.")
+    bundles = [bundle for _ in range(num_workers)]
+    pg = placement_group(bundles, strategy="PACK")
 
-    if ray_args.ray_init_kwargs is not None:
+    return pg, bundle
-        ray.init(**ray_args.ray_init_kwargs)
 
-    if ray_args.ray_storage_filesystem is not None:
-        # this means we are using s3/gcs
-        storage_path = ray_args.ray_storage_path
-    else:
-        storage_path = Path(ray_args.ray_storage_path).absolute().as_posix()
 
-    trainer = TorchTrainer(
+def get_ray_remote_config_for_worker(
-        training_function,
+    placement_group: "PlacementGroup",
-        train_loop_config=train_loop_config,
+    bundle_idx: int,
-        scaling_config=ScalingConfig(
+    rank: int,
-            num_workers=ray_args.ray_num_workers,
+    world_size: int,
-            resources_per_worker=ray_args.resources_per_worker,
+    master_addr: str,
-            placement_strategy=ray_args.placement_strategy,
+    master_port: str,
-            use_gpu=True,
+    env: dict[str, str] = None,
+) -> dict[str, Any]:
+    r"""Get the remote config for a Ray worker."""
+    env_vars = {
+        "RANK": str(rank),
+        "WORLD_SIZE": str(world_size),
+        "MASTER_ADDR": master_addr,
+        "MASTER_PORT": master_port,
+        "TORCHELASTIC_USE_AGENT_STORE": "False",
+    }
+    env.update(env_vars)
+
+    remote_config = {
+        "scheduling_strategy": PlacementGroupSchedulingStrategy(
+            placement_group=placement_group,
+            placement_group_bundle_index=bundle_idx,
         ),
-        run_config=RunConfig(
+        "runtime_env": {"env_vars": env},
-            name=ray_args.ray_run_name,
+        "num_cpus": 10,
-            storage_filesystem=ray_args.ray_storage_filesystem,
+    }
-            storage_path=storage_path,
+
-        ),
+    device_name = get_device_name()
-    )
+    if device_name == "gpu":
-    return trainer
+        remote_config["num_gpus"] = 1
+    elif device_name == "npu":
+        remote_config["resources"] = {"NPU": 1}
+
+    return remote_config
+
+
+def get_ray_head_node_ip() -> str:
+    r"""Get the IP address of the Ray head node."""
+    head_ip = next(node["NodeManagerAddress"] for node in ray.nodes() if node.get("IsHead", False))
+    return head_ip
+
+
+def sort_placement_group_by_node_ip(placement_group: "PlacementGroup", master_addr: str = None) -> list[int]:
+    r"""Sort the placement group bundles by their node IP addresses."""
+
+    @ray.remote
+    def _get_node_ip():
+        return ray.util.get_node_ip_address().strip("[]")
+
+    tasks = []
+    for bundle_idx in range(placement_group.bundle_count):
+        task = _get_node_ip.options(
+            scheduling_strategy=PlacementGroupSchedulingStrategy(
+                placement_group=placement_group,
+                placement_group_bundle_index=bundle_idx,
+            ),
+        ).remote()
+        tasks.append(task)
+
+    bundle_ips = ray.get(tasks)
+    bundle_node_ip_list = list(enumerate(bundle_ips))
+
+    sorted_bundle_node_ip_list = sorted(bundle_node_ip_list, key=lambda x: x[1])
+    sorted_bundle_indices = [item[0] for item in sorted_bundle_node_ip_list]
+
+    if master_addr is not None:
+        preferred_indices = [idx for idx, ip in bundle_node_ip_list if ip == master_addr]
+        if preferred_indices:
+            remaining = [i for i in sorted_bundle_indices if i not in preferred_indices]
+            sorted_bundle_indices = preferred_indices + remaining
+
+    return sorted_bundle_indices

src/llamafactory/train/tuner.py

@@ -23,9 +23,9 @@ from transformers import EarlyStoppingCallback, PreTrainedModel
 from ..data import get_template_and_fix_tokenizer
 from ..extras import logging
 from ..extras.constants import V_HEAD_SAFE_WEIGHTS_NAME, V_HEAD_WEIGHTS_NAME
-from ..extras.misc import infer_optim_dtype
+from ..extras.misc import find_available_port, get_device_name, get_torch_device, infer_optim_dtype
 from ..extras.packages import is_mcore_adapter_available, is_ray_available
-from ..hparams import get_infer_args, get_ray_args, get_train_args, read_args
+from ..hparams import RayArguments, get_infer_args, get_ray_args, get_train_args, read_args
 from ..model import load_model, load_tokenizer
 from .callbacks import LogCallback, PissaConvertCallback, ReporterCallback
 from .dpo import run_dpo
@@ -34,12 +34,17 @@ from .ppo import run_ppo
 from .pt import run_pt
 from .rm import run_rm
 from .sft import run_sft
-from .trainer_utils import get_ray_trainer, get_swanlab_callback
+from .trainer_utils import (
+    get_placement_group,
+    get_ray_head_node_ip,
+    get_ray_remote_config_for_worker,
+    get_swanlab_callback,
+    sort_placement_group_by_node_ip,
+)
 
 
 if is_ray_available():
     import ray
-    from ray.train.huggingface.transformers import RayTrainReportCallback
 
 
 if TYPE_CHECKING:
@@ -115,13 +120,7 @@ def run_exp(args: Optional[dict[str, Any]] = None, callbacks: Optional[list["Tra
     ray_args = get_ray_args(args)
     callbacks = callbacks or []
     if ray_args.use_ray:
-        callbacks.append(RayTrainReportCallback())
+        _ray_training_function(ray_args, config={"args": args, "callbacks": callbacks})
-        trainer = get_ray_trainer(
-            training_function=_training_function,
-            train_loop_config={"args": args, "callbacks": callbacks},
-            ray_args=ray_args,
-        )
-        trainer.fit()
     else:
         _training_function(config={"args": args, "callbacks": callbacks})
 
