huangsukai/finetuned-sentence-encoder-for-pddl-gen

SentenceTransformer based on sentence-transformers/all-roberta-large-v1

This is a sentence-transformers model finetuned from sentence-transformers/all-roberta-large-v1 on the json 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.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: sentence-transformers/all-roberta-large-v1
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 1024 tokens
• Similarity Function: Cosine Similarity
• Training Dataset:
  • json

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': 256, 'do_lower_case': False}) with Transformer model: RobertaModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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("huangsukai/finetuned-sentence-encoder-for-pddl-gen")
# Run inference
sentences = [
    "Context: Translate the given natural language description into an action schema that includes the parameters, preconditions, and effects. Ensure that only the provided predicates are used to construct the preconditions and effects.\nQuestion: Here is the task.\nA natural language description of the domain\nDomain information: Help the hero to get out of dungeon! A hero woke up in a dungeon full of monsters and traps (perhaps the party last night went wrong...) and needs your help to get out.  Here are basic facts for the dungeon domain: - The dungeon contains rooms that are **connected** by corridors (dungeon can thus be represented by undirected graph) - each room can be **empty**, or can have a **monster** in it, or can have a **trap** in it, or can have a **sword** in it - one of the empty rooms is the **goal**: it has an exit, so the hero can escape\n\nA list of available predicates\n1. (at-hero ?loc - cells) ;;  Hero's cell location\n2. (at-sword ?s - swords ?loc - cells) ;; Sword cell location\n3. (has-monster ?loc - cells) ;; Indicates if a cell location has a monster\n4. (has-trap ?loc - cells) ;; Indicates if a cell location has a trap\n5. (is-destroyed ?obj) ;; Indicates if a chell or sword has been destroyed\n6. (connected ?from ?to - cells) ;; connects cells\n7. (arm-free) ;; Hero's hand is free\n8. (holding ?s - swords) ;; Hero's holding a sword\n9. (trap-disarmed ?loc) ;; It becomes true when a trap is disarmed\n\nAction Description: **Disarm a trap** – if there is a trap in the room the hero is in and the hero is empty-handed (does not hold a sword), then the hero can disarm it\n\nAction name: disarm-trap\n\n\nYour answer:\n",
    'Parameters:\n1. ?loc - cells\n\nPreconditions:\n```\n(and\n    (at-hero ?loc)\n    (arm-free)\n    (has-trap ?loc)\n)\n```\n\nEffects:\n```\n(and\n    (trap-disarmed ?loc)\n    (not (has-trap ?loc))\n)\n```\n',
    'Parameters:\n1. ?from - cells\n2. ?to - cells\n\nPreconditions:\n```\n(and\n    (connected ?from ?to)\n    (at-hero ?from)\n    (not (has-trap ?from))\n    (not (is-destroyed ?to))\n    (not (has-trap ?to))\n    (not (has-monster ?to))\n)\n```\n\nEffects:\n```\n(and\n    (at-hero ?to)\n    (is-destroyed ?from)\n    (not (at-hero ?from))\n)\n```\n',
]
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]

Training Details

Training Dataset

json

• Dataset: json
• Size: 200,010 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 246 tokens mean: 255.74 tokens max: 256 tokens	min: 71 tokens mean: 138.42 tokens max: 248 tokens	min: 62 tokens mean: 132.66 tokens max: 250 tokens

• Samples:

