README.md

---
license: apache-2.0
base_model: google/gemma-4-E2B-it
library_name: llama.cpp
pipeline_tag: text-generation
tags:
  - gguf
  - lora
  - adapter
  - gemma4
  - text-generation
  - coding
---

# Gemopus v1 checkpoint-400 GGUF LoRA

Gemopus v1 checkpoint-400 is a GGUF LoRA adapter trained by Disc View / Disc Labs.

This repo contains the checkpoint-400 LoRA adapter only:

```text
Gemopus-v1-e2b-lora.gguf
```

It is not a standalone full model. To run it, load it together with the base model GGUF.

If `Gemopus-v1-e2b-Q4_K_M-merged.gguf` is present in this repository, that file is the standalone merged GGUF and can be loaded directly in LM Studio.

## Base Model

Base architecture/model family:

```text
google/gemma-4-E2B-it
```

Recommended local GGUF base file:

```text
gemma-4-E2B-it-Q4_K_M.gguf
```

## Identity

Gemopus v1 is a Gemma-based model fine-tuned toward a helpful, structured, Claude-like assistant style. It is not Claude, not Anthropic, and not a standalone model without the base model.

Suggested system prompt:

```text
You are Gemopus v1 by Disc View / Disc Labs, a Gemma-based model fine-tuned toward a Claude-like helpful, structured assistant style. Be helpful, accurate, clear, and strong at coding.
```

## llama.cpp Usage

Example:

```bash
llama-server \
  --model gemma-4-E2B-it-Q4_K_M.gguf \
  --lora Gemopus-v1-e2b-lora.gguf \
  --alias gemopus-v1-e2b-lora \
  --ctx-size 4096
```

Then call the OpenAI-compatible endpoint:

```text
http://127.0.0.1:8080/v1/chat/completions
```

## Important

If the model answers that it is stock Gemma or trained by Google, then the LoRA adapter is not attached. Make sure both files are loaded together:

```text
base GGUF + Gemopus-v1-e2b-lora.gguf
```

## Training Notes

This is a LoRA fine-tune checkpoint exported to GGUF adapter format. It was trained as Gemopus v1 checkpoint-400 with a focus on clearer instruction following, coding help, and structured answers.