ce-forkulous-large

This is the main reranker model for Forkulous, a free-as-in-freedom API for turning unstructured recipe data into nutrient information.

This model accepts an unparsed freetext ingredient line and a description from USDA FoodDataCentral, and outputs a logit that represents the similarity between the two inputs.

This is not a model for sequence/token classification of single ingredient lines.

This is a Cross Encoder model finetuned from cross-encoder/ettin-reranker-150m-v1 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/ettin-reranker-150m-v1
• Maximum Sequence Length: 64 tokens
• Number of Output Labels: 1 label
• Supported Modality: Text

Model Sources

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

Full Model Architecture

CrossEncoder(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'ModernBertModel'})
  (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'cls', 'include_prompt': True})
  (2): Dense({'in_features': 768, 'out_features': 768, 'bias': False, 'activation_function': 'torch.nn.modules.activation.GELU', 'module_input_name': 'sentence_embedding', 'module_output_name': 'sentence_embedding'})
  (3): LayerNorm({'dimension': 768})
  (4): Dense({'in_features': 768, 'out_features': 1, 'bias': True, 'activation_function': 'torch.nn.modules.linear.Identity', 'module_input_name': 'sentence_embedding', 'module_output_name': 'scores'})
)

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("Forkulous/ce-forkulous-large")
# Get scores for pairs of inputs
pairs = [
    ['query: 2/3 cup dark brown sugar packed soft', 'document: Cherries, dark red, sweet, raw'],
    ['query: 1/4 granulated sugar I used raw', 'document: C&H Granulated White Sugar (OK1) - NFY040Y38'],
    ['query: 1 cup yellow squash diced', 'document: Squash, yellow, raw'],
    ['query: 1/4 cup onion tops thinly sliced green', 'document: Asparagus, green, whole spear, raw'],
    ['query: 24 ounces mascarpone cheese chilled', 'document: Cheese, Mexican blend'],
]
scores = model.predict(pairs)
print(scores)
# [-5.9062  1.7188  7.0938 -0.9141  1.1953]

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'query: 2/3 cup dark brown sugar packed soft',
    [
        'document: Cherries, dark red, sweet, raw',
        'document: C&H Granulated White Sugar (OK1) - NFY040Y38',
        'document: Squash, yellow, raw',
        'document: Asparagus, green, whole spear, raw',
        'document: Cheese, Mexican blend',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Correlation

• Evaluated with CrossEncoderCorrelationEvaluator

Metric	Value
pearson	0.8598
spearman	0.9014

Training Details

Training Dataset

Unnamed Dataset

• Size: 7,845 training samples
• Columns: query, document, and score
• Approximate statistics based on the first 100 samples:

  	query	document	score
  type	string	string	float
  modality	text	text
  details	min: 7 tokens mean: 13.04 tokens max: 30 tokens	min: 6 tokens mean: 13.44 tokens max: 27 tokens	min: 0.0 mean: 0.56 max: 1.0

• Samples:

  query	document	score
  query: 2/3 cup potato starch or corn starch	document: FLOUR, CORN, YELLOW (FINE MEAL) (ENRICHED)	0.4
  query: 2 heads butter lettuce chopped	document: Lettuce, for use on a sandwich	1.0
  query: whipped cream or ice cream, for serving	document: Beef with cream or white sauce	0.0

• Loss: BinaryCrossEntropyLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

• Size: 2,055 evaluation samples
• Columns: query, document, and score
• Approximate statistics based on the first 100 samples:

  	query	document	score
  type	string	string	float
  modality	text	text
  details	min: 8 tokens mean: 14.54 tokens max: 37 tokens	min: 6 tokens mean: 14.19 tokens max: 34 tokens	min: 0.0 mean: 0.49 max: 1.0

• Samples:

  query	document	score
  query: 2/3 cup dark brown sugar packed soft	document: Cherries, dark red, sweet, raw	0.0
  query: 1/4 granulated sugar I used raw	document: C&H Granulated White Sugar (OK1) - NFY040Y38	0.9
  query: 1 cup yellow squash diced	document: Squash, yellow, raw	1.0

