Gonyai-teo2 / README.md
omdeep22's picture
Update README.md
6d54d26 verified
metadata
language:
  - kok
tags:
  - konkani
  - goa
  - causal-lm
  - text-generation
license: mit

Gonyai-TEO2 — Konkani Language Model

Gonyai (गोण्याय) is a Konkani AI assistant trained on Goan culture, history, and the Konkani language (Goan dialect, Devanagari script).

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id  = "omdeep22/Gonyai-teo2"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model     = AutoModelForCausalLM.from_pretrained(
                model_id, trust_remote_code=True).to("cuda")

response = model.chat(tokenizer, "गोंयच्या निसर्गाविशीं एक कविता बरय.")
print(response)

Multi-turn Conversation

messages = [
    {"role": "user",      "content": "गोंयचें फेमस जेवण कितें?"},
    {"role": "assistant", "content": "शित-कडी, मासळें कालवण, बेबिंका आनी सोलकडी."},
    {"role": "user",      "content": "बेबिंका कशी करतात?"},
]
response = model.chat(tokenizer, messages)
print(response)

Reading Comprehension / RAG

passage  = "गोंयांत काजूची लागवड खूब जाता. काजूपासून फेणी तयार करतात."
question = "काजूपासून कितें तयार करतात?"
prompt   = f"हो उतारो वाच:\n\n{passage}\n\nप्रस्न: {question}"
response = model.chat(tokenizer, prompt)
print(response)  # → "फेणी"

Parameters

Architecture KonkanGPT (RoPE + RMSNorm + SwiGLU)
Parameters ~251M
Layers 24 transformer blocks
Context 4096 tokens
Vocabulary 32,000 (custom Konkani BPE)
Language Konkani, Goan dialect, Devanagari