  anchor	positive	negative
  Context: Translate the given natural language description into an action schema that includes the parameters, preconditions, and effects. Ensure that only the provided predicates are used to construct the preconditions and effects. Question: Here is the task. A natural language description of the domain Domain information: This domain is structured to allow organizing and managing books within a library setting. The actions and predicates support the movement of books between tables and shelves, ensuring that conditions like accessibility and the librarian's hands being free are met. Additionally, it includes managing book categories, shelf space, and check-out/return processes to reflect a more complex library system. A list of available predicates 1. (on-shelf ?x ?y - book) ;; ?x is on top of ?y on the shelf 2. (on-table ?x - book) ;; ?x is on the table 3. (accessible ?x - book) ;; ?x is accessible (not covered) 4. (hands-free) ;; The hands of the librarian are free 5. (holding ?x - book) ;; The librarian is holding ?x 6. (belongs-to-category ?x - book ?cat - category) ;; ?x belongs to the category ?cat 7. (shelf-empty ?cat - category) ;; The shelf for category ?cat is empty 8. (shelf-overflow ?cat - category) ;; The shelf for category ?cat is full 9. (book-request ?book - book) ;; There is a request for book ?book 10. (return-due ?book - book) ;; Book ?book is due for return 11. (checked-out ?book - book) ;; Book ?book is checked out Action Description: Mark a book as borrowed by a patron, ensuring it's not already taken. Action name: check-out Your answer:	Parameters: 1. ?x - book Preconditions: <br>(and<br> (accessible ?x)<br> (not (checked-out ?x))<br>)<br> Effects: <br>(and<br> (checked-out ?x)<br> (not (accessible ?x))<br> (book-request ?x)<br> (return-due ?x)<br>)<br>	Parameters: 1. ?x - book Preconditions: <br>(and<br> (checked-out ?x)<br> (accessible ?x)<br>)<br> Effects: <br>(and<br> (checked-out ?x)<br> (book-request ?x)<br> (return-due ?x)<br>)<br>
  Context: Translate the given natural language description into an action schema that includes the parameters, preconditions, and effects. Ensure that only the provided predicates are used to construct the preconditions and effects. Question: Here is the task. A natural language description of the domain Domain information: This describes a cooking or baking process where ingredients such as eggs and flour are used. The process of baking involves putting eggs in the pan followed by the flour, mix the two and then put the pan in the oven and then remove the pan from the oven to get the baked cake. Lastly, the pan is cleaned using a soap. Well, you can also bake a souffle but in at bit different process. A list of available predicates 1. (is_egg ?egg - ingredient) ;; the ingredient is an egg 2. (is_flour ?flour - ingredient) ;; the ingredient is flour 3. (pan_has_egg ?pan - pan) ;; the pan has an egg 4. (pan_has_flour ?pan - pan) ;; the pan has flour 5. (pan_is_clean ?pan - pan) ;; the pan is clean 6. (pan_in_oven ?pan - pan) ;; the pan is in the oven 7. (in_pan ?x - ingredient ?pan - pan) ;; the ingredient is in the pan 8. (in_oven ?pan - pan ?oven - oven) ;; the pan is in the oven 9. (oven_is_full ?oven - oven) ;; the oven is full 10. (hypothetical ?new - ingredient) ;; the ingredient is hypothetical 11. (is_mixed ?pan - pan) ;; the ingredients in the pan are mixed 12. (is_cake ?new - ingredient) ;; the ingredient is a cake 13. (is_souffle ?new - ingredient) ;; the ingredient is a souffle 14. (soap_consumed ?soap - soap) ;; the soap is consumed Action Description: Action putting pan in oven needs oven and a pan. The pan is put in oven which is not full. Action name: put_pan_in_oven Your answer:	Parameters: 1. ?pan - pan 2. ?oven - oven Preconditions: <br>(and<br> (not (oven_is_full ?oven))<br> (not (pan_in_oven ?pan))<br>)<br> Effects: <br>(and<br> (oven_is_full ?oven)<br> (in_oven ?pan ?oven)<br> (pan_in_oven ?pan)<br>)<br>	Parameters: 1. ?pan - pan 2. ?oven - oven Preconditions: <br>(and<br> (not (oven_is_full ?oven))<br> (not (pan_in_oven ?pan))<br>)<br> Effects: <br>(and<br> (oven_is_full ?oven)<br> (in_oven ?pan ?oven)<br>)<br>
  Context: Translate the given natural language description into an action schema that includes the parameters, preconditions, and effects. Ensure that only the provided predicates are used to construct the preconditions and effects. Question: Here is the task. A natural language description of the domain Domain information: Help the hero to get out of dungeon! A hero woke up in a dungeon full of monsters and traps (perhaps the party last night went wrong...) and needs your help to get out. Here are basic facts for the dungeon domain: - The dungeon contains rooms that are connected by corridors (dungeon can thus be represented by undirected graph) - each room can be empty, or can have a monster in it, or can have a trap in it, or can have a sword in it - one of the empty rooms is the goal: it has an exit, so the hero can escape A list of available predicates 1. (at-hero ?loc - cells) ;; Hero's cell location 2. (at-sword ?s - swords ?loc - cells) ;; Sword cell location 3. (has-monster ?loc - cells) ;; Indicates if a cell location has a monster 4. (has-trap ?loc - cells) ;; Indicates if a cell location has a trap 5. (is-destroyed ?obj) ;; Indicates if a chell or sword has been destroyed 6. (connected ?from ?to - cells) ;; connects cells 7. (arm-free) ;; Hero's hand is free 8. (holding ?s - swords) ;; Hero's holding a sword 9. (trap-disarmed ?loc) ;; It becomes true when a trap is disarmed Action Description: Hero can move if the - hero is at current location - cells are connected, - there is no trap in current loc, and - destination does not have a trap/monster/has-been-destroyed Effects move the hero, and destroy the original cell. No need to destroy the sword. Action name: move Your answer:	Parameters: 1. ?from - cells 2. ?to - cells Preconditions: <br>(and<br> (connected ?from ?to)<br> (at-hero ?from)<br> (not (has-trap ?from))<br> (not (is-destroyed ?to))<br> (not (has-trap ?to))<br> (not (has-monster ?to))<br>)<br> Effects: <br>(and<br> (at-hero ?to)<br> (is-destroyed ?from)<br> (not (at-hero ?from))<br>)<br>	Parameters: 1. ?from - cells 2. ?to - cells Preconditions: <br>(and<br> (connected ?from ?to)<br> (at-hero ?from)<br> (not (has-trap ?from))<br> (not (is-destroyed ?to))<br> (not (has-trap ?to))<br>)<br> Effects: <br>(and<br> (at-hero ?to)<br> (is-destroyed ?from)<br> (not (at-hero ?from))<br>)<br>