• Loss: BinaryCrossEntropyLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• num_train_epochs: 4
• learning_rate: 2e-05
• lr_scheduler_type: cosine
• warmup_steps: 0.1
• weight_decay: 0.01
• bf16: True
• per_device_eval_batch_size: 16
• load_best_model_at_end: True
• dataloader_pin_memory: False

All Hyperparameters

Click to expand

• per_device_train_batch_size: 16
• num_train_epochs: 4
• max_steps: -1
• learning_rate: 2e-05
• lr_scheduler_type: cosine
• lr_scheduler_kwargs: None
• warmup_steps: 0.1
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 1
• average_tokens_across_devices: True
• max_grad_norm: 1.0
• label_smoothing_factor: 0.0
• bf16: True
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 16
• prediction_loss_only: True
• eval_on_start: False
• eval_do_concat_batches: True
• eval_use_gather_object: False
• eval_accumulation_steps: None
• include_for_metrics: []
• batch_eval_metrics: False
• save_only_model: False
• save_on_each_node: False
• enable_jit_checkpoint: False
• push_to_hub: False
• hub_private_repo: None
• hub_model_id: Forkulous/ce-forkulous-large
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: True
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 42
• data_seed: None
• use_cpu: False
• 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
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_pin_memory: False
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• dataloader_multiprocessing_context: None
• dataloader_in_order: True
• remove_unused_columns: True
• label_names: None
• train_sampling_strategy: random
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• ddp_static_graph: None
• ddp_backend: None
• ddp_timeout: 1800
• fsdp: None
• fsdp_config: None
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• local_rank: -1
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}
• warmup_ratio: None

