Sarvam-Fellow Assignment

1. baseline-inference

• Contains code for inference using meta-llama/Llama-3.2-11B-Vision.
• Works on the given dataset.

2. finetuning

• Details for fine-tuning meta-llama/Llama-3.2-11B-Vision.

3. dataset-ground-truth

• Code for generating the ground truth CSV file.
• Includes code for ground truth generation.

Processed dataset pushed to Hugging Face

4. Text Organization

• Includes code to structure text extracted from images using Mistral API

Summary

Baseline Inference Metric

• Average Sequence Accuracy: 0.330054
• WER (Average): 2.532779
• CER (Average): 1.510348

FineTuned Inference Metric

• Average Sequence Accuracy: 0.448970
• WER (Average): 1.685408
• CER (Average): 1.428369