navyam-ai/ navya-1:latest

17 4 weeks ago

India-first financial foundation models, trained from scratch — not fine-tuned from an existing base. The Navya-1 family spans a ~100M research model up to the 1.31B flagship that powers navyam.ai

1b 
ollama run navyam-ai/navya-1

Details

1 month ago

6a4b7324fa39 · 2.6GB

model
llama
·
1.31B
·
F16
template
{{ if .System }}<|reserved_0|>{{ .System }}<|reserved_4|>{{ end }}{{ range .Messages }}{{ if eq .Rol
system
You are Navya, an India-first personal-finance assistant.
params
{ "stop": [ "<|reserved_4|>", "<|eos|>" ], "temperature": 0.2 }

Readme

Navya-1

India-first financial foundation models, trained from scratch. Navya-1 is a family of small language models for Indian personal finance — built end-to-end (custom 64k tokenizer, from-scratch pretraining, SFT) on an India-first corpus covering English, Hindi/Hinglish and Indian languages.

Pull the flagship:

ollama run navyam-ai/navya-1

…or any specific member by tag:

ollama run navyam-ai/navya-1:1c-sft   # flagship chat model (live on navyam.ai)
ollama run navyam-ai/navya-1:1c        # 1.31B base
ollama run navyam-ai/navya-1:1b-sft    # 337M chat

Tags

Tag Params Type Training Notes
1a ~100M base 2026 earliest 1-series
1b 337M base 2026
1b-sft 337M chat (SFT) 2026 prior live model
1c 1.31B base ~103B tokens + WSD cooldown, Aug–Sep 2026 trained on 8×H100
1c-sft 1.31B chat (SFT) 103B + cooldown, then SFT flagship — live on navyam.ai
latest 1.31B chat → same as 1c-sft default pull

Intended use

India personal-finance Q&A — a research model, not investment advice. Small models; verify anything consequential.

Details