---
license: apache-2.0
library_name: gguf
pipeline_tag: text-generation
tags:
  - gguf
  - gemma4
  - gemma
  - log-analysis
  - log-parsing
  - troubleshooting
  - sysadmin
  - llama.cpp
  - ollama
language:
  - it
  - en
base_model:
  - unsloth/gemma-4-E2B-it-GGUF
---

# gemma4-loghub-e2b-GGUF

Gemma 4 E2B fine-tuned for infrastructure, Linux, Windows, networking, cloud and application log analysis.

## Files

```text
gemma4-loghub-e2b-loghub-Q4_K_M.gguf
```

This is the recommended release artifact:

- Base family: Gemma 4 E2B
- Format: GGUF v3
- Quantization: Q4_K_M
- Mode: text-only language model
- Tensor count: 601
- Size: ~3.2 GB
- Chat template: embedded Gemma 4 template
- SHA256: `e70b4d0009db2ec9fa4a57782d57264084572bb34bf9a4d41b76eb9559a07f73`

## Why Text-Only

Gemma 4 E2B is a multimodal model with language, audio and vision tensors. For log analysis we only need text inference. Some `llama-server` based tools fail when given a unified Gemma4 GGUF that contains the extra audio/vision tensors. This release keeps the fine-tuned language tensors and removes audio, vision and multimodal projector tensors.

## llama.cpp

Use a recent `llama.cpp` build with Gemma4 support.

```bash
llama-server \
  -m gemma4-loghub-e2b-loghub-Q4_K_M.gguf \
  -c 4096 \
  --reasoning off
```

If your `llama-server` does not support `--reasoning off`, use:

```bash
llama-server \
  -m gemma4-loghub-e2b-loghub-Q4_K_M.gguf \
  -c 4096 \
  --chat-template-kwargs '{"enable_thinking":false}'
```

## Ollama

From this directory:

```bash
ollama create gemma4-loghub-e2b:q4 -f Modelfile
ollama run gemma4-loghub-e2b:q4
```

## Unsloth Studio

Import the real `.gguf` file directly:

```text
gemma4-loghub-e2b-loghub-Q4_K_M.gguf
```

Do not import a symlink, the old LoRA adapter GGUF, or the unified multimodal GGUF.

Start with context length `2048` or `4096` on laptops.