@@ -212,3 +211,94 @@ def export_model(args: Optional[dict[str, Any]] = None) -> None:
     with open(ollama_modelfile, "w", encoding="utf-8") as f:
         f.write(template.get_ollama_modelfile(tokenizer))
         logger.info_rank0(f"Ollama modelfile saved in {ollama_modelfile}")
+
+
+class Worker:
+    def __init__(self):
+        self._setup_env_visible_devices()
+
+        local_rank = os.environ.get("LOCAL_RANK", "0")
+        get_torch_device().set_device(int(local_rank))
+
+    def _setup_env_visible_devices(self) -> None:
+        RAY_NOSET_VISIBLE_DEVICES_LIST = [
+            "RAY_EXPERIMENTAL_NOSET_CUDA_VISIBLE_DEVICES",
+            "RAY_EXPERIMENTAL_NOSET_ASCEND_RT_VISIBLE_DEVICES",
+        ]
+        is_ray_noset_visible_devices = any(os.environ.get(env_var, None) for env_var in RAY_NOSET_VISIBLE_DEVICES_LIST)
+        if is_ray_noset_visible_devices:
+            device_name = get_device_name().upper()
+            local_rank = ray.get_runtime_context().get_accelerator_ids()[device_name][0]
+            os.environ["LOCAL_RANK"] = local_rank
+        else:
+            os.environ["LOCAL_RANK"] = "0"
+
+    def _training_function(self, config: dict[str, Any]) -> None:
+        _training_function(config)
+
+
+def _ray_training_function(ray_args: "RayArguments", config: dict[str, Any]) -> None:
+    num_workers = ray_args.ray_num_workers
+    master_addr = ray_args.master_addr
+    master_port = ray_args.master_port
+    logger.info(f"Using ray.remote mode with {num_workers} workers for distributed training.")
+
+    # initialize ray
+    if not ray.is_initialized():
+        if ray_args.ray_init_kwargs is not None:
+            ray.init(**ray_args.ray_init_kwargs)
+        else:
+            ray.init()
+
+    # verify resources
+    device_name = get_device_name().upper()
+    total_devices = int(ray.cluster_resources().get(device_name, 0))
+    if num_workers > total_devices:
+        raise ValueError(
+            f"The number of devices in the Ray cluster ({total_devices}) should be greater than num_workers ({num_workers})."
+        )
+
+    # verify master_addr
+    if master_addr is None:
+        master_addr = get_ray_head_node_ip()
+        logger.info(f"`master_addr` is not specified, using head node ip: {master_addr}.")
+    else:
+        nodes = [node["NodeManagerAddress"] for node in ray.nodes() if node["Alive"]]
+        if master_addr not in nodes:
+            raise ValueError(f"The `master_addr` ({master_addr}) is not in Ray cluster or not alive ")
+
+    # create placementgroup for resource management
+    pg, bundle = get_placement_group(total_devices)
+    ray.get(pg.ready())
+    logger.info(f"Create placement group with {num_workers} bundles: {bundle}")
+
+    # get sorted_bundle_indices
+    sorted_bundle_indices = sort_placement_group_by_node_ip(pg, master_addr)
+
+    # get master port
+    if master_port is None:
+        master_port = find_available_port()
+        logger.info(f"`master_port` is not specified, using available port: {master_port}.")
+    master_port = str(master_port)
+
+    # backing up environment variables
+    current_env = dict(os.environ.items())
+
+    # launch workers
+    RayWorker = ray.remote(Worker)
+    workers = []
+    for rank in range(num_workers):
+        remote_config = get_ray_remote_config_for_worker(
+            placement_group=pg,
+            bundle_idx=sorted_bundle_indices[rank],
+            rank=rank,
+            world_size=num_workers,
+            master_addr=master_addr,
+            master_port=master_port,
+            env=current_env,
+        )
+        worker = RayWorker.options(**remote_config).remote()
+        workers.append(worker)
+
+    ray.get([worker._training_function.remote(config=config) for worker in workers])
+    ray.shutdown()

tests/conftest.py

@@ -18,7 +18,6 @@ Contains shared fixtures, pytest configuration, and custom markers.
 """
 
 import os
-import sys
 
 import pytest
 import torch
@@ -149,14 +148,7 @@ def _manage_distributed_env(request: FixtureRequest, monkeypatch: MonkeyPatch) -
             devices_str = ",".join(str(i) for i in range(required))
 
         monkeypatch.setenv(env_key, devices_str)
-
+        monkeypatch.syspath_prepend(os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
-        # add project root dir to path for mp run
-        project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
-        if project_root not in sys.path:
-            sys.path.insert(0, project_root)
-
-        os.environ["PYTHONPATH"] = project_root + os.pathsep + os.environ.get("PYTHONPATH", "")
-
     else:  # non-distributed test
         if old_value:
             visible_devices = [v for v in old_value.split(",") if v != ""]

tests/data/processor/test_feedback.py

@@ -20,6 +20,7 @@ from datasets import load_dataset
 from transformers import AutoTokenizer
 
 from llamafactory.extras.constants import IGNORE_INDEX
+from llamafactory.extras.packages import is_transformers_version_greater_than
 from llamafactory.train.test_utils import load_dataset_module
 
