Navya-1
Collection
India-first financial LMs from scratch (100M-1.31B), base + SFT chat. navya-1c-sft powers navyam.ai. • 6 items • Updated
navya-1c-sft
Navya is an India-first financial foundation model family, trained from scratch on an India-first corpus (finance + English + Hindi/Hinglish and other Indian languages), custom 64k tokenizer.
| Parameters | 1.31B |
| Training tokens | ~103B |
| Trained | Sep 2026 |
| Type | chat (SFT) |
| Author | Navyam AI (Bachatt) |
| License | Apache-2.0 |
| Code | https://github.com/bachatt-app/navyam-gpt |
1.31B base (~103B tokens + cooldown) then SFT on the India-finance corpus (val loss ~1.09). The model currently live on navyam.ai.
Intended use: India personal-finance Q&A. Research model — not investment advice.
Chat template
Role tokens <|reserved_0..4|> (system/user/assistant/tool/end_turn); generation stops on <|reserved_4|> and <|eos|>.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("navyam-ai/navya-1c-sft")
model = AutoModelForCausalLM.from_pretrained("navyam-ai/navya-1c-sft")
- Downloads last month
- 164