CharlesPing/finetuned-ce-climate-multineg-v1

CrossEncoder based on cross-encoder/ms-marco-MiniLM-L12-v2

This is a Cross Encoder model finetuned from cross-encoder/ms-marco-MiniLM-L12-v2 on the climate-cross-encoder-mixed-neg-v3 dataset using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

Model Description

• Model Type: Cross Encoder
• Base model: cross-encoder/ms-marco-MiniLM-L12-v2
• Maximum Sequence Length: 512 tokens
• Number of Output Labels: 1 label
• Training Dataset:
  • climate-cross-encoder-mixed-neg-v3

Model Sources

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

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 CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("CharlesPing/finetuned-ce-climate-multineg-v1")
# Get scores for pairs of texts
pairs = [
    ['Scientific analysis of past climates\xa0shows that greenhouse gasses, principally CO2,\xa0have controlled most ancient\xa0climate changes.', 'Greenhouse gases, in particular carbon dioxide and methane, played a significant role during the Eocene in controlling the surface temperature.'],
    ['Scientific analysis of past climates\xa0shows that greenhouse gasses, principally CO2,\xa0have controlled most ancient\xa0climate changes.', 'Climatic geomorphology is of limited use to study recent (Quaternary, Holocene) large climate changes since there are seldom discernible in the geomorphological record.'],
    ['Scientific analysis of past climates\xa0shows that greenhouse gasses, principally CO2,\xa0have controlled most ancient\xa0climate changes.', 'There is also a close correlation between CO2 and temperature, where CO2 has a strong control over global temperatures in Earth history.'],
    ['Scientific analysis of past climates\xa0shows that greenhouse gasses, principally CO2,\xa0have controlled most ancient\xa0climate changes.', 'While scientists knew of past climate change such as the ice ages, the concept of climate as unchanging was useful in the development of a general theory of what determines climate.'],
    ['Scientific analysis of past climates\xa0shows that greenhouse gasses, principally CO2,\xa0have controlled most ancient\xa0climate changes.', 'Some long term modifications along the history of the planet have been significant, such as the incorporation of oxygen to the atmosphere.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'Scientific analysis of past climates\xa0shows that greenhouse gasses, principally CO2,\xa0have controlled most ancient\xa0climate changes.',
    [
        'Greenhouse gases, in particular carbon dioxide and methane, played a significant role during the Eocene in controlling the surface temperature.',
        'Climatic geomorphology is of limited use to study recent (Quaternary, Holocene) large climate changes since there are seldom discernible in the geomorphological record.',
        'There is also a close correlation between CO2 and temperature, where CO2 has a strong control over global temperatures in Earth history.',
        'While scientists knew of past climate change such as the ice ages, the concept of climate as unchanging was useful in the development of a general theory of what determines climate.',
        'Some long term modifications along the history of the planet have been significant, such as the incorporation of oxygen to the atmosphere.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

• Dataset: climate-rerank-multineg
• Evaluated with CrossEncoderRerankingEvaluator with these parameters:

  {
      "at_k": 1,
      "always_rerank_positives": false
  }
  

Metric	Value
map	0.6809 (-0.3191)
mrr@1	0.6748 (-0.3252)
ndcg@1	0.6748 (-0.3252)

Training Details

Training Dataset

climate-cross-encoder-mixed-neg-v3

• Dataset: climate-cross-encoder-mixed-neg-v3 at cd49b57
• Size: 41,052 training samples
• Columns: query, doc, and label
• Approximate statistics based on the first 1000 samples:

  	query	doc	label
  type	string	string	float
  details	min: 49 characters mean: 140.03 characters max: 306 characters	min: 4 characters mean: 136.03 characters max: 731 characters	min: 0.0 mean: 0.09 max: 1.0

• Samples:

  query	doc	label
  “A leading Canadian authority on polar bears, Mitch Taylor, said: ‘We’re seeing an increase in bears that’s really unprecedented, and in places where we’re seeing a decrease in the population	Warnings about the future of the polar bear are often contrasted with the fact that worldwide population estimates have increased over the past 50 years and are relatively stable today.	1.0
  “A leading Canadian authority on polar bears, Mitch Taylor, said: ‘We’re seeing an increase in bears that’s really unprecedented, and in places where we’re seeing a decrease in the population	Species distribution models of recent years indicate that the deer tick, known as "I. scapularis," is pushing its distribution to higher latitudes of the Northeastern United States and Canada, as well as pushing and maintaining populations in the South Central and Northern Midwest regions of the United States.	0.0
  “A leading Canadian authority on polar bears, Mitch Taylor, said: ‘We’re seeing an increase in bears that’s really unprecedented, and in places where we’re seeing a decrease in the population	Bear and deer are among the animals present.	0.0

• Loss: BinaryCrossEntropyLoss with these parameters:

  {
      "activation_fn": "torch.nn.modules.linear.Identity",
      "pos_weight": null
  }
  

Evaluation Dataset

climate-cross-encoder-mixed-neg-v3

• Dataset: climate-cross-encoder-mixed-neg-v3 at cd49b57
• Size: 4,290 evaluation samples
• Columns: query, doc, and label
• Approximate statistics based on the first 1000 samples:

  	query	doc	label
  type	string	string	float
  details	min: 39 characters mean: 116.67 characters max: 240 characters	min: 18 characters mean: 132.92 characters max: 731 characters	min: 0.0 mean: 0.09 max: 1.0

