Edit model card

Model Card for INDUS-Retriever-small

INDUS-Retriever-small (previously nasa-smd-ibm-st.38m) is a Bi-encoder sentence transformer model, that is fine-tuned from distilled version of nasa-smd-ibm-v0.1 encoder model. it is a smaller version of nasa-smd-ibm-st with better performance, using fewer parameters (shown below). It's trained with 362 million examples along with a domain-specific dataset of 2.6 million examples from documents curated by NASA Science Mission Directorate (SMD). With this model, we aim to enhance natural language technologies like information retrieval and intelligent search as it applies to SMD NLP applications.

A bigger model is also available here: https://huggingface.co/nasa-impact/nasa-smd-ibm-st-v2

Model Details

  • Base Encoder Model: INDUS(https://huggingface.co/nasa-impact/nasa-smd-ibm-v0.1)
  • Tokenizer: Custom
  • Parameters: 38M
  • Training Strategy: Sentence Pairs, and score indicating relevancy. The model encodes the two sentence pairs independently and cosine similarity is calculated. the similarity is optimized using the relevance score.

Training Data

image/png

Figure: dataset sources for sentence transformers (362M in total)

Additionally, 2.6M abstract + title pairs collected from NASA SMD documents.

Training Procedure

  • Framework: PyTorch 1.9.1
  • sentence-transformers version: 4.30.2
  • Strategy: Sentence Pairs

Evaluation

Following models are evaluated:

  1. All-MiniLM-l6-v2 [sentence-transformers/all-MiniLM-L6-v2]
  2. BGE-base [BAAI/bge-base-en-v1.5]
  3. RoBERTa-base [roberta-base]
  4. nasa-smd-ibm-rtvr_v0.1 [nasa-impact/nasa-smd-ibm-st]

image/png

Figure: BEIR and NASA-IR Evaluation Metrics

Uses

  • Information Retreival
  • Sentence Similarity Search

For NASA SMD related, scientific usecases.

Usage


from sentence_transformers import SentenceTransformer, Util

model = SentenceTransformer("nasa-impact/nasa-smd-ibm-st.38m")

input_queries = [
'query: how much protein should a female eat', 'query: summit define']
input_passages = [
"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day.
But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments."]
query_embeddings = model.encode(input_queries)
passage_embeddings = model.encode(input_passages)
print(util.cos_sim(query_embeddings, passage_embeddings))

Note

This Sentence Transformer Model is released in support of the training and evaluation of the encoder language model "Indus".

Accompanying paper can be found here: https://arxiv.org/abs/2405.10725

Citation

If you find this work useful, please cite using the following bibtex citation:

@misc {nasa-impact_2024,
    author       = { {NASA-IMPACT} },
    title        = { nasa-ibm-st.38m (Revision 9c1989c) },
    year         = 2024,
    url          = { https://huggingface.co/nasa-impact/nasa-ibm-st.38m },
    doi          = { 10.57967/hf/1875 },
    publisher    = { Hugging Face }
}

Attribution

IBM Research

  • Aashka Trivedi
  • Masayasu Muraoka
  • Bishwaranjan Bhattacharjee
  • Takuma Udagawa

NASA SMD

  • Muthukumaran Ramasubramanian
  • Iksha Gurung
  • Rahul Ramachandran
  • Manil Maskey
  • Kaylin Bugbee
  • Mike Little
  • Elizabeth Fancher
  • Lauren Sanders
  • Sylvain Costes
  • Sergi Blanco-Cuaresma
  • Kelly Lockhart
  • Thomas Allen
  • Felix Grazes
  • Megan Ansdell
  • Alberto Accomazzi
  • Sanaz Vahidinia
  • Ryan McGranaghan
  • Armin Mehrabian
  • Tsendgar Lee

Disclaimer

This sentence-transformer model is currently in an experimental phase. We are working to improve the model's capabilities and performance, and as we progress, we invite the community to engage with this model, provide feedback, and contribute to its evolution.

Downloads last month
2,181
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.