# Publish to Hugging Face

This directory is the clean Hugging Face release folder.

Final model file:

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

Recommended repository name:

```text
<your-hf-username-or-org>/gemma4-loghub-e2b-GGUF
```

## Upload with the HF CLI

Install the CLI:

```bash
python3 -m pip install -U huggingface_hub
```

Log in with a Hugging Face write token:

```bash
hf auth login
```

Create the model repository:

```bash
hf repos create <your-hf-username-or-org>/gemma4-loghub-e2b-GGUF --type model --exist-ok
```

Upload this folder:

```bash
cd /Users/balt/resources/llmlog/release/gemma4-loghub-e2b-GGUF
hf upload <your-hf-username-or-org>/gemma4-loghub-e2b-GGUF . . --type model
```

## First public test

After upload, test with llama.cpp:

```bash
hf download <your-hf-username-or-org>/gemma4-loghub-e2b-GGUF gemma4-loghub-e2b-loghub-Q4_K_M.gguf

llama-server \
  -m gemma4-loghub-e2b-loghub-Q4_K_M.gguf \
  -c 2048 \
  --reasoning off
```

For tools that do not support `--reasoning off`, use:

```bash
llama-server \
  -m gemma4-loghub-e2b-loghub-Q4_K_M.gguf \
  -c 2048 \
  --chat-template-kwargs '{"enable_thinking":false}'
```

## What not to publish

Do not publish these intermediate files as the main release:

- old adapter-only GGUF files;
- MLX adapter directories;
- unified multimodal GGUF files;
- symlink import folders;
- smoke-test datasets.