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: T5EncoderModel 
    (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): Dense({'in_features': 1024, 'out_features': 768, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
    (3): Normalize()
  ), 'temperature': 0.01}
  

Evaluation Dataset

json

• Dataset: json
• Size: 34 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 34 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 256 tokens mean: 256.0 tokens max: 256 tokens	min: 82 tokens mean: 127.94 tokens max: 176 tokens	min: 64 tokens mean: 121.15 tokens max: 176 tokens

• Samples:

  anchor	positive	negative
  Context: Translate the given natural language description into an action schema that includes the parameters, preconditions, and effects. Ensure that only the provided predicates are used to construct the preconditions and effects. Question: Here is the task. A natural language description of the domain Domain information: This domain is structured to allow organizing and managing books within a library setting. The actions and predicates support the movement of books between tables and shelves, ensuring that conditions like accessibility and the librarian's hands being free are met. Additionally, it includes managing book categories, shelf space, and check-out/return processes to reflect a more complex library system. A list of available predicates 1. (on-shelf ?x ?y - book) ;; ?x is on top of ?y on the shelf 2. (on-table ?x - book) ;; ?x is on the table 3. (accessible ?x - book) ;; ?x is accessible (not covered) 4. (hands-free) ;; The hands of the librarian are free 5. (holding ?x - book) ;; The librarian is holding ?x 6. (belongs-to-category ?x - book ?cat - category) ;; ?x belongs to the category ?cat 7. (shelf-empty ?cat - category) ;; The shelf for category ?cat is empty 8. (shelf-overflow ?cat - category) ;; The shelf for category ?cat is full 9. (book-request ?book - book) ;; There is a request for book ?book 10. (return-due ?book - book) ;; Book ?book is due for return 11. (checked-out ?book - book) ;; Book ?book is checked out Action Description: Consider a librarian holding a book and standing near a shelf. The 'place-on-shelf' action involves placing the held book on top of another book on the shelf, given that the book on the shelf is accessible. This action results in the held book becoming accessible, the book on the shelf becoming inaccessible, and the librarian's hands becoming free. Action name: place-on-shelf Your answer:	Parameters: 1. ?x - book 2. ?y - book 3. ?cat - category Preconditions: <br>(and<br> (holding ?x)<br> (accessible ?y)<br> (belongs-to-category ?x ?cat)<br> (not (shelf-overflow ?cat))<br>)<br> Effects: <br>(and<br> (not (holding ?x))<br> (not (accessible ?y))<br> (accessible ?x)<br> (hands-free)<br> (on-shelf ?x ?y)<br> (shelf-empty ?cat)<br>)<br>	Parameters: 1. ?x - book 2. ?y - book 3. ?cat - category Preconditions: <br>(and<br> (accessible ?y)<br> (belongs-to-category ?x ?cat)<br> (not (shelf-overflow ?cat))<br>)<br> Effects: <br>(and<br> (not (shelf-empty ?cat))<br> (not (holding ?x))<br> (not (accessible ?y))<br> (accessible ?x)<br> (hands-free)<br> (on-shelf ?x ?y)<br>)<br>
  Context: Translate the given natural language description into an action schema that includes the parameters, preconditions, and effects. Ensure that only the provided predicates are used to construct the preconditions and effects. Question: Here is the task. A natural language description of the domain Domain information: This domain is structured to allow organizing and managing books within a library setting. The actions and predicates support the movement of books between tables and shelves, ensuring that conditions like accessibility and the librarian's hands being free are met. Additionally, it includes managing book categories, shelf space, and check-out/return processes to reflect a more complex