 
@@ -52,7 +53,12 @@ def test_feedback_data(num_samples: int):
     for index in indexes:
         messages = original_data["messages"][index]
         ref_input_ids = ref_tokenizer.apply_chat_template(messages)
-        prompt_len = len(ref_tokenizer.apply_chat_template(messages[:-1], add_generation_prompt=True))
+        ref_prompt_ids = ref_tokenizer.apply_chat_template(messages[:-1], add_generation_prompt=True)
+        if is_transformers_version_greater_than("5.0.0"):
+            ref_input_ids = ref_input_ids["input_ids"]
+            ref_prompt_ids = ref_prompt_ids["input_ids"]
+
+        prompt_len = len(ref_prompt_ids)
         ref_labels = [IGNORE_INDEX] * prompt_len + ref_input_ids[prompt_len:]
         assert train_dataset["input_ids"][index] == ref_input_ids
         assert train_dataset["labels"][index] == ref_labels

tests/data/processor/test_pairwise.py

@@ -20,6 +20,7 @@ from datasets import load_dataset
 from transformers import AutoTokenizer
 
 from llamafactory.extras.constants import IGNORE_INDEX
+from llamafactory.extras.packages import is_transformers_version_greater_than
 from llamafactory.train.test_utils import load_dataset_module
 
 
@@ -63,13 +64,21 @@ def test_pairwise_data(num_samples: int):
         rejected_messages = original_data["conversations"][index] + [original_data["rejected"][index]]
         chosen_messages = _convert_sharegpt_to_openai(chosen_messages)
         rejected_messages = _convert_sharegpt_to_openai(rejected_messages)
+
         ref_chosen_input_ids = ref_tokenizer.apply_chat_template(chosen_messages)
-        chosen_prompt_len = len(ref_tokenizer.apply_chat_template(chosen_messages[:-1], add_generation_prompt=True))
+        ref_chosen_prompt_ids = ref_tokenizer.apply_chat_template(chosen_messages[:-1], add_generation_prompt=True)
-        ref_chosen_labels = [IGNORE_INDEX] * chosen_prompt_len + ref_chosen_input_ids[chosen_prompt_len:]
         ref_rejected_input_ids = ref_tokenizer.apply_chat_template(rejected_messages)
-        rejected_prompt_len = len(
+        ref_rejected_prompt_ids = ref_tokenizer.apply_chat_template(rejected_messages[:-1], add_generation_prompt=True)
-            ref_tokenizer.apply_chat_template(rejected_messages[:-1], add_generation_prompt=True)
+
-        )
+        if is_transformers_version_greater_than("5.0.0"):
+            ref_chosen_input_ids = ref_chosen_input_ids["input_ids"]
+            ref_rejected_input_ids = ref_rejected_input_ids["input_ids"]
+            ref_chosen_prompt_ids = ref_chosen_prompt_ids["input_ids"]
+            ref_rejected_prompt_ids = ref_rejected_prompt_ids["input_ids"]
+
+        chosen_prompt_len = len(ref_chosen_prompt_ids)
+        rejected_prompt_len = len(ref_rejected_prompt_ids)
+        ref_chosen_labels = [IGNORE_INDEX] * chosen_prompt_len + ref_chosen_input_ids[chosen_prompt_len:]
         ref_rejected_labels = [IGNORE_INDEX] * rejected_prompt_len + ref_rejected_input_ids[rejected_prompt_len:]
         assert train_dataset["chosen_input_ids"][index] == ref_chosen_input_ids
         assert train_dataset["chosen_labels"][index] == ref_chosen_labels

tests/data/processor/test_supervised.py

@@ -20,6 +20,7 @@ from datasets import load_dataset
 from transformers import AutoTokenizer
 
 from llamafactory.extras.constants import IGNORE_INDEX
+from llamafactory.extras.packages import is_transformers_version_greater_than
 from llamafactory.train.test_utils import load_dataset_module
 
