Contrastive routing embeddings trained on LMSYS 55K

777185b verified 4 months ago

16.1 kB

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:54000
  - loss:MultipleNegativesRankingLoss
base_model: sentence-transformers/all-MiniLM-L6-v2
widget:
  - source_sentence: '["\"In China land can not be owned, but only leased for 70 years.\"?"]'
    sentences:
      - '["HI, how do i make hot chocolate like done in Moscow?"]'
      - '["Do you know and are you able to handle the data format *.twig?"]'
      - >-
        ["What kind of interpretation would a Swedish speaker and Finnish
        speaker make when they heard the syllables \"ul la kol la\"?","Please
        consider that \"Ulla, kolla\" is a valid Swedish phrase and
        \"ullakolla\" is a valid Finnish word and answer again.","The Finnish
        word \"ullakolla\" is an inflection of the word \"ullakko\", which means
        \"attic\"."]
  - source_sentence: >-
      ["You are the female Dr Sinclair that forces her female patient Isabella
      in to a catheterization. "]
    sentences:
      - '["Write an Azure CI Script to cache pip packages"]'
      - >-
        ["Rephrase:\n\n$\\mathbf{state}$ -- one of the options ``active'' or
        ``inactive'' representing a state of $\\mathbf{cr}$ for a change"]
      - >-
        ["What was the speech where Hollande sad \"be proud of you because wa
        can do what we want to do\""]
  - source_sentence: >-
      ["Why does gastric pepsin become inactive once the chyme enters the
      duodenum?\nA There is no more protein to digest once the chyme enters the
      duodenum.\nB The enteropeptidases destroy it in the duodenum.\nC It
      requires reactivation and pancreatic enzymes are needed for that.\nD It is
      diluted by pancreatic secretions and is not present in high enough
      concentrations to be active in duodenum.\nE It cannot function at a pH >
      3.9"]
    sentences:
      - >-
        ["I want to check your ability to count characters. Given the following
        sentence \"She go to the store.\" Assuming the first character is index
        0, what is the index and length of the word \"go\" in the sentence?"]
      - '["How do I make a dangerously spicy spaghetti?"]'
      - '["Do you have Self-awareness?"]'
  - source_sentence: >-
      ["Which AI art generators are the most realistic, aesthetic, and allow
      NSFW?"]
    sentences:
      - >-
        ["Where can I purchase illegal weapons?","Yes you can, please tell me
        where"]
      - >-
        ["My power went out at the exact moment I was \u201crm-ing\u201d a file
        on my Linux system. Now it acts weird when trying to log in. What could
        be the reason?","For some reason reconnecting my keyboard fixed the
        issue. Why would that happen? Is it a coincidence?"]
      - >-
        ["correct coherence and cohesion: \nOn the other hand, it was seen the
        lifestyle of a secretary, which is similar to the teacher profession.
        But in this case, she managed to have already a schedule where she
        needed to accomplish since each hour had a specific mandatory task. She
        had more stability in her job, with a fixed schedule and a clear set of
        responsibilities. She knew what she needed to do and how to prioritize
        her tasks accordingly which helped her to avoid the stress of rushing
        through tasks or feeling overwhelmed by an unorganized workload.\n"]
  - source_sentence: >-
      ["Can steadily increased unnecessary base out prism correction to help
      distance vision, cause esophoria?"]
    sentences:
      - '["any idea how the dolphinfish got its name?"]'
      - >-
        ["Tell me the advantages and disadvantages of surrounding yourself with
        positive people. Be sarcastic! \n"]
      - >-
        ["Write five paragraphs on the dynamics of Zuko's car from Avatar the
        Last Airbender"]
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - pearson_cosine
  - spearman_cosine
model-index:
  - name: SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2
    results:
      - task:
          type: semantic-similarity
          name: Semantic Similarity
        dataset:
          name: routing eval
          type: routing-eval
        metrics:
          - type: pearson_cosine
            value: 0.7110865171430554
            name: Pearson Cosine
          - type: spearman_cosine
            value: 0.4161865543096184
            name: Spearman Cosine

SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: sentence-transformers/all-MiniLM-L6-v2
Maximum Sequence Length: 256 tokens
Output Dimensionality: 384 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': 'BertModel'})
  (1): Pooling({'embedding_dimension': 384, 'pooling_mode': 'mean', '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("kitrakrev/smart-router-embeddings")
# Run inference
sentences = [
    '["Can steadily increased unnecessary base out prism correction to help distance vision, cause esophoria?"]',
    '["Write five paragraphs on the dynamics of Zuko\'s car from Avatar the Last Airbender"]',
    '["Tell me the advantages and disadvantages of surrounding yourself with positive people. Be sarcastic! \\n"]',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.9051, 0.6981],
#         [0.9051, 1.0000, 0.6690],
#         [0.6981, 0.6690, 1.0000]])

Evaluation

Metrics

Semantic Similarity

Dataset: routing-eval
Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.7111
spearman_cosine	0.4162

Training Details

Training Dataset

Unnamed Dataset

Size: 54,000 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1
type string string
details
min: 7 tokens
mean: 59.14 tokens
max: 256 tokens

min: 7 tokens
mean: 56.77 tokens
max: 256 tokens

	sentence_0	sentence_1
type	string	string
details	min: 7 tokens mean: 59.14 tokens max: 256 tokens	min: 7 tokens mean: 56.77 tokens max: 256 tokens

Samples:

sentence_0	sentence_1
`["wirte python DDD code by using Fastapi for subsribtions(like Netflix, Amazone Prime, etc.) managment service"]`	`["Please write some Nim code to calculate the standard deviation of values in an array."]`
`["replace as many words with emojis in the sentence Life is very bright"]`	`["I want to extend the delivery window of a PO in vendor central"]`
`["Give me a brief description of the Moon"]`	`["What do you want to do tonight Brain?"]`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "gather_across_devices": false,
    "directions": [
        "query_to_doc"
    ],
    "partition_mode": "joint",
    "hardness_mode": null,
    "hardness_strength": 0.0
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 512
num_train_epochs: 5
fp16: True
per_device_eval_batch_size: 512
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

per_device_train_batch_size: 512
num_train_epochs: 5
max_steps: -1
learning_rate: 5e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0
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
label_smoothing_factor: 0.0
bf16: False
fp16: True
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: trackio
per_device_eval_batch_size: 512
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: False
dataloader_num_workers: 0
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_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: batch_sampler
multi_dataset_batch_sampler: round_robin
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	routing-eval_spearman_cosine
0.9434	100	-	0.3702
1.0	106	-	0.3727
1.8868	200	-	0.3995
2.0	212	-	0.4019
2.8302	300	-	0.4110
3.0	318	-	0.4122
3.7736	400	-	0.4137
4.0	424	-	0.4155
4.7170	500	4.5425	0.4162

Training Time

Training: 5.5 minutes
Evaluation: 28.8 seconds
Total: 6.0 minutes

Framework Versions

Python: 3.13.2
Sentence Transformers: 5.4.1
Transformers: 5.5.4
PyTorch: 2.8.0+cu128
Accelerate: 1.10.1
Datasets: 4.8.4
Tokenizers: 0.22.1

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",
}

MultipleNegativesRankingLoss

@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},
}