---
title: Apochat Gemma 4 E2B Trainer
emoji: 🏋️
colorFrom: purple
colorTo: pink
sdk: gradio
sdk_version: 4.0.0
app_file: app.py
pinned: false
---

# Apochat Gemma 4 E2B Trainer

Fine-tune `google/gemma-4-E2B-it` on the Apochat chat dataset (`apoapps/apochat-gemma4-e2b-chat-v1`)
using PEFT/LoRA.

**You must upgrade the Space hardware to a GPU before starting training.** Free `cpu-basic`
will not be able to train a 2B model in a reasonable time.

Recommended hardware:
- `t4-small` or `l4x1` for QLoRA.
- `a10g-small` or larger for faster training.

When training finishes, the adapter is pushed to `apoapps/apochat-gemma4-e2b-apochat-tuned-v2`.
Use the LiteRT export script to convert the trained adapter + base model to `.litertlm`.