• Samples:

  query	doc	label
  Scientific analysis of past climates shows that greenhouse gasses, principally CO2, have controlled most ancient climate changes.	Greenhouse gases, in particular carbon dioxide and methane, played a significant role during the Eocene in controlling the surface temperature.	1.0
  Scientific analysis of past climates shows that greenhouse gasses, principally CO2, have controlled most ancient climate changes.	Climatic geomorphology is of limited use to study recent (Quaternary, Holocene) large climate changes since there are seldom discernible in the geomorphological record.	0.0
  Scientific analysis of past climates shows that greenhouse gasses, principally CO2, have controlled most ancient climate changes.	There is also a close correlation between CO2 and temperature, where CO2 has a strong control over global temperatures in Earth history.	0.0

• Loss: BinaryCrossEntropyLoss with these parameters:

  {
      "activation_fn": "torch.nn.modules.linear.Identity",
      "pos_weight": null
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 32
• learning_rate: 2e-05
• warmup_ratio: 0.1
• fp16: 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: 32
• 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: 2e-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
• 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: 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	Validation Loss	climate-rerank-multineg_ndcg@1
0.0390	100	0.5097	-	-
0.0779	200	0.3662	-	-
0.1169	300	0.3034	-	-
0.1559	400	0.2655	-	-
0.1949	500	0.2651	0.2262	0.6585 (-0.3415)
0.2338	600	0.2161	-	-
0.2728	700	0.227	-	-
0.3118	800	0.235	-	-
0.3507	900	0.2243	-	-
0.3897	1000	0.2081	0.2174	0.6992 (-0.3008)
0.4287	1100	0.1961	-	-
0.4677	1200	0.207	-	-
0.5066	1300	0.2375	-	-
0.5456	1400	0.2117	-	-
0.5846	1500	0.2058	0.2253	0.6748 (-0.3252)
0.6235	1600	0.2163	-	-
0.6625	1700	0.2235	-	-
0.7015	1800	0.2193	-	-
0.7405	1900	0.1924	-	-
0.7794	2000	0.2084	0.2095	0.6748 (-0.3252)
0.8184	2100	0.2113	-	-
0.8574	2200	0.2276	-	-
0.8963	2300	0.2071	-	-
0.9353	2400	0.2374	-	-
0.9743	2500	0.2173	0.2172	0.6667 (-0.3333)
1.0133	2600	0.2011	-	-
1.0522	2700	0.1634	-	-
1.0912	2800	0.1807	-	-
1.1302	2900	0.1878	-	-
1.1691	3000	0.2037	0.2147	0.6911 (-0.3089)
1.2081	3100	0.1904	-	-
1.2471	3200	0.1911	-	-
1.2860	3300	0.1828	-	-
1.3250	3400	0.1686	-	-
1.3640	3500	0.1892	0.2179	0.6992 (-0.3008)
1.4030	3600	0.188	-	-
1.4419	3700	0.1691	-	-
1.4809	3800	0.1946	-	-
1.5199	3900	0.1938	-	-
1.5588	4000	0.211	0.2088	0.6992 (-0.3008)
1.5978	4100	0.1826	-	-
1.6368	4200	0.1608	-	-
1.6758	4300	0.1782	-	-
1.7147	4400	0.1803	-	-
1.7537	4500	0.1804	0.2160	0.6911 (-0.3089)
1.7927	4600	0.1823	-	-
1.8316	4700	0.1844	-	-
1.8706	4800	0.1727	-	-
1.9096	4900	0.1937	-	-
1.9486	5000	0.1662	0.2219	0.6829 (-0.3171)
1.9875	5100	0.1653	-	-
2.0265	5200	0.1658	-	-
2.0655	5300	0.1316	-	-
2.1044	5400	0.1379	-	-
2.1434	5500	0.152	0.2513	0.6504 (-0.3496)
2.1824	5600	0.1848	-	-
2.2214	5700	0.1507	-	-
2.2603	5800	0.1495	-	-
2.2993	5900	0.1469	-	-
2.3383	6000	0.1596	0.2407	0.6585 (-0.3415)
2.3772	6100	0.1518	-	-
2.4162	6200	0.1351	-	-
2.4552	6300	0.1706	-	-
2.4942	6400	0.1538	-	-
2.5331	6500	0.1329	0.2505	0.6911 (-0.3089)
2.5721	6600	0.147	-	-
2.6111	6700	0.1289	-	-
2.6500	6800	0.1698	-	-
2.6890	6900	0.1456	-	-
2.7280	7000	0.141	0.2618	0.6748 (-0.3252)
2.7670	7100	0.1413	-	-
2.8059	7200	0.1474	-	-
2.8449	7300	0.1381	-	-
2.8839	7400	0.1252	-	-
2.9228	7500	0.1384	0.2608	0.6748 (-0.3252)
2.9618	7600	0.1826	-	-

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.12
• Sentence Transformers: 4.1.0
• Transformers: 4.51.3
• PyTorch: 2.6.0+cu124
• Accelerate: 1.6.0
• Datasets: 3.6.0
• Tokenizers: 0.21.1

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",
}