README.md

metadata

language:
  - en
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - dense
  - generated_from_trainer
  - dataset_size:197462
  - loss:MSELoss
base_model: Qwen/Qwen3-Embedding-0.6B
widget:
  - source_sentence: >-
      Instruct: Given a web search query, retrieve relevant passages that answer
      the query

      Query:who sings the song i don't want to work
    sentences:
      - >-
        The Invisible Man Griffin is the surname of the story's protagonist. His
        name is not mentioned until about halfway through the book. Consumed
        with his greed for power and fame, he is the model of science without
        humanity. A gifted young student, he becomes interested in the science
        of refraction. During his experiments, he accidentally discovers
        chemicals (combined with an unspecified kind of radiation) that would
        make living tissue invisible. Obsessed with his discovery, he tries the
        experiment on himself and becomes invisible. However, he does not know
        how to reverse the process, and he slowly discovers that the advantages
        of being invisible do not outweigh the disadvantages and the problems he
        faces. Thus begins his downfall as he takes the road to crime for his
        survival, revealing in the process his lack of conscience, inhumanity
        and complete selfishness. He progresses from obsession to fanaticism, to
        insanity, and finally to his fateful end.
      - >-
        Instruct: Given a web search query, retrieve relevant passages that
        answer the query

        Query:who did the united states become independent from
      - >-
        Jordan Belfort Jordan Ross Belfort (/ˈbɛlfɔːrt/; born July 9, 1962) is
        an American author, motivational speaker, and former stockbroker. In
        1999, he pleaded guilty to fraud and related crimes in connection with
        stock-market manipulation and running a boiler room as part of a
        penny-stock scam. Belfort spent 22 months in prison as part of an
        agreement under which he gave testimony against numerous partners and
        subordinates in his fraud scheme.[5] He published the memoir The Wolf of
        Wall Street, which was adapted into a film and released in 2013.
  - source_sentence: >-
      London water supply infrastructure Most of London's water comes from
      non-tidal parts of the Thames and Lea, with the remainder being abstracted
      from underground sources.[22]
    sentences:
      - >-
        Instruct: Given a web search query, retrieve relevant passages that
        answer the query

        Query:what is the number on the hogwarts express
      - >-
        Instruct: Given a web search query, retrieve relevant passages that
        answer the query

        Query:when did roughing the kicker become a rule
      - >-
        Agora Early in Greek history (18th century–8th century BC), free-born
        citizens would gather in the agora for military duty or to hear
        statements of the ruling king or council. Later, the Agora also served
        as a marketplace where merchants kept stalls or shops to sell their
        goods amid colonnades. This attracted artisans who built workshops
        nearby.[2]
  - source_sentence: >-
      Instruct: Given a web search query, retrieve relevant passages that answer
      the query

      Query:what is meant by lagging and leading current in ac circuit
    sentences:
      - >-
        .org The domain name org is a generic top-level domain (gTLD) of the
        Domain Name System (DNS) used in the Internet. The name is truncated
        from organization. It was one of the original domains established in
        1985, and has been operated by the Public Interest Registry since 2003.
        The domain was originally intended for non-profit entities, but this
        restriction was not enforced and has been removed. The domain is
        commonly used by schools, open-source projects, and communities, but
        also by some for-profit entities. The number of registered domains in
        org has increased from fewer than one million in the 1990s, to ten
        million as of June 2013.
      - >-
        Instruct: Given a web search query, retrieve relevant passages that
        answer the query

        Query:how many episode in season 1 game of thrones
      - >-
        Instruct: Given a web search query, retrieve relevant passages that
        answer the query

