Mercity/memory-retrieval-qwen3-0.6b-lora

SentenceTransformer based on Qwen/Qwen3-Embedding-0.6B

This is a sentence-transformers model finetuned from Qwen/Qwen3-Embedding-0.6B. It maps sentences & paragraphs to a 1024-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-0.6B
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 1024 dimensions
• Similarity Function: Cosine Similarity

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': 512, 'do_lower_case': False, 'architecture': 'PeftModelForFeatureExtraction'})
  (1): Pooling({'word_embedding_dimension': 1024, '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("Mercity/memory-retrieval-qwen3-0.6b-lora")
# Run inference
queries = [
    "Freelancer\u0027s on board, but I\u0027m anxious about the budget stretching for these fixes.",
]
documents = [
    'Victor strongly values authenticity and has explicitly stated he would rather deliver 4 excellent, deeply resonant ads than 6 mediocre ones that rely on viral gimmicks.',
    'Carlos successfully mastered the past tense (Passé Composé) last month by focusing solely on verb conjugation tables rather than contextual sentences.',
    'Raj recently won a $500 gift certificate specifically for experiences/tours redeemable through a regional travel aggregator.',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 1024] [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.3463, -0.1885, -0.0302]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 369,891 training samples
• Columns: sentence_0, sentence_1, and sentence_2
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1	sentence_2
  type	string	string	string
  details	min: 9 tokens mean: 29.84 tokens max: 67 tokens	min: 16 tokens mean: 34.6 tokens max: 66 tokens	min: 16 tokens mean: 30.9 tokens max: 56 tokens

• Samples:

  sentence_0	sentence_1	sentence_2
  Our budget's stretched thin with the condo mortgage and student loans, yet we're locked into this $4k Ireland trip next summer—am I being irresponsible by not prioritizing career growth over these family splurges?	Liam's father, who was rarely present due to travel, recently reconciled with Liam and offered to pay for 100% of Ava's future college tuition, regardless of Liam's career path.	Liam is extremely dedicated to his career and has always prioritized professional advancement over personal comfort.
  I've got $1800 left after flights for my Tokyo conference trip—can you recommend budget vegan hotels close to the venue to keep things simple?	Lena's primary networking goal for this trip is securing a mentorship with Kenji Tanaka, a speaker at the conference known for hosting small, private dinners for potential mentees.	Lena is intensely private about her financial struggles and views asking for help or discussing debt openly as a personal failure.
  Inflation's killing my shop, but I wanna surprise Lena with milestone magic on a dime.	Last year, Marcus successfully organized a surprise 'Family Fun Day' at a local park for Lena's birthday, which she cited as the best day they'd had since before Jamal got sick, despite being low-cost.	Maria recently spent an hour reviewing the history of Mandarin tonal development since the Han Dynasty.

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.COSINE",
      "triplet_margin": 0.5
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• num_train_epochs: 5
• fp16: True
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 5e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1
• num_train_epochs: 5
• max_steps: -1
• lr_scheduler_type: linear
• 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: True
• 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: True
• 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: True
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• 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: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.0865	500	0.1927
0.1730	1000	0.1023
0.2595	1500	0.0841
0.3460	2000	0.0772
0.4325	2500	0.0704
0.5190	3000	0.0635
0.6055	3500	0.0576
0.6920	4000	0.0545
0.7785	4500	0.0522
0.8651	5000	0.0484
0.9516	5500	0.0465
1.0	5780	-
1.0381	6000	0.0388
1.1246	6500	0.0323
1.2111	7000	0.0318
1.2976	7500	0.0314
1.3841	8000	0.0294
1.4706	8500	0.0301
1.5571	9000	0.0283
1.6436	9500	0.0272
1.7301	10000	0.0247
1.8166	10500	0.0248
1.9031	11000	0.0233
1.9896	11500	0.0228
2.0	11560	-
2.0761	12000	0.0154
2.1626	12500	0.0141
2.2491	13000	0.0146
2.3356	13500	0.0135
2.4221	14000	0.0143
2.5087	14500	0.0139
2.5952	15000	0.014
2.6817	15500	0.0128
2.7682	16000	0.0126
2.8547	16500	0.0122
2.9412	17000	0.0114
3.0	17340	-
3.0277	17500	0.0104
3.1142	18000	0.007
3.2007	18500	0.0069
3.2872	19000	0.0067
3.3737	19500	0.0064
3.4602	20000	0.0067
3.5467	20500	0.0063
3.6332	21000	0.0061
3.7197	21500	0.0058
3.8062	22000	0.0058
3.8927	22500	0.0051
3.9792	23000	0.005
4.0	23120	-
4.0657	23500	0.0035
4.1522	24000	0.0027
4.2388	24500	0.0028
4.3253	25000	0.0025
4.4118	25500	0.0025
4.4983	26000	0.0024

Framework Versions

• Python: 3.12.3
• Sentence Transformers: 5.1.2
• Transformers: 4.57.1
• PyTorch: 2.8.0+cu128
• Accelerate: 1.11.0
• Datasets: 4.4.1
• Tokenizers: 0.22.1

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",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}