Customize
Fine-tuning & datasets
Adapt TNSA models to your domain by training on your own examples. Datasets are uploaded as JSON Lines and managed from the web console.
Dataset format
Datasets are .jsonl files — one training example per line. Each line is a chat conversation with system, user and assistant turns.
json
{"messages": [{"role": "system", "content": "You are a SEBI policy advisor."}, {"role": "user", "content": "What is the filing window for mutual funds?"}, {"role": "assistant", "content": "The window is 30 days..."}]}
{"messages": [{"role": "system", "content": "You are a SEBI policy advisor."}, {"role": "user", "content": "When are quarterly disclosures due?"}, {"role": "assistant", "content": "Within 45 days of quarter end..."}]}One JSON object per line
Each line must be a complete, valid JSON object. Do not wrap the file in an array or add trailing commas.
Manage in the console
Datasets and training jobs are managed from the web console:
- Fine-tuning portal — create and monitor tuning jobs.
- Dataset manager — upload and organize
.jsonlfiles. - RLHF evaluations — preference and reward-model workflows.
Dataset tips
- Keep the system prompt consistent across examples for a given behavior.
- Include diverse, representative examples covering the cases you care about.
- Prefer quality and consistency over sheer volume.