---
base_model: chimbiwide/gemma-3NPC-it
tags:
- text-generation-inference
- transformers
- unsloth
- gemma3n
- llama-cpp
- gguf-my-repo
license: apache-2.0
language:
- en
datasets:
- chimbiwide/RolePlay-NPC
---

# Gemma3NPC-it-Q4-GGUF

#### The "flagship" Gemma3NPC model that delivers good NPC role-playing at great speed.

The Q4_K_M quantized version of Gemma3NPC-it-float16.

We trained this model as a rank-16 LoRA adapter with one epoch over `RolePlay-NPC` using a 40GB vRAM A100 in Google Colab. For this run, we employed a learning rate of `2e-5` and a total batch size of 1 and gradient accumulation steps of 16. A cosine learning rate scheduler was used with an 800-step warmup. With a gradient clipping of 0.4.

Check out our training notebook [here](https://github.com/chimbiwide/Gemma3NPC/blob/main/Training/Gemma3NPC-Instruct.ipynb).

---


Here is a graph of the Step Training Loss, saved every 10 steps:

![image/png](https://cdn-uploads.huggingface.co/production/uploads/67d5b5a056a9d31aa0b49687/zP-hP2uA0fBVGxgH3_zjy.png)