        Query:when is season 11 of doctor who coming out
  - source_sentence: >-
      Gabriel Vlad (born April 9, 1969) in Bucharest, is a former Romanian
      former rugby union football player.
    sentences:
      - >-
        As of May 2013, The Jewish Tribune had a circulation of 60,500 copies a
        week which made it, for a time, the largest Jewish weekly publication in
        Canada.
      - >-
        Cunjamba Dima is a city and commune of Angola, located in the province
        of Cuando Cubango.
      - >-
        He also acted in the National award winning Tamil movie Vazhakku Enn
        18/9, directed by Balaji Sakthivel.
  - source_sentence: The actress was thirteen when she was offered the role of Annie.
    sentences:
      - >-
        All profits from the sale and streaming of the song go to music
        education supported by the CMA Foundation.
      - >-
        Narsingh Temple is situated at the across of the village just across
        confluence of Magri State village.
      - >-
        Contrasting significantly from other soccer leagues in the U.S., WLS
        intends to be an open entry, promotion and relegation competition.
datasets:
  - sentence-transformers/natural-questions
  - sentence-transformers/gooaq
  - sentence-transformers/wikipedia-en-sentences
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
  - negative_mse
model-index:
  - name: SentenceTransformer based on Qwen/Qwen3-Embedding-0.6B
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: NanoMSMARCO
          type: NanoMSMARCO
        metrics:
          - type: cosine_accuracy@1
            value: 0.42
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.64
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.76
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.82
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.42
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.21333333333333335
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.15200000000000002
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08199999999999999
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.42
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.64
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.76
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.82
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.620918816092183
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.5567777777777778
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.5664067325709117
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: NanoNFCorpus
          type: NanoNFCorpus
        metrics:
          - type: cosine_accuracy@1
            value: 0.38
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.44
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.52
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.66
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.38
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.31333333333333335
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.29200000000000004
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.254
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.041275151654868704
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.06868331254409366
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.08524350018847202
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.11409038508225758
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.30429750607308503
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.44163492063492066
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.1254808602198398
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: NanoNQ
          type: NanoNQ
        metrics:
          - type: cosine_accuracy@1
            value: 0.4
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.72
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.76
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.82
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.4
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.24666666666666665
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.16
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.088
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.39
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.69
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.73
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.79
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.6214012092294585
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.571047619047619
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.564828869259454
            name: Cosine Map@100
      - task:
          type: nano-beir
          name: Nano BEIR
        dataset:
          name: NanoBEIR mean
          type: NanoBEIR_mean
        metrics:
          - type: cosine_accuracy@1
            value: 0.4000000000000001
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.6
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.68
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.7666666666666666
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.4000000000000001
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.25777777777777783
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.20133333333333336
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.1413333333333333
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.2837583838849562
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.4662277708480312
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.5250811667294907
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.5746967950274192
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.5155391771315755
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.5231534391534391
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.41890548735006855
            name: Cosine Map@100
      - task:
          type: knowledge-distillation
          name: Knowledge Distillation
        dataset:
          name: Unknown
          type: unknown
        metrics:
          - type: negative_mse
            value: -0.016825005412101746
            name: Negative Mse

SentenceTransformer based on Qwen/Qwen3-Embedding-0.6B

This is a sentence-transformers model finetuned from Qwen/Qwen3-Embedding-0.6B on the nq dataset. 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.

This is a first experiment attempting to distil the powerful Qwen/Qwen3-Embedding-0.6B model from 28 layers down to some lower layer count, in an attempt to speed up inference with minimal performance reductions.

To be specific, this early model falls from ~0.55 NDCG@10 average across NanoMSMARCO, NanoNFCorpus, and NanoNQ with the full 28 layers, to 0.5155 NDCG@10 on that selection with just 18 layers. Early tests indicate that using only 18 layers results in a 1.51x speedup compared to the full model.

This model was distilled using only 200k texts from one domain, reaching superior performance should be possible, especially with stronger distillation techniques like MarginMSE.

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
• Training Dataset:
  • nq
• 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': 512, 'do_lower_case': False}) with Transformer model: Qwen3Model 
  (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("tomaarsen/Qwen3-Embedding-0.6B-18-layers")
# Run inference
sentences = [
    'The actress was thirteen when she was offered the role of Annie.',
    'Contrasting significantly from other soccer leagues in the U.S., WLS intends to be an open entry, promotion and relegation competition.',
    'Narsingh Temple is situated at the across of the village just across confluence of Magri State village.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

• Datasets: NanoMSMARCO, NanoNFCorpus and NanoNQ
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "query_prompt": "Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery:"
  }
  

Metric	NanoMSMARCO	NanoNFCorpus	NanoNQ
cosine_accuracy@1	0.42	0.38	0.4
cosine_accuracy@3	0.64	0.44	0.72
cosine_accuracy@5	0.76	0.52	0.76
cosine_accuracy@10	0.82	0.66	0.82
cosine_precision@1	0.42	0.38	0.4
cosine_precision@3	0.2133	0.3133	0.2467
cosine_precision@5	0.152	0.292	0.16
cosine_precision@10	0.082	0.254	0.088
cosine_recall@1	0.42	0.0413	0.39
cosine_recall@3	0.64	0.0687	0.69
cosine_recall@5	0.76	0.0852	0.73
cosine_recall@10	0.82	0.1141	0.79
cosine_ndcg@10	0.6209	0.3043	0.6214
cosine_mrr@10	0.5568	0.4416	0.571
cosine_map@100	0.5664	0.1255	0.5648

Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with NanoBEIREvaluator with these parameters:

  {
      "dataset_names": [
          "msmarco",
          "nfcorpus",
          "nq"
      ],
      "query_prompts": {
          "msmarco": "Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery:",
          "nfcorpus": "Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery:",
          "nq": "Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery:"
      }
  }
  

Metric	Value
cosine_accuracy@1	0.4
cosine_accuracy@3	0.6
cosine_accuracy@5	0.68
cosine_accuracy@10	0.7667
cosine_precision@1	0.4
cosine_precision@3	0.2578
cosine_precision@5	0.2013
cosine_precision@10	0.1413
cosine_recall@1	0.2838
cosine_recall@3	0.4662
cosine_recall@5	0.5251
cosine_recall@10	0.5747
cosine_ndcg@10	0.5155
cosine_mrr@10	0.5232
cosine_map@100	0.4189

Knowledge Distillation

• Evaluated with MSEEvaluator

Metric	Value
negative_mse	-0.0168

Training Details

Training Dataset

nq

• Dataset: nq at f9e894e
• Size: 197,462 training samples
• Columns: text and label
• Approximate statistics based on the first 1000 samples:

  	text	label
  type	string	list
  details	min: 27 tokens mean: 89.38 tokens max: 505 tokens	size: 1024 elements

• Samples:

  text	label
  Instruct: Given a web search query, retrieve relevant passages that answer the query Query:the movie bernie based on a true story	[-0.05126953125, -0.0020294189453125, 0.00152587890625, 0.060791015625, 0.022216796875, ...]
  College World Series The College World Series, or CWS, is an annual June baseball tournament held in Omaha, Nebraska. The CWS is the culmination of the National Collegiate Athletic Association (NCAA) Division I Baseball Championship tournament—featuring 64 teams in the first round—which determines the NCAA Division I college baseball champion. The eight participating teams are split into two, four-team, double-elimination brackets, with the winners of each bracket playing in a best-of-three championship series.	[0.033935546875, -0.0908203125, -0.010498046875, 0.0625, -0.01263427734375, ...]
  Instruct: Given a web search query, retrieve relevant passages that answer the query Query:does the femoral nerve turn into the saphenous nerve	[0.052978515625, -0.0028228759765625, -0.0022430419921875, 0.0732421875, 0.044677734375, ...]

• Loss: MSELoss

Evaluation Datasets

nq

• Dataset: nq at f9e894e
• Size: 3,000 evaluation samples
• Columns: text and label
• Approximate statistics based on the first 1000 samples:

  	text	label
  type	string	list
  details	min: 21 tokens mean: 87.24 tokens max: 410 tokens	size: 1024 elements

• Samples:

  text	label
  Instruct: Given a web search query, retrieve relevant passages that answer the query Query:who was the heir apparent of the austro-hungarian empire in 1914	[0.0262451171875, 0.0556640625, -0.0, -0.03076171875, -0.05712890625, ...]
  Instruct: Given a web search query, retrieve relevant passages that answer the query Query:who played tommy in coward of the county	[-0.00848388671875, -0.02294921875, -0.00182342529296875, 0.060546875, -0.021240234375, ...]
  Vertebra The vertebral arch is formed by pedicles and laminae. Two pedicles extend from the sides of the vertebral body to join the body to the arch. The pedicles are short thick processes that extend, one from each side, posteriorly, from the junctions of the posteriolateral surfaces of the centrum, on its upper surface. From each pedicle a broad plate, a lamina, projects backwards and medialwards to join and complete the vertebral arch and form the posterior border of the vertebral foramen, which completes the triangle of the vertebral foramen.[6] The upper surfaces of the laminae are rough to give attachment to the ligamenta flava. These ligaments connect the laminae of adjacent vertebra along the length of the spine from the level of the second cervical vertebra. Above and below the pedicles are shallow depressions called vertebral notches (superior and inferior). When the vertebrae articulate the notches align with those on adjacent vertebrae and these form the openings of the int...	[0.062255859375, -0.005706787109375, -0.009765625, 0.035400390625, -0.0125732421875, ...]