library system. A list of available predicates 1. (on-shelf ?x ?y - book) ;; ?x is on top of ?y on the shelf 2. (on-table ?x - book) ;; ?x is on the table 3. (accessible ?x - book) ;; ?x is accessible (not covered) 4. (hands-free) ;; The hands of the librarian are free 5. (holding ?x - book) ;; The librarian is holding ?x 6. (belongs-to-category ?x - book ?cat - category) ;; ?x belongs to the category ?cat 7. (shelf-empty ?cat - category) ;; The shelf for category ?cat is empty 8. (shelf-overflow ?cat - category) ;; The shelf for category ?cat is full 9. (book-request ?book - book) ;; There is a request for book ?book 10. (return-due ?book - book) ;; Book ?book is due for return 11. (checked-out ?book - book) ;; Book ?book is checked out Action Description: Put a book you're holding on top of another accessible book on the shelf. Action name: place-on-shelf Your answer:	Parameters: 1. ?x - book 2. ?y - book 3. ?cat - category Preconditions: <br>(and<br> (holding ?x)<br> (accessible ?y)<br> (belongs-to-category ?x ?cat)<br> (not (shelf-overflow ?cat))<br>)<br> Effects: <br>(and<br> (not (holding ?x))<br> (not (accessible ?y))<br> (accessible ?x)<br> (hands-free)<br> (on-shelf ?x ?y)<br> (shelf-empty ?cat)<br>)<br>	Parameters: 1. ?x - book 2. ?y - book 3. ?cat - category Preconditions: <br>(and<br> (holding ?x)<br> (accessible ?y)<br> (belongs-to-category ?x ?cat)<br> (not (shelf-overflow ?cat))<br>)<br> Effects: <br>(and<br> (not (hands-free))<br> (not (holding ?x))<br> (not (accessible ?y))<br> (accessible ?x)<br> (on-shelf ?x ?y)<br> (shelf-empty ?cat)<br>)<br>
  Context: Translate the given natural language description into an action schema that includes the parameters, preconditions, and effects. Ensure that only the provided predicates are used to construct the preconditions and effects. Question: Here is the task. A natural language description of the domain Domain information: This domain is structured to allow organizing and managing books within a library setting. The actions and predicates support the movement of books between tables and shelves, ensuring that conditions like accessibility and the librarian's hands being free are met. Additionally, it includes managing book categories, shelf space, and check-out/return processes to reflect a more complex library system. A list of available predicates 1. (on-shelf ?x ?y - book) ;; ?x is on top of ?y on the shelf 2. (on-table ?x - book) ;; ?x is on the table 3. (accessible ?x - book) ;; ?x is accessible (not covered) 4. (hands-free) ;; The hands of the librarian are free 5. (holding ?x - book) ;; The librarian is holding ?x 6. (belongs-to-category ?x - book ?cat - category) ;; ?x belongs to the category ?cat 7. (shelf-empty ?cat - category) ;; The shelf for category ?cat is empty 8. (shelf-overflow ?cat - category) ;; The shelf for category ?cat is full 9. (book-request ?book - book) ;; There is a request for book ?book 10. (return-due ?book - book) ;; Book ?book is due for return 11. (checked-out ?book - book) ;; Book ?book is checked out Action Description: Imagine a patron returning a borrowed book to the library. The 'return-book' action enables the librarian to process the return, updating the book's status and removing any return due date. This action is applicable when the librarian is holding the book that needs to be returned. Action name: return-book Your answer:	Parameters: 1. ?x - book Preconditions: <br>(and<br> (checked-out ?x)<br> (holding ?x)<br>)<br> Effects: <br>(and<br> (not (checked-out ?x))<br> (not (holding ?x))<br> (not (book-request ?x))<br> (not (return-due ?x))<br> (accessible ?x)<br> (hands-free)<br>)<br>	Parameters: 1. ?x - book Preconditions: <br>(and<br> (not (checked-out ?x))<br> (holding ?x)<br>)<br> Effects: <br>(and<br> (return-due ?x)<br> (checked-out ?x)<br> (not (holding ?x))<br> (not (book-request ?x))<br> (accessible ?x)<br> (hands-free)<br>)<br>

