Edit model card

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

Model Card for akṣara

akṣara, is a Micro Language Model ((µ-LM)) and is a fine-tuned version of the Mistral-7B-Instruct-v0.2 model. The model is fine-tuned using proprietary and open-source agricultural data. The data is specific to five countries in the global south: India, Bangladesh, Sri Lanka, Pakistan and Nepal, and covers nine of the major crops of the region like paddy/rice, wheat, maize, barley, sorghum, cotton, sugarcane, millets, soybean. It was compressed by using the QLoRA technique that compresses the model from 16-bit to 4-bit precision leading to a 60% reduction in the model size! It gives about 40% more relevant ROUGE score than GPT-4 Turbo on randomly selected test datasets.

The knowledge domain of the model is specific to the agricultural best practices, including climate-smart agricultural practices (CSA) and regenerative agricultural practices (RA) for the above-mentioned focus countries and crops. More geographies and crops will be added later. The model is trained on a database containing information from seed sowing to harvesting, covering every phenological stage of the crop growth cycle and different aspects like crop health management, soil management, disease control, and others. The end-to-end pipeline incorporates various aspects of Responsible AI (RAI), like considering local features and preventing harmful content or misinformation.

The following hyperparameters were used during training:

  • learning_rate: 2e-4
  • train_batch_size: 4
  • eval_batch_size: 4
  • optimizer: paged_adam_32bit
  • lr_scheduler_type: cosine
  • num_epochs: 1
  • lora_r: 32
  • weight_decay: 0.001

For full details of this model please read our release blog post here. The technical details for the background, fine-tuning and reproducibility will be shared as a pre-print on arxiv very soon. The model can be inferenced on our Huggingface Space aksara spaces For framework versions, please refer to the requirements.txt file.

Limitations

The knowledge of akṣara is limited to the specific region and crops. We are looking forward to engage with the community on ways to make it better on the model, data, pipeline and "Responsible AI".

Disclaimer

Beta Test version #1.0 - aksara is your agricultural AI advisor. Expect inaccuracies. We’re in active development stage to constantly learn & improve.

The Akshara Team

Downloads last month
142
Safetensors
Model size
7.24B params
Tensor type
BF16
·

Finetuned from

Spaces using cropinailab/aksara_v1 2