• Loss: MSELoss

gooaq

• Dataset: gooaq at b089f72
• Size: 3,000 evaluation samples
• Columns: text and label
• Approximate statistics based on the first 1000 samples:

  	text	label
  type	string	list
  details	min: 10 tokens mean: 43.88 tokens max: 117 tokens	size: 1024 elements

• Samples:

  text	label
  Instruct: Given a web search query, retrieve relevant passages that answer the query Query:what essential oils are soothing?	[-0.025146484375, 0.06591796875, -0.0025634765625, 0.0732421875, -0.046630859375, ...]
  Titles of books should be underlined or put in italics . (Titles of stories, essays and poems are in "quotation marks.") Refer to the text specifically as a novel, story, essay, memoir, or poem, depending on what it is.	[-0.006988525390625, -0.050537109375, -0.007476806640625, -0.07177734375, -0.049560546875, ...]
  Dakine Cyclone Wet/Dry 32L Backpack. Born from the legacy of our most iconic surf pack, the Cyclone Collection is a family of super-technical and durable wet/dry packs and bags.	[0.0016632080078125, 0.04150390625, -0.01324462890625, 0.0234375, 0.03173828125, ...]

• Loss: MSELoss

wikipedia

• Dataset: wikipedia at 4a0972d
• Size: 3,000 evaluation samples
• Columns: text and label
• Approximate statistics based on the first 1000 samples:

  	text	label
  type	string	list
  details	min: 5 tokens mean: 28.1 tokens max: 105 tokens	size: 1024 elements

• Samples:

  text	label
  The daughter of Vice-admiral George Davies and Julia Hume, she spent her younger years on board the ship he was stationed, the Griper.	[0.0361328125, 0.01904296875, -0.003662109375, 0.0247802734375, 0.0140380859375, ...]
  The impetus for the project began when Amalgamated Dynamics, hired to provide the practical effects for The Thing, a prequel to John Carpenter's 1982 classic film-renowned for its almost exclusive use of practical effects-became disillusioned upon discovering the theatrical release had the bulk of their effects digitally replaced with computer-generated imagery.	[-0.0106201171875, -0.0439453125, -0.01104736328125, 0.00946044921875, 0.0322265625, ...]
  Lost Angeles, his second feature film, starring Joelle Carter and Kelly Blatz, had its world premiere at the Oldenburg International Film Festival in 2012.	[0.0272216796875, 0.0263671875, -0.007110595703125, 0.0294189453125, 0.01129150390625, ...]

