This is a sentence-transformers model finetuned from AI4free/JARVIS-tool-search-v1. It maps sentences & paragraphs to a 256-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: AI4free/JARVIS-tool-search-v1
Maximum Sequence Length: inf tokens
Output Dimensionality: 256 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): StaticEmbedding({})
)

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("sentence_transformers_model_id")
# Run inference
sentences = [
    "Find tool: If I roll 3 dice, what's the probability of getting a sum of 10? Additionally, if I invested $5000 and it grew to $7500 over 3 years, what was the CAGR?",
    'Tools: [{"name": "cagr", "description": "Calculates the Compound Annual Growth Rate (CAGR) of an investment.", "parameters": {"start_value": {"description": "The initial value of the investment.", "type": "float"}, "end_value": {"description": "The final value of the investment.", "type": "float"}, "years": {"description": "The number of years between the initial and final values.", "type": "int"}}}, {"name": "binomial_probability", "description": "Calculates the probability of getting exactly k successes in n independent trials,", "parameters": {"n": {"description": "The total number of trials.", "type": "int"}, "k": {"description": "The number of successes.", "type": "int"}, "p": {"description": "The probability of success in each trial.", "type": "float"}}}, {"name": "dice_roll_probability", "\nCalled: [{"name": "dice_roll_probability", "arguments": {"target_sum": 10, "num_dice": 3}}, {"name": "cagr", "arguments": {"start_value": 5000, "end_value": 7500, "years": 3}}]',
    "Review: <think>\nOkay, let's see the problem here. So we have a pyramid with N steps, and the bottom step (step N) has a permutation of numbers from 1 to 2N-1. The rule is that each block above is the median of the three blocks directly below it. And we need to find the number in the top block (step 1). \n\nHmm, the sample input for N=4 gives output 4, and for N=2 it's 2. Wait a second, those outputs are exactly equal to N. So maybe the answer is always N? But why?\n\nLet me think. Let's consider the process. The top block's value depends on the medians as we go down. But since every step's value is the me\n```python\nn = int(input())\na = list(map(int, input().split()))\nprint(n)\n```",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 256]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.7395, 0.0061],
#         [0.7395, 1.0000, 0.0494],
#         [0.0061, 0.0494, 1.0000]])

Training Details

Training Dataset

Unnamed Dataset

Size: 597,350 training samples
Columns: anchor and positive
Approximate statistics based on the first 100 samples:
anchor positive
type string string
modality text text
details
min: 44 characters
mean: 557.53 characters
max: 815 characters
min: 191 characters
mean: 1470.12 characters
max: 1909 characters

Samples:

anchor	positive
Solve: Consider a rooted tree. A rooted tree has one special vertex called the root. All edges are directed from the root. Vertex u is called a child of vertex v and vertex v is called a parent of vertex u if there exists a directed edge from v to u. A vertex is called a leaf if it doesn't have children and has a parent.[object Object][object Object]Let's call a rooted tree a spruce if its every non-leaf vertex has at least 3 leaf children. You are given a rooted tree, check whether it's a spruce.[object Object][object Object]The definition of a rooted tree can be found [object Object].[object Object]Input[object Object][object Object]The first line contains one integer n — the numb	`[object Object]`
Solve: There are N balls placed in a row. AtCoDeer the deer is painting each of these in one of the K colors of his paint cans. For aesthetic reasons, any two adjacent balls must be painted in different colors.[object Object][object Object]Find the number of the possible ways to paint the balls.[object Object][object Object]Constraints[object Object][object Object]* 1≦N≦1000[object Object]* 2≦K≦1000[object Object]* The correct answer is at most 2^{31}-1.[object Object]Input[object Object][object Object]The input is given from Standard Input in the following format:[object Object][object Object][object Object]N K[object Object][object Object][object Object]Output[object Object][object Object]Print the number of the possible ways to paint the balls.[object Object][object Object]Examples[object Object][object Object]Input[object Object][object Object]2 2[object Object][object Object][object Object]Output[object Object][object Object]2[object Object][object Object][object Object]Input[object Object][object Object]1 10[object Object][object Object][object Object]Output[object Object][object Object]10	`[object Object]`
Solve: Little Pandey is someone who is lazy, and when he's around his best friend GJ, he becomes super lazy. Pandey thinks that he is a Math-wizard, so he picks up a number S and asks GJ to throw him a challenge around that number. [object Object][object Object]GJ explains Little Pandey a property called nothingness and decides to ask him Q queries based on it.[object Object]In mathematical terms, Nothingness(A, B) is defined as the maximum M that (A%M==0 and B%M==0). (You can read about the Modulus Operator here.)[object Object]In the i-th query GJ says a number Ai and Little Pandey must find Nothingness(S, Ai).[object Object][object Object]But, Pandey is the laziest of all.[object Object]He wo	`[object Object]`