 
@@ -59,7 +60,16 @@ def test_supervised_single_turn(num_samples: int):
             {"role": "assistant", "content": original_data["output"][index]},
         ]
         ref_input_ids = ref_tokenizer.apply_chat_template(messages)
+        ref_prompt_ids = ref_tokenizer.apply_chat_template(messages[:-1], add_generation_prompt=True)
+
+        if is_transformers_version_greater_than("5.0.0"):
+            ref_input_ids = ref_input_ids["input_ids"]
+            ref_prompt_ids = ref_prompt_ids["input_ids"]
+
+        prompt_len = len(ref_prompt_ids)
+        ref_label_ids = [IGNORE_INDEX] * prompt_len + ref_input_ids[prompt_len:]
         assert train_dataset["input_ids"][index] == ref_input_ids
+        assert train_dataset["labels"][index] == ref_label_ids
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
@@ -73,6 +83,10 @@ def test_supervised_multi_turn(num_samples: int):
     indexes = random.choices(range(len(original_data)), k=num_samples)
     for index in indexes:
         ref_input_ids = ref_tokenizer.apply_chat_template(original_data["messages"][index])
+        if is_transformers_version_greater_than("5.0.0"):
+            ref_input_ids = ref_input_ids["input_ids"]
+
+        # cannot test the label ids in multi-turn case
         assert train_dataset["input_ids"][index] == ref_input_ids
 
 
@@ -86,9 +100,12 @@ def test_supervised_train_on_prompt(num_samples: int):
     original_data = load_dataset(DEMO_DATA, name="system_chat", split="train")
     indexes = random.choices(range(len(original_data)), k=num_samples)
     for index in indexes:
-        ref_ids = ref_tokenizer.apply_chat_template(original_data["messages"][index])
+        ref_input_ids = ref_tokenizer.apply_chat_template(original_data["messages"][index])
-        assert train_dataset["input_ids"][index] == ref_ids
+        if is_transformers_version_greater_than("5.0.0"):
-        assert train_dataset["labels"][index] == ref_ids
+            ref_input_ids = ref_input_ids["input_ids"]
+
+        assert train_dataset["input_ids"][index] == ref_input_ids
+        assert train_dataset["labels"][index] == ref_input_ids
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
@@ -103,7 +120,13 @@ def test_supervised_mask_history(num_samples: int):
     for index in indexes:
         messages = original_data["messages"][index]
         ref_input_ids = ref_tokenizer.apply_chat_template(messages)
-        prompt_len = len(ref_tokenizer.apply_chat_template(messages[:-1], add_generation_prompt=True))
+        ref_prompt_ids = ref_tokenizer.apply_chat_template(messages[:-1], add_generation_prompt=True)
+
+        if is_transformers_version_greater_than("5.0.0"):
+            ref_input_ids = ref_input_ids["input_ids"]
+            ref_prompt_ids = ref_prompt_ids["input_ids"]
+
+        prompt_len = len(ref_prompt_ids)
         ref_label_ids = [IGNORE_INDEX] * prompt_len + ref_input_ids[prompt_len:]
         assert train_dataset["input_ids"][index] == ref_input_ids
         assert train_dataset["labels"][index] == ref_label_ids

tests/data/processor/test_unsupervised.py

@@ -19,6 +19,7 @@ import pytest
 from datasets import load_dataset
 from transformers import AutoTokenizer
 
+from llamafactory.extras.packages import is_transformers_version_greater_than
 from llamafactory.train.test_utils import load_dataset_module
 
 
@@ -55,8 +56,13 @@ def test_unsupervised_data(num_samples: int):
     indexes = random.choices(range(len(original_data)), k=num_samples)
     for index in indexes:
         messages = original_data["messages"][index]
-        ref_ids = ref_tokenizer.apply_chat_template(messages)
+        ref_input_ids = ref_tokenizer.apply_chat_template(messages)
-        ref_input_ids = ref_tokenizer.apply_chat_template(messages[:-1], add_generation_prompt=True)
+        ref_prompt_ids = ref_tokenizer.apply_chat_template(messages[:-1], add_generation_prompt=True)
-        ref_labels = ref_ids[len(ref_input_ids) :]
+
-        assert train_dataset["input_ids"][index] == ref_input_ids
+        if is_transformers_version_greater_than("5.0.0"):
+            ref_input_ids = ref_input_ids["input_ids"]
+            ref_prompt_ids = ref_prompt_ids["input_ids"]
+
+        ref_labels = ref_input_ids[len(ref_prompt_ids) :]
+        assert train_dataset["input_ids"][index] == ref_prompt_ids
         assert train_dataset["labels"][index] == ref_labels

tests/data/test_collator.py

@@ -17,7 +17,7 @@ import os
 import pytest
 import torch
 from PIL import Image
-from transformers import AutoConfig, AutoModelForVision2Seq
+from transformers import AutoConfig, AutoModelForImageTextToText
 
 from llamafactory.data import get_template_and_fix_tokenizer
 from llamafactory.data.collator import MultiModalDataCollatorForSeq2Seq, prepare_4d_attention_mask
@@ -82,7 +82,7 @@ def test_multimodal_collator():
     template = get_template_and_fix_tokenizer(tokenizer_module["tokenizer"], data_args)
     config = AutoConfig.from_pretrained(model_args.model_name_or_path)
     with torch.device("meta"):
-        model = AutoModelForVision2Seq.from_config(config)
+        model = AutoModelForImageTextToText.from_config(config)
 
     data_collator = MultiModalDataCollatorForSeq2Seq(
         template=template,

tests/data/test_template.py

@@ -20,6 +20,7 @@ from transformers import AutoTokenizer
 
 from llamafactory.data import get_template_and_fix_tokenizer
 from llamafactory.data.template import parse_template
+from llamafactory.extras.packages import is_transformers_version_greater_than
 from llamafactory.hparams import DataArguments
 
 
@@ -65,7 +66,6 @@ def _check_template(
     template_name: str,
     prompt_str: str,
     answer_str: str,
-    use_fast: bool,
     messages: list[dict[str, str]] = MESSAGES,
 ) -> None:
     r"""Check template.
@@ -75,13 +75,15 @@ def _check_template(
         template_name: the template name.
         prompt_str: the string corresponding to the prompt part.
         answer_str: the string corresponding to the answer part.
-        use_fast: whether to use fast tokenizer.
         messages: the list of messages.
 
