music_ent_classification

This model is a fine-tuned version of hfl/chinese-roberta-wwm-ext on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1230
• Accuracy: 0.9662

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 128
• total_eval_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 5.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.4408	1.0	54	0.1763	0.9459
0.1407	2.0	108	0.1221	0.9628
0.0762	3.0	162	0.1123	0.9640
0.0563	4.0	216	0.1226	0.9718
0.0423	5.0	270	0.1230	0.9662

Framework versions

• Transformers 4.57.5
• Pytorch 2.6.0+cu124
• Datasets 2.19.0
• Tokenizers 0.22.2

训练

export WANDB_MODE=disabled # 禁用交互式登录
export CUDA_VISIBLE_DEVICES=0,1,2,3 # 确保识别 4 张 V100
# 变量定义
model="hfl/chinese-roberta-wwm-ext"
transformers_root="transformers"

output_dir="./models/music_ent_classification"
mkdir ${output_dir} -p

# 使用 torchrun 启动
torchrun --nproc_per_node=4 \
    ${transformers_root}/examples/pytorch/text-classification/run_classification.py \
    --model_name_or_path ${model} \
    --train_file "./data/*.train.json" \
    --validation_file "./data/*.test.json" \
    --trust_remote_code True \
    --do_train \
    --do_eval \
    --shuffle_train_dataset \
    --metric_name accuracy \
    --text_column_name sentence1 \
    --label_column_name label \
    --max_seq_length 256 \
    --per_device_train_batch_size 32 \
    --learning_rate 2e-5 \
    --num_train_epochs 5 \
    --logging_steps 50 \
    --save_strategy epoch \
    --eval_strategy epoch \
    --fp16 True \
    --output_dir ${output_dir} \
    --overwrite_output_dir 

推理

# 启动分布式推理
torchrun --nproc_per_node=$(echo $CUDA_DEVICES | tr ',' '\n' | wc -l) \
    transformers/examples/pytorch/text-classification/run_classification.py \
    --model_name_or_path "${MODEL_PATH}" \
    --train_file "${TRAIN_DATA}" \
    --validation_file "${TRAIN_DATA}" \
    --test_file "${INPUT_FILE}" \
    --text_column_name "sentence1" \
    --label_column_name "label" \
    --do_predict \
    --max_seq_length 128 \
    --per_device_eval_batch_size 256 \
    --output_dir "${OUTPUT_DIR}" \
    --fp16 True \
    --trust_remote_code True \
    --overwrite_output_dir

if [ $? -eq 0 ]; then
    echo "✅ [Infer] 推理完成。"
else
    echo "❌ [Infer] 推理失败。"
    exit 1
fi