Self-service improvements - phase 2

[mahalakshme]

Phase 1:
Bug fixes

Phase 2:
Rules

• which model - decoder only or encoder-decoder model
• what size model
• how many GPUs required to fine tune -
• Is pre-training from scratch needed because we use specific libraries.
• fine tuned models for each rules, or a single fine-tuned model for all rules
  - may be categorise based on contexts, ie., one fine tuned model for form level rules and rest for form element rules, etc.,
  - or can parallelise on the same model

Tasks

Beta Give feedback

1. [AI in AvnI] Rule generation with RAG avni-server#772
   +0
2. [AI in AvnI] Rule generation with fine tuning avni-server#773
   +0

Phase 3:
easiness of flow