     """
-    tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=use_fast, token=HF_TOKEN)
+    tokenizer = AutoTokenizer.from_pretrained(model_id)
     content_str = tokenizer.apply_chat_template(messages, tokenize=False)
     content_ids = tokenizer.apply_chat_template(messages, tokenize=True)
+    if is_transformers_version_greater_than("5.0.0"):
+        content_ids = content_ids["input_ids"]
+
     template = get_template_and_fix_tokenizer(tokenizer, DataArguments(template=template_name))
     prompt_ids, answer_ids = template.encode_oneturn(tokenizer, messages)
     assert content_str == prompt_str + answer_str
@@ -90,9 +92,8 @@ def _check_template(
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
-@pytest.mark.parametrize("use_fast", [True, False])
+def test_encode_oneturn():
-def test_encode_oneturn(use_fast: bool):
+    tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3)
-    tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3, use_fast=use_fast)
     template = get_template_and_fix_tokenizer(tokenizer, DataArguments(template="llama3"))
     prompt_ids, answer_ids = template.encode_oneturn(tokenizer, MESSAGES)
     prompt_str = (
@@ -106,9 +107,8 @@ def test_encode_oneturn(use_fast: bool):
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
-@pytest.mark.parametrize("use_fast", [True, False])
+def test_encode_multiturn():
-def test_encode_multiturn(use_fast: bool):
+    tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3)
-    tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3, use_fast=use_fast)
     template = get_template_and_fix_tokenizer(tokenizer, DataArguments(template="llama3"))
     encoded_pairs = template.encode_multiturn(tokenizer, MESSAGES)
     prompt_str_1 = (
@@ -128,11 +128,10 @@ def test_encode_multiturn(use_fast: bool):
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
-@pytest.mark.parametrize("use_fast", [True, False])
 @pytest.mark.parametrize("cot_messages", [True, False])
 @pytest.mark.parametrize("enable_thinking", [True, False, None])
-def test_reasoning_encode_oneturn(use_fast: bool, cot_messages: bool, enable_thinking: bool):
+def test_reasoning_encode_oneturn(cot_messages: bool, enable_thinking: bool):
-    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B", use_fast=use_fast)
+    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B")
     data_args = DataArguments(template="qwen3", enable_thinking=enable_thinking)
     template = get_template_and_fix_tokenizer(tokenizer, data_args)
     prompt_ids, answer_ids = template.encode_oneturn(tokenizer, MESSAGES_WITH_THOUGHT if cot_messages else MESSAGES)
@@ -155,11 +154,10 @@ def test_reasoning_encode_oneturn(use_fast: bool, cot_messages: bool, enable_thi
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
-@pytest.mark.parametrize("use_fast", [True, False])
 @pytest.mark.parametrize("cot_messages", [True, False])
 @pytest.mark.parametrize("enable_thinking", [True, False, None])
-def test_reasoning_encode_multiturn(use_fast: bool, cot_messages: bool, enable_thinking: bool):
+def test_reasoning_encode_multiturn(cot_messages: bool, enable_thinking: bool):
-    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B", use_fast=use_fast)
+    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B")
     data_args = DataArguments(template="qwen3", enable_thinking=enable_thinking)
     template = get_template_and_fix_tokenizer(tokenizer, data_args)
     encoded_pairs = template.encode_multiturn(tokenizer, MESSAGES_WITH_THOUGHT if cot_messages else MESSAGES)
@@ -185,10 +183,9 @@ def test_reasoning_encode_multiturn(use_fast: bool, cot_messages: bool, enable_t
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
-@pytest.mark.parametrize("use_fast", [True, False])
+def test_jinja_template():
-def test_jinja_template(use_fast: bool):
+    tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3)
-    tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3, use_fast=use_fast)
+    ref_tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3)
-    ref_tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3, use_fast=use_fast)
     template = get_template_and_fix_tokenizer(tokenizer, DataArguments(template="llama3"))
     tokenizer.chat_template = template._get_jinja_template(tokenizer)  # llama3 template no replace
     assert tokenizer.chat_template != ref_tokenizer.chat_template
@@ -222,8 +219,7 @@ def test_get_stop_token_ids():
 