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss	spearman
0.0204	10	1.6208	-	-
0.0407	20	1.4429	-	-
0.0611	30	1.0490	-	-
0.0815	40	0.5679	-	-
0.1018	50	0.5497	-	-
0.1222	60	0.5334	-	-
0.1426	70	0.5868	-	-
0.1629	80	0.4883	-	-
0.1833	90	0.4745	-	-
0.2037	100	0.4767	0.4467	0.8302
0.2240	110	0.4124	-	-
0.2444	120	0.4810	-	-
0.2648	130	0.4300	-	-
0.2851	140	0.4689	-	-
0.3055	150	0.3960	-	-
0.3259	160	0.4221	-	-
0.3462	170	0.4342	-	-
0.3666	180	0.4317	-	-
0.3870	190	0.4269	-	-
0.4073	200	0.4022	0.4376	0.8620
0.4277	210	0.4581	-	-
0.4481	220	0.4255	-	-
0.4684	230	0.4297	-	-
0.4888	240	0.4564	-	-
0.5092	250	0.4250	-	-
0.5295	260	0.3977	-	-
0.5499	270	0.4122	-	-
0.5703	280	0.3910	-	-
0.5906	290	0.3930	-	-
0.6110	300	0.4239	0.4146	0.8745
0.6314	310	0.4190	-	-
0.6517	320	0.4212	-	-
0.6721	330	0.4102	-	-
0.6925	340	0.4513	-	-
0.7128	350	0.3958	-	-
0.7332	360	0.4110	-	-
0.7536	370	0.3697	-	-
0.7739	380	0.4526	-	-
0.7943	390	0.4417	-	-
0.8147	400	0.3545	0.4165	0.8824
0.8350	410	0.3983	-	-
0.8554	420	0.3533	-	-
0.8758	430	0.3654	-	-
0.8961	440	0.3874	-	-
0.9165	450	0.3816	-	-
0.9369	460	0.3923	-	-
0.9572	470	0.3717	-	-
0.9776	480	0.4023	-	-
0.9980	490	0.4285	-	-
1.0183	500	0.3794	0.4080	0.8872
1.0387	510	0.3357	-	-
1.0591	520	0.2863	-	-
1.0794	530	0.3658	-	-
1.0998	540	0.3649	-	-
1.1202	550	0.4076	-	-
1.1405	560	0.3354	-	-
1.1609	570	0.3882	-	-
1.1813	580	0.3731	-	-
1.2016	590	0.3672	-	-
1.2220	600	0.3987	0.4114	0.8954
1.2424	610	0.4156	-	-
1.2627	620	0.3384	-	-
1.2831	630	0.3951	-	-
1.3035	640	0.3695	-	-
1.3238	650	0.3838	-	-
1.3442	660	0.3622	-	-
1.3646	670	0.3820	-	-
1.3849	680	0.3633	-	-
1.4053	690	0.3564	-	-
1.4257	700	0.3516	0.4089	0.8893
1.4460	710	0.3841	-	-
1.4664	720	0.3823	-	-
1.4868	730	0.3567	-	-
1.5071	740	0.3644	-	-
1.5275	750	0.4058	-	-
1.5479	760	0.3474	-	-
1.5682	770	0.3692	-	-
1.5886	780	0.3836	-	-
1.6090	790	0.3581	-	-
1.6293	800	0.3675	0.4039	0.8947
1.6497	810	0.3366	-	-
1.6701	820	0.3354	-	-
1.6904	830	0.4028	-	-
1.7108	840	0.3859	-	-
1.7312	850	0.3046	-	-
1.7515	860	0.3429	-	-
1.7719	870	0.3857	-	-
1.7923	880	0.3485	-	-
1.8126	890	0.3832	-	-
1.8330	900	0.4025	0.3989	0.8942
1.8534	910	0.3480	-	-
1.8737	920	0.3625	-	-
1.8941	930	0.3900	-	-
1.9145	940	0.3804	-	-
1.9348	950	0.3413	-	-
1.9552	960	0.3600	-	-
1.9756	970	0.4013	-	-
1.9959	980	0.3806	-	-
2.0163	990	0.3380	-	-
2.0367	1000	0.3269	0.4063	0.8977
2.0570	1010	0.3237	-	-
2.0774	1020	0.3287	-	-
2.0978	1030	0.3342	-	-
2.1181	1040	0.3151	-	-
2.1385	1050	0.3219	-	-
2.1589	1060	0.3601	-	-
2.1792	1070	0.3548	-	-
2.1996	1080	0.3342	-	-
2.2200	1090	0.3918	-	-
2.2403	1100	0.3419	0.3969	0.8975
2.2607	1110	0.3515	-	-
2.2811	1120	0.3274	-	-
2.3014	1130	0.3447	-	-
2.3218	1140	0.3262	-	-
2.3422	1150	0.3223	-	-
2.3625	1160	0.3541	-	-
2.3829	1170	0.3219	-	-
2.4033	1180	0.3526	-	-
2.4236	1190	0.3106	-	-
2.4440	1200	0.3169	0.4025	0.8995
2.4644	1210	0.3134	-	-
2.4847	1220	0.3326	-	-
2.5051	1230	0.3669	-	-
2.5255	1240	0.3433	-	-
2.5458	1250	0.3299	-	-
2.5662	1260	0.3784	-	-
2.5866	1270	0.3436	-	-
2.6069	1280	0.3636	-	-
2.6273	1290	0.2903	-	-
2.6477	1300	0.3264	0.4039	0.8999
2.6680	1310	0.3589	-	-
2.6884	1320	0.3355	-	-
2.7088	1330	0.3434	-	-
2.7291	1340	0.3430	-	-
2.7495	1350	0.3399	-	-
2.7699	1360	0.3572	-	-
2.7902	1370	0.3102	-	-
2.8106	1380	0.3460	-	-
2.8310	1390	0.4116	-	-
2.8513	1400	0.3252	0.3946	0.9005
2.8717	1410	0.3421	-	-
2.8921	1420	0.3196	-	-
2.9124	1430	0.3131	-	-
2.9328	1440	0.4006	-	-
2.9532	1450	0.3267	-	-
2.9735	1460	0.3391	-	-
2.9939	1470	0.3545	-	-
3.0143	1480	0.3362	-	-
3.0346	1490	0.3144	-	-
3.0550	1500	0.3422	0.3990	0.9014

Training Time

• Training: 10.5 minutes
• Evaluation: 3.0 minutes
• Total: 13.6 minutes

Framework Versions

• Python: 3.13.11
• Sentence Transformers: 5.5.1
• Transformers: 5.15.0
• PyTorch: 2.14.0.dev20260708
• Accelerate: 1.14.0
• Datasets: 5.0.1
• Tokenizers: 0.22.2

Additional Resources

• Training and Finetuning Reranker Models with Sentence Transformers: the end-to-end guide for training or finetuning Cross Encoder (reranker) models.
• Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video reranker models through the same API.
• Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: training multimodal Cross Encoders.

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",
}