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}

Evaluation Dataset

Unnamed Dataset

Size: 2,000 evaluation samples
Columns: anchor and positive
Approximate statistics based on the first 100 samples:
anchor positive
type string string
modality text text
details
min: 45 characters
mean: 537.05 characters
max: 815 characters
min: 36 characters
mean: 1455.73 characters
max: 1909 characters

Samples:

anchor	positive
Solve: Shaass thinks a kitchen with all white floor tiles is so boring. His kitchen floor is made of n·m square tiles forming a n × m rectangle. Therefore he's decided to color some of the tiles in black so that the floor looks like a checkerboard, which is no two side-adjacent tiles should have the same color.[object Object][object Object]Shaass wants to use a painter robot to color the tiles. In the beginning the robot is standing in a border tile (xs, ys) facing a diagonal direction (i.e. upper-left, upper-right, down-left or down-right). As the robot walks in the kitchen he paints every tile he passes even if it's painted b	`[object Object]`
`Find tool: Fetch the dialog strings for users 'user14' and 'user15', who are contributors to an open-source project.`	`Tools: [{"name": "get_user_dialog_string", "description": "Fetches the dialog string for a given user from the Waifu API.", "parameters": {"user_id": {"description": "A unique identifier for the user.", "type": "str", "default": "sample_user_id"}}}][object Object]Called: [{"name": "get_user_dialog_string", "arguments": {"user_id": "user14"}}, {"name": "get_user_dialog_string", "arguments": {"user_id": "user15"}}]`
`Code (python): [object Object]`	`Review: [object Object]`

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

per_device_train_batch_size: 4096
num_train_epochs: 5
learning_rate: 0.04
lr_scheduler_type: cosine
warmup_steps: 0.05
disable_tqdm: True
per_device_eval_batch_size: 4096
dataloader_drop_last: True
dataloader_num_workers: 4
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

per_device_train_batch_size: 4096
num_train_epochs: 5
max_steps: -1
learning_rate: 0.04
lr_scheduler_type: cosine
lr_scheduler_kwargs: None
warmup_steps: 0.05
optim: adamw_torch_fused
optim_args: None
weight_decay: 0.0
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: False
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: True
project: huggingface
trackio_space_id: None
trackio_bucket_id: None
trackio_static_space_id: None
per_device_eval_batch_size: 4096
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: None
hub_strategy: every_save
hub_always_push: False
hub_revision: None
load_best_model_at_end: False
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: True
dataloader_num_workers: 4
dataloader_pin_memory: True
dataloader_persistent_workers: False
dataloader_prefetch_factor: None
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: []
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
warmup_ratio: None
local_rank: -1
prompts: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.0069	1	3.1332
0.0690	10	3.0859
0.1379	20	2.7984
0.2069	30	2.4399
0.2759	40	2.1068
0.3448	50	1.8330
0.4138	60	1.6418
0.4828	70	1.4892
0.5517	80	1.3397
0.6207	90	1.2159
0.6897	100	1.1323
0.7586	110	1.0537
0.8276	120	0.9845
0.8966	130	0.9245
0.9655	140	0.8589
0.9793	142	-
1.0552	150	0.7909
1.1241	160	0.7432
1.1931	170	0.7173
1.2621	180	0.6876
1.3310	190	0.6648
1.4	200	0.6410
1.4690	210	0.6102
1.5379	220	0.5917
1.6069	230	0.5745
1.6759	240	0.5561
1.7448	250	0.5461
1.8138	260	0.5416
1.8828	270	0.5205
1.9517	280	0.5045
1.9793	284	-
2.0414	290	0.4867
2.1103	300	0.4625
2.1793	310	0.4571
2.2483	320	0.4541
2.3172	330	0.4423
2.3862	340	0.4436
2.4552	350	0.4323
2.5241	360	0.4263
2.5931	370	0.4258
2.6621	380	0.4155
2.7310	390	0.4083
2.8	400	0.4131
2.8690	410	0.4059
2.9379	420	0.3996

Training Time

Training: 4.5 hours
Evaluation: 3.3 seconds
Total: 4.5 hours

Framework Versions

Python: 3.13.13
Sentence Transformers: 5.5.1
Transformers: 5.9.0
PyTorch: 2.12.0+cu130
Accelerate: 1.13.0
Datasets: 4.8.5
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
type	string	string
modality	text	text
details	min: 44 characters mean: 557.53 characters max: 815 characters	min: 191 characters mean: 1470.12 characters max: 1909 characters

AI4free/JARVIS-tool-search-v2

SentenceTransformer based on AI4free/JARVIS-tool-search-v1

Model Details

Model Description

Model Sources

Full Model Architecture

Usage

Direct Usage (Sentence Transformers)

Training Details

Training Dataset

Unnamed Dataset

Evaluation Dataset

Unnamed Dataset

Training Hyperparameters

Non-Default Hyperparameters

All Hyperparameters

Training Logs

Training Time

Framework Versions

Citation

BibTeX

Sentence Transformers

MultipleNegativesRankingLoss

AI4free/
JARVIS-tool-search-v2