 @pytest.mark.runs_on(["cpu", "mps"])
 @pytest.mark.skipif(not HF_TOKEN, reason="Gated model.")
-@pytest.mark.parametrize("use_fast", [True, False])
+def test_gemma_template():
-def test_gemma_template(use_fast: bool):
     prompt_str = (
         f"<bos><start_of_turn>user\n{MESSAGES[0]['content']}<end_of_turn>\n"
         f"<start_of_turn>model\n{MESSAGES[1]['content']}<end_of_turn>\n"
@@ -231,13 +227,12 @@ def test_gemma_template(use_fast: bool):
         "<start_of_turn>model\n"
     )
     answer_str = f"{MESSAGES[3]['content']}<end_of_turn>\n"
-    _check_template("google/gemma-3-4b-it", "gemma", prompt_str, answer_str, use_fast)
+    _check_template("google/gemma-3-4b-it", "gemma", prompt_str, answer_str)
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
 @pytest.mark.skipif(not HF_TOKEN, reason="Gated model.")
-@pytest.mark.parametrize("use_fast", [True, False])
+def test_gemma2_template():
-def test_gemma2_template(use_fast: bool):
     prompt_str = (
         f"<bos><start_of_turn>user\n{MESSAGES[0]['content']}<end_of_turn>\n"
         f"<start_of_turn>model\n{MESSAGES[1]['content']}<end_of_turn>\n"
@@ -245,13 +240,12 @@ def test_gemma2_template(use_fast: bool):
         "<start_of_turn>model\n"
     )
     answer_str = f"{MESSAGES[3]['content']}<end_of_turn>\n"
-    _check_template("google/gemma-2-2b-it", "gemma2", prompt_str, answer_str, use_fast)
+    _check_template("google/gemma-2-2b-it", "gemma2", prompt_str, answer_str)
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
 @pytest.mark.skipif(not HF_TOKEN, reason="Gated model.")
-@pytest.mark.parametrize("use_fast", [True, False])
+def test_llama3_template():
-def test_llama3_template(use_fast: bool):
     prompt_str = (
         f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n{MESSAGES[0]['content']}<|eot_id|>"
         f"<|start_header_id|>assistant<|end_header_id|>\n\n{MESSAGES[1]['content']}<|eot_id|>"
@@ -259,14 +253,11 @@ def test_llama3_template(use_fast: bool):
         "<|start_header_id|>assistant<|end_header_id|>\n\n"
     )
     answer_str = f"{MESSAGES[3]['content']}<|eot_id|>"
-    _check_template("meta-llama/Meta-Llama-3-8B-Instruct", "llama3", prompt_str, answer_str, use_fast)
+    _check_template("meta-llama/Meta-Llama-3-8B-Instruct", "llama3", prompt_str, answer_str)
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
-@pytest.mark.parametrize(
+def test_llama4_template():
-    "use_fast", [True, pytest.param(False, marks=pytest.mark.xfail(reason="Llama 4 has no slow tokenizer."))]
-)
-def test_llama4_template(use_fast: bool):
     prompt_str = (
         f"<|begin_of_text|><|header_start|>user<|header_end|>\n\n{MESSAGES[0]['content']}<|eot|>"
         f"<|header_start|>assistant<|header_end|>\n\n{MESSAGES[1]['content']}<|eot|>"
@@ -274,18 +265,11 @@ def test_llama4_template(use_fast: bool):
         "<|header_start|>assistant<|header_end|>\n\n"
     )
     answer_str = f"{MESSAGES[3]['content']}<|eot|>"
-    _check_template(TINY_LLAMA4, "llama4", prompt_str, answer_str, use_fast)
+    _check_template(TINY_LLAMA4, "llama4", prompt_str, answer_str)
 
 
-@pytest.mark.parametrize(
-    "use_fast",
-    [
-        pytest.param(True, marks=pytest.mark.xfail(not HF_TOKEN, reason="Authorization.")),
-        pytest.param(False, marks=pytest.mark.xfail(reason="Phi-4 slow tokenizer is broken.")),
-    ],
-)
 @pytest.mark.runs_on(["cpu", "mps"])
-def test_phi4_template(use_fast: bool):
+def test_phi4_template():
     prompt_str = (
         f"<|im_start|>user<|im_sep|>{MESSAGES[0]['content']}<|im_end|>"
         f"<|im_start|>assistant<|im_sep|>{MESSAGES[1]['content']}<|im_end|>"
@@ -293,13 +277,12 @@ def test_phi4_template(use_fast: bool):
         "<|im_start|>assistant<|im_sep|>"
     )
     answer_str = f"{MESSAGES[3]['content']}<|im_end|>"
-    _check_template("microsoft/phi-4", "phi4", prompt_str, answer_str, use_fast)
+    _check_template("microsoft/phi-4", "phi4", prompt_str, answer_str)
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
 @pytest.mark.xfail(not HF_TOKEN, reason="Authorization.")
-@pytest.mark.parametrize("use_fast", [True, False])
+def test_qwen2_5_template():
-def test_qwen2_5_template(use_fast: bool):
     prompt_str = (
         "<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n"
         f"<|im_start|>user\n{MESSAGES[0]['content']}<|im_end|>\n"
@@ -308,13 +291,12 @@ def test_qwen2_5_template(use_fast: bool):
         "<|im_start|>assistant\n"
     )
     answer_str = f"{MESSAGES[3]['content']}<|im_end|>\n"
-    _check_template("Qwen/Qwen2.5-7B-Instruct", "qwen", prompt_str, answer_str, use_fast)
+    _check_template("Qwen/Qwen2.5-7B-Instruct", "qwen", prompt_str, answer_str)
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
-@pytest.mark.parametrize("use_fast", [True, False])
 @pytest.mark.parametrize("cot_messages", [True, False])
-def test_qwen3_template(use_fast: bool, cot_messages: bool):
+def test_qwen3_template(cot_messages: bool):
     prompt_str = (
         f"<|im_start|>user\n{MESSAGES[0]['content']}<|im_end|>\n"
         f"<|im_start|>assistant\n{MESSAGES[1]['content']}<|im_end|>\n"
@@ -328,12 +310,12 @@ def test_qwen3_template(use_fast: bool, cot_messages: bool):
         answer_str = f"{MESSAGES_WITH_THOUGHT[3]['content']}<|im_end|>\n"
         messages = MESSAGES_WITH_THOUGHT
 
