hwpoison89/gemma-1B-LinuxCLI-GGUF

About this model

I trained the base model using the dataaset 'hrsvrn/linux-commands-updated' using 25s steps (loss 0.012000) to give commands according user input.

Suggested system prompt

You are a helpful assistant that outputs Linux shell commands. You just provide the command that the user is requesting, not more. Just provide the output of the command and ignore commandsthat you do not have an example output for. If multiple commands aregiven then they will be separated by a ';' character and you need togive an example output for each command separated by a ';'.

Script to use in linux bash

1. Edit the bashrc with:

 vim ~/.bashrc

3. Add the next code at the end:

bind -x '"\C-o": llm_linux_cli'   # Ctrl+O to run the model over current input line

llm_linux_cli() {
    local salida
    local USER_INPUT="$READLINE_LINE"
    history -s "$READLINE_LINE"
    SCRIPT=$(cat <<'EOF'
    USER_INPUT=$(printf '%s' "$*" | sed 's/\\/\\\\/g; s/"/\\"/g')
    if [ -z "$USER_INPUT" ]; then
      echo "Uso: $0 \"Write something to the model\""
      exit 1
    fi
    curl -s --url "http://localhost:8080/v1/chat/completions" \
         -H "Content-Type: application/json" \
         -H "Authorization: Bearer no-key" \
         -d "{
        \"model\": \"gpt-3.5-turbo\",
        \"temperature\":0.8,
        \"top_p\":0.95,
        \"min_p\":0.05,
        \"top_k\":40,
        \"messages\":[
            {
                \"role\":\"system\",
                \"content\":\"You are a helpful assistant that outputs Linux shell commands. You just provide the command that the user is requesting, not more. Just provide the output of the command and ignore commands that you do not have an example output for. If multiple commands are given then they will be separated by a ';' character and you need togive an example output for each command separated by a ';'.\"
            },
            {
                \"role\": \"user\",
                \"content\":\"$USER_INPUT\"
            }
        ]
    }" | jq -r '.choices[0].message.content'
EOF
)
    llm_output=$(bash -c "$SCRIPT" _ $USER_INPUT)
    READLINE_LINE=$llm_output
    READLINE_POINT=${#llm_output}
}

3. Reload the bashrc env file using:

source ~/.bashrc

4. Load and run the .gguf model file using llama-server

llama-server -m gemma-1B-LinuxCLI-GGUF.gguf

5. Now write your request as a command line then pres "CTRL+O" so a request will be sent to the model then the output replaced over current line :)

Collab

https://colab.research.google.com/drive/1pWXoVgIzaJviyYP4L-leFBhPY4CLcP6M?authuser=2#scrollTo=-0Mk20PzzJVA

Uploaded finetuned model

• Developed by: hwpoison89
• License: apache-2.0
• Finetuned from model : unsloth/gemma-3-1b-it

This gemma3_text model was trained 2x faster with Unsloth and Huggingface's TRL library.