chatlas/all-mpnet-base-v2-combined_4400-400vs1000

SentenceTransformer based on sentence-transformers/all-mpnet-base-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-mpnet-base-v2. It maps sentences & paragraphs to a 768-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: sentence-transformers/all-mpnet-base-v2
• Maximum Sequence Length: 384 tokens
• Output Dimensionality: 768 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': 384, 'do_lower_case': False, 'architecture': 'MPNetModel'})
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, '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("chatlas/all-mpnet-base-v2-combined_4400-400vs1000")
# Run inference
sentences = [
    'Which file could not be opened according to the xAOD::TFileMerger::addFile error message?',
    'Metadata:\nsource: AtlasTalk\n\nChunk text:\nError in <xAOD::TFileMerger::addFile>: /build1/atnight/localbuilds/nightlies/AnalysisBase-2.3.X/AnalysisBase/rel_nightly/xAODRootAccess/Root/TFileMerger.cxx:105 Couldn\'t open file "user.pottgen.5855794._000003.hist-output.root"',
    "Metadata:\nsource: GitLabMarkdown\nproject path: acc-co/ucap/ucap-core\nproject description: \nfile path: docs/src/docs/reference/device-behavior.md\nheader path: 'Device Behavior' > 'Acquisition properties' > 'First updates'\n\nChunk text:\nAs of May 2024, UCAP retains converter outputs (for each selector) within an in-memory data structure, paired with the\nrelevant selector. Thus, UCAP nodes provide first-updates as needed for `get` and `subscribe` operations; however,",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000,  0.7130, -0.0958],
#         [ 0.7130,  1.0000, -0.1120],
#         [-0.0958, -0.1120,  1.0000]])

Evaluation

Metrics

Information Retrieval

• Dataset: validation
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.745
cosine_accuracy@3	0.8583
cosine_accuracy@5	0.8867
cosine_accuracy@10	0.9183
cosine_precision@1	0.745
cosine_precision@3	0.2861
cosine_precision@5	0.1773
cosine_precision@10	0.0918
cosine_recall@1	0.745
cosine_recall@3	0.8583
cosine_recall@5	0.8867
cosine_recall@10	0.9183
cosine_ndcg@10	0.8348
cosine_mrr@10	0.8078
cosine_map@100	0.8109
dot_accuracy@1	0.745
dot_accuracy@3	0.8583
dot_accuracy@5	0.8867
dot_accuracy@10	0.9183
dot_precision@1	0.745
dot_precision@3	0.2861
dot_precision@5	0.1773
dot_precision@10	0.0918
dot_recall@1	0.745
dot_recall@3	0.8583
dot_recall@5	0.8867
dot_recall@10	0.9183
dot_ndcg@10	0.8348
dot_mrr@10	0.8078
dot_map@100	0.8109

Training Details

Training Dataset

Unnamed Dataset

• Size: 12,000 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 11 tokens mean: 29.43 tokens max: 93 tokens	min: 33 tokens mean: 142.95 tokens max: 356 tokens

• Samples:

  anchor	positive
  On the ATLAS Trigger Developer Pages, what do two-digit version numbers (e.g., 21.3) and three-digit version numbers (e.g., 21.3.9) indicate?	Metadata: source: twiki name: version: 51 last modification: 09-09-2024 category: trigger parents_structure: W, e, b, H, o, m, e, /, A, t, l, a, s, T, r, i, g, g, e, r, /, T, r, i, g, g, e, r, D, e, v, e, l, o, p, e, r, P, a, g, e, s Chunk text: * Two-digit version numbers correspond the branch used to build the nightly (e.g. 21.3) while three digit version numbers correspond to built releases (21.3.9).
  How can I list all available nox sessions using the uv runner?	Metadata: source: GitLabMarkdown project path: particlepredatorinvasion/digout project description: Configurable Python library that automates the conversion of LHCb DIGI files into parquet dataframes by managing a sequence of dependent steps and scheduling their parallel execution on local or distributed systems. file path: docs/source/development/tests.md header path: 'Testing & Automation' > 'Running Sessions' Chunk text: To list all available sessions: bash<br>uv run nox --list<br> To run a specific session: bash<br>uv run nox -s <session_name><br> For example, to run the linter: uv run nox -s lint_check.
  Which setupATLAS -c options will set up the default CentOS6 container used by ATLAS?	Metadata: source: AtlasTalk Chunk text: Answer 5: Hi, You can also do setupATLAS -c centos6 setupATLAS -c sl6 setupATLAS -c rhel6 and it will always setup the default centos6 container that is used by ATLAS.

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 1.0,
      "similarity_fct": "dot_score",
      "gather_across_devices": false
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 1,200 evaluation samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 8 tokens mean: 28.49 tokens max: 100 tokens	min: 38 tokens mean: 142.38 tokens max: 384 tokens

• Samples:

  anchor	positive
  Which copytool was used when the file transfer failed according to the error message?	Metadata: source: AtlasTalk Chunk text: No matching replicas were found in list_replicas() output: [ReplicasNotFound(('No replica found for lfn=panda.0911140145.367865.lib._30337145.30050914397.lib.tgz (allow_lan=True, allow_wan=False)',), {})]:failed to transfer files using copytools=['rucio']
  What are the dimensions of the single conductor wire used in SMC_set10 model set #10?	Metadata: source: GitLabMarkdown project path: steam/analyses/esc-on-smc project description: file path: SMC_set10/README.md header path: 'Model set #10' Chunk text: Its conductor is a single 2 mm * 0.5 mm wire, but in ROXIE it has 4x4 current lines.
  Where should an author go to submit an ATLAS internal note to the CERN Document Server (CDS)?	Metadata: source: twiki name: version: 5 last modification: 19-04-2022 category: pubcom parents_structure: P, u, b, C, o, m Chunk text: For each ATLAS internal note the following should be done: * go to the [[https://cds.cern.ch/submit?ln=en&doctype=ATN][CDS submission page for ATLAS notes]]: =https://cds.cern.ch/submit?ln=en&doctype=ATN=

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 1.0,
      "similarity_fct": "dot_score",
      "gather_across_devices": false
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 4
• learning_rate: 5e-07
• warmup_ratio: 0.1
• fp16: True
• batch_sampler: no_duplicates

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: 4
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 5e-07
• 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
• 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: 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}
• 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
• 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: False
• 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: False
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	validation_cosine_ndcg@10
0.5333	100	2.3423	2.2474	0.7773
1.064	200	2.2441	2.1880	0.8141
1.5973	300	2.208	2.1673	0.8285
2.128	400	2.1906	2.1575	0.8343
2.6613	500	2.1826	2.1530	0.8348

Framework Versions

• Python: 3.12.11
• Sentence Transformers: 5.1.0
• Transformers: 4.55.2
• PyTorch: 2.2.2+cu121
• Accelerate: 1.10.1
• Datasets: 4.0.0
• Tokenizers: 0.21.4

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",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}