-    _check_template("Qwen/Qwen3-8B", "qwen3", prompt_str, answer_str, use_fast, messages=messages)
+    _check_template("Qwen/Qwen3-8B", "qwen3", prompt_str, answer_str, messages=messages)
 
 
 @pytest.mark.runs_on(["cpu", "mps"])
 def test_parse_llama3_template():
-    tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3, token=HF_TOKEN)
+    tokenizer = AutoTokenizer.from_pretrained(TINY_LLAMA3)
     template = parse_template(tokenizer)
     assert template.format_user.slots == [
         "<|start_header_id|>user<|end_header_id|>\n\n{{content}}<|eot_id|>"
@@ -348,7 +330,7 @@ def test_parse_llama3_template():
 @pytest.mark.runs_on(["cpu", "mps"])
 @pytest.mark.xfail(not HF_TOKEN, reason="Authorization.")
 def test_parse_qwen_template():
-    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct", token=HF_TOKEN)
+    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct")
     template = parse_template(tokenizer)
     assert template.__class__.__name__ == "Template"
     assert template.format_user.slots == ["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]
@@ -361,7 +343,7 @@ def test_parse_qwen_template():
 @pytest.mark.runs_on(["cpu", "mps"])
 @pytest.mark.xfail(not HF_TOKEN, reason="Authorization.")
 def test_parse_qwen3_template():
-    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B", token=HF_TOKEN)
+    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B")
     template = parse_template(tokenizer)
     assert template.__class__.__name__ == "ReasoningTemplate"
     assert template.format_user.slots == ["<|im_start|>user\n{{content}}<|im_end|>\n<|im_start|>assistant\n"]

tests/model/model_utils/test_visual.py

@@ -16,7 +16,8 @@ import os
 
 import pytest
 import torch
-from transformers import AutoConfig, AutoModelForVision2Seq
+from safetensors.torch import load_file
+from transformers import AutoConfig, AutoModelForImageTextToText
 
 from llamafactory.extras.packages import is_transformers_version_greater_than
 from llamafactory.hparams import FinetuningArguments, ModelArguments
@@ -36,7 +37,7 @@ def test_visual_full(freeze_vision_tower: bool, freeze_multi_modal_projector: bo
     )
     config = AutoConfig.from_pretrained(model_args.model_name_or_path)
     with torch.device("meta"):
-        model = AutoModelForVision2Seq.from_config(config)
+        model = AutoModelForImageTextToText.from_config(config)
 
     model = init_adapter(config, model, model_args, finetuning_args, is_trainable=True)
     for name, param in model.named_parameters():
@@ -56,7 +57,7 @@ def test_visual_lora(freeze_vision_tower: bool, freeze_language_model: bool):
     )
     config = AutoConfig.from_pretrained(model_args.model_name_or_path)
     with torch.device("meta"):
-        model = AutoModelForVision2Seq.from_config(config)
+        model = AutoModelForImageTextToText.from_config(config)
 
     model = init_adapter(config, model, model_args, finetuning_args, is_trainable=True)
     trainable_params, frozen_params = set(), set()
@@ -86,13 +87,14 @@ def test_visual_model_save_load():
     finetuning_args = FinetuningArguments(finetuning_type="full")
     config = AutoConfig.from_pretrained(model_args.model_name_or_path)
     with torch.device("meta"):
-        model = AutoModelForVision2Seq.from_config(config)
+        model = AutoModelForImageTextToText.from_config(config)
 
     model = init_adapter(config, model, model_args, finetuning_args, is_trainable=False)
+    model.to_empty(device="cpu")
     loaded_model_weight = dict(model.named_parameters())
 
-    model.save_pretrained(os.path.join("output", "qwen2_vl"), max_shard_size="10GB", safe_serialization=False)
+    model.save_pretrained(os.path.join("output", "qwen2_vl"), max_shard_size="10GB", safe_serialization=True)
-    saved_model_weight = torch.load(os.path.join("output", "qwen2_vl", "pytorch_model.bin"), weights_only=False)
+    saved_model_weight = load_file(os.path.join("output", "qwen2_vl", "model.safetensors"))
 
     if is_transformers_version_greater_than("4.52.0"):
         assert "model.language_model.layers.0.self_attn.q_proj.weight" in loaded_model_weight

tests/version.txt

@@ -1,2 +1,2 @@
 # change if test fails or cache is outdated
-0.9.5.105
+0.9.5.106

tests_v1/core/utils/test_rendering.py

@@ -23,6 +23,13 @@ from llamafactory.v1.core.utils.rendering import Renderer
 from llamafactory.v1.utils.types import Processor
 
