zyc-zju/qwen3-embedding-4b_qwen3-8b_nq_lsr

SentenceTransformer based on Qwen/Qwen3-Embedding-4B

This is a sentence-transformers model finetuned from Qwen/Qwen3-Embedding-4B on the flash_rag_datasets dataset. It maps sentences & paragraphs to a 2560-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: Qwen/Qwen3-Embedding-4B
• Maximum Sequence Length: 40960 tokens
• Output Dimensionality: 2560 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • flash_rag_datasets
• Language: en

Model Sources

• Documentation: Sentence Transformers Documentation
• Repository: Sentence Transformers on GitHub
• Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 40960, 'do_lower_case': False, 'architecture': 'PeftModelForFeatureExtraction'})
  (1): Pooling({'word_embedding_dimension': 2560, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': True, 'include_prompt': True})
  (2): Normalize()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("qwen3-embedding-4b_qwen3-8b_nq_lsr")
# Run inference
queries = [
    "when did captain crunch oops all berries come out",
]
documents = [
    'First released in 1997',
    'Shel Silverstein',
    'Notre Dame',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 2560] [3, 2560]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.2967, 0.2414, 0.1715]])

Training Details

Training Dataset

flash_rag_datasets

• Dataset: flash_rag_datasets at bcafb8d
• Size: 79,168 training samples
• Columns: query and response
• Approximate statistics based on the first 1000 samples:

  	query	response
  type	string	string
  details	min: 9 tokens mean: 11.35 tokens max: 25 tokens	min: 2 tokens mean: 5.31 tokens max: 16 tokens

• Samples:

  query	response
  total number of death row inmates in the us	2,718
  big little lies season 2 how many episodes	seven
  who sang waiting for a girl like you	Foreigner

• Loss: fed_rag.loss.pytorch.lsr.LSRLoss

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 1
• gradient_accumulation_steps: 16
• learning_rate: 1e-05
• max_steps: 100
• lr_scheduler_type: constant
• remove_unused_columns: False
• dataloader_pin_memory: False
• push_to_hub: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 1
• per_device_eval_batch_size: 8
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 16
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 1e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 3.0
• max_steps: 100
• lr_scheduler_type: constant
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• bf16: False
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: False
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• project: huggingface
• trackio_space_id: trackio
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: False
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: True
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• hub_revision: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.0002	1	0.0004
0.0004	2	0.0004
0.0006	3	0.0008
0.0008	4	0.0003
0.0010	5	0.0002
0.0012	6	0.0002
0.0014	7	0.0003
0.0016	8	0.0004
0.0018	9	0.0007
0.0020	10	0.0003
0.0022	11	0.0007
0.0024	12	0.0003
0.0026	13	0.0004
0.0028	14	0.0004
0.0030	15	0.0004
0.0032	16	0.0006
0.0034	17	0.0005
0.0036	18	0.0005
0.0038	19	0.0003
0.0040	20	0.0006
0.0042	21	0.0003
0.0044	22	0.0004
0.0046	23	0.0004
0.0049	24	0.0003
0.0051	25	0.0004
0.0053	26	0.0005
0.0055	27	0.0004
0.0057	28	0.0003
0.0059	29	0.0003
0.0061	30	0.0005
0.0063	31	0.0004
0.0065	32	0.0003
0.0067	33	0.0003
0.0069	34	0.0007
0.0071	35	0.0002
0.0073	36	0.0003
0.0075	37	0.0003
0.0077	38	0.0004
0.0079	39	0.0011
0.0081	40	0.0004
0.0083	41	0.0004
0.0085	42	0.0002
0.0087	43	0.0003
0.0089	44	0.0004
0.0091	45	0.0003
0.0093	46	0.0004
0.0095	47	0.0006
0.0097	48	0.0004
0.0099	49	0.0003
0.0101	50	0.0003
0.0103	51	0.0004
0.0105	52	0.0002
0.0107	53	0.0003
0.0109	54	0.0003
0.0111	55	0.0004
0.0113	56	0.0009
0.0115	57	0.0012
0.0117	58	0.0003
0.0119	59	0.0003
0.0121	60	0.0004
0.0123	61	0.0005
0.0125	62	0.0006
0.0127	63	0.0003
0.0129	64	0.0004
0.0131	65	0.0004
0.0133	66	0.0005
0.0135	67	0.0003
0.0137	68	0.0006
0.0139	69	0.0004
0.0141	70	0.0003
0.0143	71	0.0005
0.0146	72	0.0003
0.0148	73	0.0003
0.0150	74	0.0003
0.0152	75	0.0004
0.0154	76	0.0005
0.0156	77	0.0002
0.0158	78	0.0005
0.0160	79	0.0003
0.0162	80	0.0003
0.0164	81	0.0004
0.0166	82	0.0005
0.0168	83	0.0003
0.0170	84	0.0003
0.0172	85	0.0003
0.0174	86	0.0004
0.0176	87	0.0001
0.0178	88	0.0004
0.0180	89	0.0004
0.0182	90	0.0003
0.0184	91	0.0005
0.0186	92	0.0003
0.0188	93	0.0003
0.0190	94	0.0003
0.0192	95	0.0003
0.0194	96	0.0005
0.0196	97	0.0006
0.0198	98	0.0003
0.0200	99	0.0003
0.0202	100	0.0004

Framework Versions

• Python: 3.11.14
• Sentence Transformers: 5.2.0
• Transformers: 4.57.2
• PyTorch: 2.9.1+cu128
• Accelerate: 1.12.0
• Datasets: 4.5.0
• Tokenizers: 0.22.2

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}