[Experiments] Various experiments related to MindGPT

[dudeperf3ct]

In this PR, we add 4 notebooks each focusing on a different experiment. Each notebook is a fully reproducible experiment environment.

• big_models_experiment.ipynb: This notebook compares responses for NHS and Mind across various LLM models. Starting with flan-t5-base and keeping the rest of the variables constant, 5 additional LLM models were tested flan-t5-large, flan-t5-xl, fastchat-t5, openllam-3b and redpajama-3b-instruct.
• Data Chunking: There are two parts of this experiments captured in two separate notebook
  • data_chunking_experiments.ipynb: This notebook handpick small subset of dataset relevant for the question and experiments with different sizes of chunk_size and chunk_overlap.
  • data_chunking_experiments_third_version_openllm.ipynb: This notebook takes similar approach above where instead of handpicking datapoints entire dataset (the latest version) is used. Two additional improvements have been added i.e. flan-t5-xl LLM and advanced template is used.
• prompt_engineering_experiment.ipynb: This notebook experiments with 5 different prompts for 3 LLMs ( flan-t5-base , flan-t5-large and flan-t5-xl).