This is a sentence-transformers model finetuned from BAAI/bge-m3. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: BAAI/bge-m3
Maximum Sequence Length: 8192 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity
Supported Modality: Text

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'XLMRobertaModel'})
  (1): Pooling({'embedding_dimension': 1024, 'pooling_mode': 'cls', '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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("pameydorke/arcanum-quests-retriever")
# Run inference
sentences = [
    'steps to rescue prisoner',
    "Rescue Tollo Underhill. If the party finds and rescues Tollo randomly (without being assigned by K'an Hua) then the rewards are still the same The Pits has some interesting loot, so exploring it in its entirety may be a consideration",
    "Clean Up the Boil. Damian Maug's Path. The Closed Sewers (optional) If the player does not want to conduct a direct frontal assault on Damian Maug, then they may try a stealth approach The Bentley (see below) and the Port of Tarant have access to the closed sewers of Tarant Point 1 is the Port entrance and Point 3 is the access to the Bentley This route can be used to avoid the bulk of Maug's defenses The Bentley The player can find Maug (point 1) in his house on the west side of The Boil If the party goes for a frontal assault then they will have to slowly work their way through a maze of rooms with about a dozen powerful and well armed gang members scattered a few at a time Defeating them and getting to the final room will let the party finish off Maug and his bodyguard",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.3610, 0.2773],
#         [0.3610, 1.0000, 0.1856],
#         [0.2773, 0.1856, 1.0000]])

Evaluation

Metrics

Information Retrieval

Dataset: arcanum-test
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7456
cosine_accuracy@3	0.9079
cosine_accuracy@5	0.9737
cosine_accuracy@10	0.9825
cosine_precision@1	0.7456
cosine_precision@3	0.3026
cosine_precision@5	0.1947
cosine_precision@10	0.0982
cosine_recall@1	0.7456
cosine_recall@3	0.9079
cosine_recall@5	0.9737
cosine_recall@10	0.9825
cosine_ndcg@10	0.8729
cosine_mrr@10	0.8365
cosine_map@100	0.8378

Training Details

Training Dataset

Unnamed Dataset

Size: 912 training samples
Columns: anchor, positive, and negative
Approximate statistics based on the first 912 samples:
anchor positive negative
type string string string
details
min: 5 tokens
mean: 9.62 tokens
max: 19 tokens
min: 15 tokens
mean: 180.92 tokens
max: 1044 tokens
min: 15 tokens
mean: 237.09 tokens
max: 1044 tokens

Samples:

anchor	positive	negative
`kill king caladon guide`	Kill the King of Caladon. After talking with Mr Willoughsby, a man, Heinrich Jenks appears in the room outside. He wants you to assassinate the King Of Caladon, he tells you to meet Vernon in the Kitchen for more details. When the job is done he will meet you again in Grant’s Tavern and give you 20,000 gold.You can tell Mr Willoughsby and he will send Lorham, his bodyguard to deal with him (this botches the quest).Travelling to Caladon, you need to talk with Renard first and then you will find Vernon who gives you a key.You can also get into the palace by hiding in a crate on the Caladon waterfront.Upstairs you will find a guard patrolling, he will try and kill you on sight (getting past him depends on your style prowling and backstab work well here). Make sure you loot the rooms as there is a lot of jewellery. Go through the ventilation system and explore.It is worthwhile making your way to the treasure vault. The door can be difficult to pick (or break down without damaging weapons)....	Find Dudley Crosston. This quest is found in the "rumor" tab of the player's logbook, rather than in the "quest" tab If the player has expert training in Bow when he talks to Kietzal in Black Root, then this quest can be done concurrently with the Bow Mastery Quest Otherwise the Mastery Quest cannot be started until reaching Caladon, and talking to Kietzal a second time And since access to Caladon is only available later in the game, it can be advantageous to get expert training in Bow BEFORE taking the Dudley quest if Bow Mastery is desired, or simply if the player wishes to get the Bow of Ecclesiastes early
`gambliing quest walkthrough`	Skill mastery quest. Gambling mastery. Gamble with Gurin Rockharrow. Women are refused entry to the Gentlemen's Club, but they still have a few avenues: Pick Pocket the key from the doorman Use Unlocking Cantrip to unbar the door Find out the name of the Club Owner from the doorman, getting his address from the Hall of Records, and then paying him a visit Once you meet with Wendell Wellington the options branch further: Pay a bribe and sleep with him for a special invitation Pick Pocket the invite Kill him and his bodyguard and loot his corpse for the invite With 8 points (2 ranks) of melee, you can threaten him for it With 4 points (1 rank) of persuasion and a Reaction Modifier of 81 you can sweet talk him for it Using this invitation will allow a lady to come and go as she pleases	Skill mastery quest. Persuasion mastery. Negotiations with Caladon. Willoughsby will comment about Orator of Ashbury reputation from the quest Monument Planning. Willoughsby will remain in Caladon for the rest of the game. The City Hall dialogue will be inaccessible, making it impossible for Sebastian to appear in The Boil. If the copy of the Agreement is somehow lost, the living one can still report its content to Willoughsby. After the negotiations are completed, independently of the result, King Farad will be "Leaving immediately for Tarant" and disappear; furthermore, the quest will be botched if the King is killed. Making it mutually exclusive with Assassinate King Farad of Caladon. There is a cut outcome left in the game code. If Tarant's counter was lower or equal to 1, currently an impossible result, Willoughsby would have been furious and wouldn't have given a reward,
`where is the key for big warehouse`	`Rid Mr. Plough's Warehouses of Rats. Talk to Simon Plough at the warehouse entrance and ask about his problem. Visit the smaller warehouse to the left first and kill the rats inside. You will also find a key in one of the barrels. Use the key to enter the bigger warehouse and kill everything inside. Return to Simon to finish the quest.`	`Caladon Thieves Underground. Adam Maxwell. Steal the book "Divinations and Magicks For The Adept Wizard" The book is found in a locked case in the library of the Misk Residence. The front room is guarded by the butler at night, and also by Mrs. Misk by day. Steal the Necklace of Queen Kna Sea The necklace is found in a locked safe in the Caladon Museum. The museum has no other visitors, but it is guarded by several mechanical arachnids.`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false, 5 "directions": [ 6 "query_to_doc" 7 ], 8 "partition_mode": "joint", 9 "hardness_mode": null, 10 "hardness_strength": 0.0 11}

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: 1e-05
weight_decay: 0.01
num_train_epochs: 1
lr_scheduler_type: cosine
warmup_steps: 0.1

All Hyperparameters

Click to expand

do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
gradient_accumulation_steps: 4
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 1e-05
weight_decay: 0.01
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: cosine
lr_scheduler_kwargs: None
warmup_ratio: None
warmup_steps: 0.1
log_level: passive
log_level_replica: warning
log_on_each_node: True
logging_nan_inf_filter: True
enable_jit_checkpoint: False
save_on_each_node: False
save_only_model: False
restore_callback_states_from_checkpoint: False
use_cpu: False
seed: 42
data_seed: None
bf16: False
fp16: False
bf16_full_eval: False
fp16_full_eval: False
tf32: None
local_rank: -1
ddp_backend: None
debug: []
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_prefetch_factor: None
disable_tqdm: False
remove_unused_columns: True
label_names: None
load_best_model_at_end: False
ignore_data_skip: False
fsdp: []
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': 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
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
optim_args: None
group_by_length: False
length_column_name: length
project: huggingface
trackio_space_id: trackio
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
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_for_metrics: []
eval_do_concat_batches: True
auto_find_batch_size: False
full_determinism: False
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
include_num_input_tokens_seen: no
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: True
use_cache: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	arcanum-test_cosine_ndcg@10
0.7018	10	0.8921	0.8700
1.0	15	-	0.8729

Training Time

Training: 4.2 minutes
Evaluation: 2.4 seconds
Total: 4.3 minutes

Framework Versions

Python: 3.12.13
Sentence Transformers: 5.4.0
Transformers: 5.0.0
PyTorch: 2.10.0+cu128
Accelerate: 1.13.0
Datasets: 4.0.0
Tokenizers: 0.22.2

Citation

BibTeX

Sentence Transformers

bibtex

@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

bibtex

@misc{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}

	anchor	positive	negative
type	string	string	string
details	min: 5 tokens mean: 9.62 tokens max: 19 tokens	min: 15 tokens mean: 180.92 tokens max: 1044 tokens	min: 15 tokens mean: 237.09 tokens max: 1044 tokens