• Loss: GISTEmbedLoss with these parameters:

  {'guide': SentenceTransformer(
    (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: T5EncoderModel 
    (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): Dense({'in_features': 1024, 'out_features': 768, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
    (3): Normalize()
  ), 'temperature': 0.01}
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 256
• num_train_epochs: 40
• max_steps: 31251
• 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: 256
• per_device_eval_batch_size: 8
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_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: 1.0
• num_train_epochs: 40
• max_steps: 31251
• 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: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• 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
• dispatch_batches: None
• split_batches: 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
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss
0.0013	1	13.401	-
0.0128	10	12.9494	-
0.0256	20	11.0057	-
0.0384	30	6.5825	-
0.0512	40	3.0534	-
0.0639	50	2.38	3.4251
0.0767	60	1.7338	-
0.0895	70	1.4789	-
0.1023	80	1.3218	-
0.1151	90	1.042	-
0.1279	100	1.06	0.8531
0.1407	110	0.9958	-
0.1535	120	0.9441	-
0.1662	130	0.7988	-
0.1790	140	0.7177	-
0.1918	150	0.8212	0.4972
0.2046	160	0.6318	-
0.2174	170	0.507	-
0.2302	180	0.6205	-
0.2430	190	0.5646	-
0.2558	200	0.5621	0.2187
0.2685	210	0.5915	-
0.2813	220	0.4457	-
0.2941	230	0.481	-
0.3069	240	0.4547	-
0.3197	250	0.443	0.1757
0.3325	260	0.3636	-
0.3453	270	0.4799	-
0.3581	280	0.3901	-
0.3708	290	0.3003	-
0.3836	300	0.25	0.1038
0.3964	310	0.2678	-
0.4092	320	0.3444	-
0.4220	330	0.2106	-
0.4348	340	0.2709	-
0.4476	350	0.2826	0.3395
0.4604	360	0.2434	-
0.4731	370	0.2208	-
0.4859	380	0.2434	-
0.4987	390	0.2766	-
0.5115	400	0.2067	0.0890
0.5243	410	0.2169	-
0.5371	420	0.2233	-
0.5499	430	0.1617	-
0.5627	440	0.1772	-
0.5754	450	0.1642	0.1299
0.5882	460	0.1493	-
0.6010	470	0.1507	-
0.6138	480	0.145	-
0.6266	490	0.1137	-
0.6394	500	0.1797	0.1373
0.6522	510	0.142	-
0.6650	520	0.1299	-
0.6777	530	0.0861	-
0.6905	540	0.1347	-
0.7033	550	0.0868	0.3025
0.7161	560	0.2161	-
0.7289	570	0.1281	-
0.7417	580	0.1241	-
0.7545	590	0.0554	-
0.7673	600	0.1829	0.7387
0.7801	610	0.1516	-
0.7928	620	0.094	-
0.8056	630	0.0902	-
0.8184	640	0.1677	-
0.8312	650	0.0541	0.3037
0.8440	660	0.1283	-
0.8568	670	0.1334	-
0.8696	680	0.1791	-
0.8824	690	0.1431	-
0.8951	700	0.0935	0.4147
0.9079	710	0.04	-
0.9207	720	0.1699	-
0.9335	730	0.1293	-
0.9463	740	0.1027	-
0.9591	750	0.1299	0.0023
0.9719	760	0.088	-
0.9847	770	0.0886	-
0.9974	780	0.0636	-
1.0102	790	0.1167	-
1.0230	800	0.0653	0.1323
1.0358	810	0.1378	-
1.0486	820	0.0778	-
1.0614	830	0.1212	-
1.0742	840	0.0472	-
1.0870	850	0.0861	0.0161

Framework Versions

• Python: 3.10.14
• Sentence Transformers: 3.1.0.dev0
• Transformers: 4.42.3
• PyTorch: 2.3.1+cu121
• Accelerate: 0.32.1
• Datasets: 2.20.0
• Tokenizers: 0.19.1

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",
}

GISTEmbedLoss

@misc{solatorio2024gistembed,
    title={GISTEmbed: Guided In-sample Selection of Training Negatives for Text Embedding Fine-tuning},
    author={Aivin V. Solatorio},
    year={2024},
    eprint={2402.16829},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}