 
+def _get_input_ids(inputs: list | dict) -> list:
+    if not isinstance(inputs, list):
+        return inputs["input_ids"]
+    else:
+        return inputs
+
+
 HF_MESSAGES = [
     {"role": "system", "content": "You are a helpful assistant."},
     {"role": "user", "content": "What is LLM?"},
@@ -81,15 +88,15 @@ def test_chatml_rendering():
     tokenizer: Processor = AutoTokenizer.from_pretrained("llamafactory/tiny-random-qwen3")
     renderer = Renderer(template="chatml", processor=tokenizer)
 
-    hf_inputs = tokenizer.apply_chat_template(HF_MESSAGES[:-1], add_generation_prompt=True)
+    hf_inputs = _get_input_ids(tokenizer.apply_chat_template(HF_MESSAGES[:-1], add_generation_prompt=True))
     v1_inputs = renderer.render_messages(V1_MESSAGES[:-1], is_generate=True)
     assert v1_inputs["input_ids"] == hf_inputs
     assert v1_inputs["attention_mask"] == [1] * len(hf_inputs)
     assert v1_inputs["labels"] == [-100] * len(hf_inputs)
     assert v1_inputs["loss_weights"] == [0.0] * len(hf_inputs)
 
-    hf_inputs_part = tokenizer.apply_chat_template(HF_MESSAGES[:-1], add_generation_prompt=False)
+    hf_inputs_part = _get_input_ids(tokenizer.apply_chat_template(HF_MESSAGES[:-1], add_generation_prompt=False))
-    hf_inputs_full = tokenizer.apply_chat_template(HF_MESSAGES, add_generation_prompt=False)
+    hf_inputs_full = _get_input_ids(tokenizer.apply_chat_template(HF_MESSAGES, add_generation_prompt=False))
     v1_inputs_full = renderer.render_messages(V1_MESSAGES, is_generate=False)
     assert v1_inputs_full["input_ids"] == hf_inputs_full
     assert v1_inputs_full["attention_mask"] == [1] * len(hf_inputs_full)
@@ -124,17 +131,21 @@ def test_qwen3_nothink_rendering():
     tokenizer: Processor = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
     renderer = Renderer(template="qwen3_nothink", processor=tokenizer)
 
-    hf_inputs = tokenizer.apply_chat_template(HF_MESSAGES_WITH_TOOLS[:-1], tools=V1_TOOLS, add_generation_prompt=True)
+    hf_inputs = _get_input_ids(
+        tokenizer.apply_chat_template(HF_MESSAGES_WITH_TOOLS[:-1], tools=V1_TOOLS, add_generation_prompt=True)
+    )
     v1_inputs = renderer.render_messages(V1_MESSAGES_WITH_TOOLS[:-1], tools=json.dumps(V1_TOOLS), is_generate=True)
     assert v1_inputs["input_ids"] == hf_inputs
     assert v1_inputs["attention_mask"] == [1] * len(hf_inputs)
     assert v1_inputs["labels"] == [-100] * len(hf_inputs)
     assert v1_inputs["loss_weights"] == [0.0] * len(hf_inputs)
 
-    hf_inputs_part = tokenizer.apply_chat_template(
+    hf_inputs_part = _get_input_ids(
-        HF_MESSAGES_WITH_TOOLS[:-1], tools=V1_TOOLS, add_generation_prompt=False
+        tokenizer.apply_chat_template(HF_MESSAGES_WITH_TOOLS[:-1], tools=V1_TOOLS, add_generation_prompt=False)
+    )
+    hf_inputs_full = _get_input_ids(
+        tokenizer.apply_chat_template(HF_MESSAGES_WITH_TOOLS, tools=V1_TOOLS, add_generation_prompt=False)
     )
-    hf_inputs_full = tokenizer.apply_chat_template(HF_MESSAGES_WITH_TOOLS, tools=V1_TOOLS, add_generation_prompt=False)
     v1_inputs_full = renderer.render_messages(V1_MESSAGES_WITH_TOOLS, tools=json.dumps(V1_TOOLS), is_generate=False)
     assert v1_inputs_full["input_ids"] == hf_inputs_full
     assert v1_inputs_full["attention_mask"] == [1] * len(hf_inputs_full)
@@ -187,7 +198,7 @@ def test_qwen3_nothink_rendering_remote(num_samples: int):
 def test_process_sft_samples():
     tokenizer: Processor = AutoTokenizer.from_pretrained("llamafactory/tiny-random-qwen3")
     renderer = Renderer(template="chatml", processor=tokenizer)
-    hf_inputs = tokenizer.apply_chat_template(HF_MESSAGES)
+    hf_inputs = _get_input_ids(tokenizer.apply_chat_template(HF_MESSAGES))
 
     samples = [{"messages": V1_MESSAGES, "extra_info": "test", "_dataset_name": "default"}]
     model_inputs = renderer.process_samples(samples)
@@ -200,7 +211,7 @@ def test_process_sft_samples():
 def test_process_dpo_samples():
     tokenizer: Processor = AutoTokenizer.from_pretrained("llamafactory/tiny-random-qwen3")
     renderer = Renderer(template="chatml", processor=tokenizer)
-    hf_inputs = tokenizer.apply_chat_template(HF_MESSAGES)
+    hf_inputs = _get_input_ids(tokenizer.apply_chat_template(HF_MESSAGES))
 
     samples = [
         {