• Loss: MSELoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• learning_rate: 0.0001
• num_train_epochs: 1
• warmup_ratio: 0.1
• bf16: True
• load_best_model_at_end: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• 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: 0.0001
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 1
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.1
• 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
• use_ipex: False
• bf16: True
• 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: True
• label_names: None
• load_best_model_at_end: True
• 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}
• tp_size: 0
• 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}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• 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
• 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: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	nq loss	gooaq loss	wikipedia loss	NanoMSMARCO_cosine_ndcg@10	NanoNFCorpus_cosine_ndcg@10	NanoNQ_cosine_ndcg@10	NanoBEIR_mean_cosine_ndcg@10	negative_mse
-1	-1	-	-	-	-	0.2033	0.0972	0.1638	0.1548	-0.0985
0.0162	200	0.0008	-	-	-	-	-	-	-	-
0.0324	400	0.0004	-	-	-	-	-	-	-	-
0.0486	600	0.0003	-	-	-	-	-	-	-	-
0.0648	800	0.0003	-	-	-	-	-	-	-	-
0.0810	1000	0.0002	0.0002	0.0003	0.0003	0.5482	0.2864	0.5995	0.4780	-0.0280
0.0972	1200	0.0002	-	-	-	-	-	-	-	-
0.1134	1400	0.0002	-	-	-	-	-	-	-	-
0.1296	1600	0.0002	-	-	-	-	-	-	-	-
0.1458	1800	0.0002	-	-	-	-	-	-	-	-
0.1620	2000	0.0002	0.0002	0.0003	0.0003	0.6136	0.2926	0.6028	0.5030	-0.0218
0.1783	2200	0.0002	-	-	-	-	-	-	-	-
0.1945	2400	0.0001	-	-	-	-	-	-	-	-
0.2107	2600	0.0001	-	-	-	-	-	-	-	-
0.2269	2800	0.0001	-	-	-	-	-	-	-	-
0.2431	3000	0.0001	0.0001	0.0002	0.0002	0.6169	0.2990	0.5781	0.4980	-0.0199
0.2593	3200	0.0001	-	-	-	-	-	-	-	-
0.2755	3400	0.0001	-	-	-	-	-	-	-	-
0.2917	3600	0.0001	-	-	-	-	-	-	-	-
0.3079	3800	0.0001	-	-	-	-	-	-	-	-
0.3241	4000	0.0001	0.0001	0.0002	0.0002	0.6137	0.3000	0.5987	0.5041	-0.0187
0.3403	4200	0.0001	-	-	-	-	-	-	-	-
0.3565	4400	0.0001	-	-	-	-	-	-	-	-
0.3727	4600	0.0001	-	-	-	-	-	-	-	-
0.3889	4800	0.0001	-	-	-	-	-	-	-	-
0.4051	5000	0.0001	0.0001	0.0002	0.0002	0.6235	0.2945	0.6105	0.5095	-0.0182
0.4213	5200	0.0001	-	-	-	-	-	-	-	-
0.4375	5400	0.0001	-	-	-	-	-	-	-	-
0.4537	5600	0.0001	-	-	-	-	-	-	-	-
0.4699	5800	0.0001	-	-	-	-	-	-	-	-
0.4861	6000	0.0001	0.0001	0.0002	0.0002	0.6183	0.2999	0.6141	0.5108	-0.0175
0.5023	6200	0.0001	-	-	-	-	-	-	-	-
0.5186	6400	0.0001	-	-	-	-	-	-	-	-
0.5348	6600	0.0001	-	-	-	-	-	-	-	-
0.5510	6800	0.0001	-	-	-	-	-	-	-	-
0.5672	7000	0.0001	0.0001	0.0002	0.0002	0.6129	0.3005	0.6201	0.5112	-0.0173
0.5834	7200	0.0001	-	-	-	-	-	-	-	-
0.5996	7400	0.0001	-	-	-	-	-	-	-	-
0.6158	7600	0.0001	-	-	-	-	-	-	-	-
0.6320	7800	0.0001	-	-	-	-	-	-	-	-
0.6482	8000	0.0001	0.0001	0.0002	0.0002	0.6258	0.3032	0.6099	0.5130	-0.0170
0.6644	8200	0.0001	-	-	-	-	-	-	-	-
0.6806	8400	0.0001	-	-	-	-	-	-	-	-
0.6968	8600	0.0001	-	-	-	-	-	-	-	-
0.7130	8800	0.0001	-	-	-	-	-	-	-	-
0.7292	9000	0.0001	0.0001	0.0002	0.0002	0.6209	0.3043	0.6214	0.5155	-0.0168
0.7454	9200	0.0001	-	-	-	-	-	-	-	-
0.7616	9400	0.0001	-	-	-	-	-	-	-	-
0.7778	9600	0.0001	-	-	-	-	-	-	-	-
0.7940	9800	0.0001	-	-	-	-	-	-	-	-
0.8102	10000	0.0001	0.0001	0.0002	0.0002	0.6224	0.3015	0.6183	0.5141	-0.0168
0.8264	10200	0.0001	-	-	-	-	-	-	-	-
0.8427	10400	0.0001	-	-	-	-	-	-	-	-
0.8589	10600	0.0001	-	-	-	-	-	-	-	-
0.8751	10800	0.0001	-	-	-	-	-	-	-	-
0.8913	11000	0.0001	0.0001	0.0002	0.0002	0.6224	0.3014	0.6155	0.5131	-0.0167
0.9075	11200	0.0001	-	-	-	-	-	-	-	-
0.9237	11400	0.0001	-	-	-	-	-	-	-	-
0.9399	11600	0.0001	-	-	-	-	-	-	-	-
0.9561	11800	0.0001	-	-	-	-	-	-	-	-
0.9723	12000	0.0001	0.0001	0.0002	0.0002	0.6247	0.3020	0.6133	0.5133	-0.0167
0.9885	12200	0.0001	-	-	-	-	-	-	-	-
-1	-1	-	-	-	-	0.6209	0.3043	0.6214	0.5155	-0.0168

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.10
• Sentence Transformers: 4.2.0.dev0
• Transformers: 4.51.2
• PyTorch: 2.5.1+cu124
• Accelerate: 1.5.2
• Datasets: 3.5.0
• Tokenizers: 0.21.0

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",
}

MSELoss

@inproceedings{reimers-2020-multilingual-sentence-bert,
    title = "Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2020",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/2004.09813",
}