nixie-suggest-small-v1

nixie-suggest-small-v1

This is a sentence-transformers model: It maps sentences & paragraphs to a 384 dimensional dense vector space and can be used for tasks like clustering or semantic search.

This model is based on E5-small-v2 model, fine-tuned for typical suggester-like workloads:

• for a partial and noisy input of the query, it tries to minimize the cosine distance to the correct query
• 'mil' should be close to 'milk'
• model also prone to typical typos like letter drops/swaps/duplications. So 'mikl' is still close to 'milk'.
• the model is asymmetrical (as the original E5), so you need to prepend your prefixes with 'query: ' and full queries with 'passage: '

Usage (Sentence-Transformers)

Using this model becomes easy when you have sentence-transformers installed:

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer
sentences = ["query: mil", "passage: milk"]

model = SentenceTransformer('nixiesearch/nixie-suggest-small-v1')
embeddings = model.encode(sentences)
print(embeddings)

Training dataset

The training dataset was syntetically generated from the following corpora:

• top-100k most frequent English words, from Google N-Gram project: https://github.com/hackerb9/gwordlist
• top-1M 2-grams and 3-grams from MultiLex

We did the following permutations to the original 1/2/3-grams:

• letter swaps: milk-mikl
• letter drops: milk-ilk
• qwerty-aware replacements: milk-nilk
• duplications: milk-miilk

The original generation code is available on github: https://github.com/nixiesearch/autocomplete-playground

Training

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 220359 with parameters:

{'batch_size': 2048, 'sampler': 'torch.utils.data.dataloader._InfiniteConstantSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss:

sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss with parameters:

{'scale': 20.0, 'similarity_fct': 'cos_sim'}

Parameters of the fit()-Method:

{
    "epochs": 1,
    "evaluation_steps": 3000,
    "evaluator": "sentence_transformers.evaluation.RerankingEvaluator.RerankingEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": 220358,
    "warmup_steps": 1000,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
  (2): Normalize()
)

Citing & Authors

<!--- Describe where people can find more information -->