KHROTU/titlegemma

KHROTU/ titlegemma

13 Downloads Updated 2 weeks ago

A fine-tuned Gemma 3 270M instruct model specialized in generating short, descriptive titles from the first message of a conversation.

ollama run KHROTU/titlegemma

curl http://localhost:11434/api/chat \
  -d '{
    "model": "KHROTU/titlegemma",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='KHROTU/titlegemma',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'KHROTU/titlegemma',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

1 model

Size / Usage

Context

Input

titlegemma:latest

551MB · 32K context window · Text · 2 weeks ago

titlegemma:latest

551MB

32K

Text

Readme

A fine-tuned Gemma 3 270M instruct model specialized in generating short, descriptive titles from the first message of a conversation, built for Arc.

What it does

Given a user’s opening message, titlegemma generates a concise 3–8 word title in Title Case with no emojis, no quotes, no markdown.

Base model

google/gemma-3-270m-it

Training

Dataset: 48,513 chat first-message -> title pairs derived from lmsys/lmsys-chat-1m (turn=1 conversations only), with titles generated by inclusionai/ling-2.6-flash via OpenRouter
Method: LoRA fine-tuning (r=16, alpha=32) with completion-only loss masking, 2 epochs, effective batch size 32, max sequence length 256, BF16
Hardware: 2× NVIDIA T4

Usage

<start_of_turn>user
Generate a title for a conversation that starts with:

how can identity protection services help protect me against identity theft<end_of_turn>
<start_of_turn>model
Identity Protection Services Benefits<end_of_turn>

License

Same as base Gemma model. See https://ai.google.dev/gemma/terms