Qwen3-VL-Reranker-8B LoRA adapter trained on vehicle re-identification hard negatives

Model Details

Model Description

• Model Type: Cross Encoder
• Base model: Qwen/Qwen3-VL-Reranker-8B
• Maximum Sequence Length: 262144 tokens
• Number of Output Labels: 1 label
• Supported Modalities: Text, Image, Video, Message

• Language: en
• License: apache-2.0

Model Sources

• Documentation: Sentence Transformers Documentation
• Documentation: Cross Encoder Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Cross Encoders on Hugging Face

Full Model Architecture

CrossEncoder(
  (0): Transformer({'transformer_task': 'any-to-any', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'logits'}, 'image': {'method': 'forward', 'method_output_name': 'logits'}, 'video': {'method': 'forward', 'method_output_name': 'logits'}, 'message': {'method': 'forward', 'method_output_name': 'logits', 'format': 'structured'}}, 'module_output_name': 'causal_logits', 'processing_kwargs': {'chat_template': {'chat_template': 'reranker', 'add_generation_prompt': True}}, 'architecture': 'Qwen3VLForConditionalGeneration'})
  (1): LogitScore({'true_token_id': 9693, 'false_token_id': 2152, 'module_input_name': 'causal_logits'})
)

Usage

Direct Usage (Sentence Transformers)

Evaluation

Metrics

Cross Encoder Reranking

• Dataset: vehicle-reid-eval-hard
• Evaluated with CrossEncoderRerankingEvaluator with these parameters:
  json

  1{ 2 "at_k": 10 3}

Training Details

Training Dataset

Unnamed Dataset

• Size: 5,766 training samples
• Columns: query, document, and label
• Approximate statistics based on the first 100 samples:

  	query	document	label
  type	string	string	int
  modality	image	image
  details			0: ~82.69% 1: ~17.31%

• Samples:

  query	document	label
  /mnt/smt_nas/huwei/test_data/genvict/qwen3_vl_reranker_8b_hard_negative_dataset/samples/sample_000793/query.jpg	/mnt/smt_nas/huwei/test_data/genvict/qwen3_vl_reranker_8b_hard_negative_dataset/samples/sample_000793/positive.jpg	1
  /mnt/smt_nas/huwei/test_data/genvict/qwen3_vl_reranker_8b_hard_negative_dataset/samples/sample_000793/query.jpg	/mnt/smt_nas/huwei/test_data/genvict/qwen3_vl_reranker_8b_hard_negative_dataset/samples/sample_000793/hard_negative_01.jpg	0
  /mnt/smt_nas/huwei/test_data/genvict/qwen3_vl_reranker_8b_hard_negative_dataset/samples/sample_000793/query.jpg	/mnt/smt_nas/huwei/test_data/genvict/qwen3_vl_reranker_8b_hard_negative_dataset/samples/sample_000793/hard_negative_02.jpg	0

• Loss: BinaryCrossEntropyLoss with these parameters:
  json

  1{ 2 "activation_fn": "torch.nn.modules.linear.Identity", 3 "pos_weight": 5.0 4}

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 1
• num_train_epochs: 2
• warmup_steps: 0.1
• gradient_accumulation_steps: 8
• bf16: True
• gradient_checkpointing: True
• gradient_checkpointing_kwargs: {'use_reentrant': False}
• per_device_eval_batch_size: 1
• save_only_model: True
• load_best_model_at_end: True

All Hyperparameters

Training Logs

• The bold row denotes the saved checkpoint.

Training Time

• Training: 1.3 hours
• Evaluation: 4.9 minutes
• Total: 1.4 hours

Framework Versions

• Python: 3.11.13
• Sentence Transformers: 6.1.0.dev0
• Transformers: 5.15.1
• PyTorch: 2.8.0+cu128
• Accelerate: 1.14.0
• Datasets: 5.0.1
• Tokenizers: 0.22.2

Additional Resources

• Training and Finetuning Reranker Models with Sentence Transformers: the end-to-end guide for training or finetuning Cross Encoder (reranker) models.
• Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video reranker models through the same API.
• Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: training multimodal Cross Encoders.

Citation

BibTeX

Sentence Transformers