README.md

---
base_model: nvidia/Nemotron-Mini-4B-Instruct
license: other
license_name: nvidia-community-model-license
inference: false
model_creator: nvidia
model_name: Nemotron-Mini-4B-Instruct
quantized_by: Second State Inc.
library_name: nemo
---

<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://github.com/LlamaEdge/LlamaEdge/raw/dev/assets/logo.svg" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->

# Nemotron-Mini-4B-Instruct-GGUF

## Original Model

[nvidia/Nemotron-Mini-4B-Instruct](https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct)

## Run with LlamaEdge

- LlamaEdge version: coming soon

<!-- - LlamaEdge version: [v0.14.2](https://github.com/LlamaEdge/LlamaEdge/releases/tag/0.14.2) and above -->

- Prompt template

  - Prompt type: `nemotron-chat`

  - Prompt string

    ```text
    <extra_id_0>System
    {system_message}
    <extra_id_1>User
    {user_message_1}<extra_id_1>Assistant
    {assistant_message_1}
    <extra_id_1>User
    {user_message_2}<extra_id_1>Assistant
    {assistant_message_2}
    <extra_id_1>User
    {user_message_3}
    <extra_id_1>Assistant\n
    ```

    - Tool use

      ```text
      <extra_id_0>System
      {system_message}
      <tool> {tool_1} </tool>
      <tool> {tool_2} </tool>
    
    
      <extra_id_1>User
      {user_message_1}<extra_id_1>Assistant
      <toolcall> {tool_call_message} </toolcall>
      <extra_id_1>Tool
      {tool_result_message}
      <extra_id_1>Assistant\n
      ```

- Context size: `4096`

- Run as LlamaEdge service

  ```bash
  wasmedge --dir .:. --nn-preload default:GGML:AUTO:Nemotron-Mini-4B-Instruct-Q5_K_M.gguf \
    llama-api-server.wasm \
    --prompt-template nemotron-chat \
    --ctx-size 4096 \
    --model-name Nemotron-Mini-4B-Instruct
  ```

  - Tool use

    ```bash
    wasmedge --dir .:. --nn-preload default:GGML:AUTO:Nemotron-Mini-4B-Instruct-Q5_K_M.gguf \
      llama-api-server.wasm \
      --prompt-template nemotron-tool \
      --ctx-size 4096 \
      --model-name Nemotron-Mini-4B-Instruct
    ```

- Run as LlamaEdge command app

  ```bash
  wasmedge --dir .:. --nn-preload default:GGML:AUTO:Nemotron-Mini-4B-Instruct-Q5_K_M.gguf \
    llama-chat.wasm \
    --prompt-template nemotron-chat \
    --ctx-size 4096
  ```

## Quantized GGUF Models

| Name | Quant method | Bits | Size | Use case |
| ---- | ---- | ---- | ---- | ----- |
| [Nemotron-Mini-4B-Instruct-Q2_K.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q2_K.gguf)     | Q2_K   | 2 | 3.35 GB| smallest, significant quality loss - not recommended for most purposes |
| [Nemotron-Mini-4B-Instruct-Q3_K_L.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q3_K_L.gguf) | Q3_K_L | 3 | 4.69 GB| small, substantial quality loss |
| [Nemotron-Mini-4B-Instruct-Q3_K_M.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q3_K_M.gguf) | Q3_K_M | 3 | 4.32 GB| very small, high quality loss |
| [Nemotron-Mini-4B-Instruct-Q3_K_S.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q3_K_S.gguf) | Q3_K_S | 3 | 3.90 GB| very small, high quality loss |
| [Nemotron-Mini-4B-Instruct-Q4_0.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q4_0.gguf)     | Q4_0   | 4 | 5.04 GB| legacy; small, very high quality loss - prefer using Q3_K_M |
| [Nemotron-Mini-4B-Instruct-Q4_K_M.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q4_K_M.gguf) | Q4_K_M | 4 | 5.33 GB| medium, balanced quality - recommended |
| [Nemotron-Mini-4B-Instruct-Q4_K_S.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q4_K_S.gguf) | Q4_K_S | 4 | 5.07 GB| small, greater quality loss |
| [Nemotron-Mini-4B-Instruct-Q5_0.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q5_0.gguf)     | Q5_0   | 5 | 6.11 GB| legacy; medium, balanced quality - prefer using Q4_K_M |
| [Nemotron-Mini-4B-Instruct-Q5_K_M.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q5_K_M.gguf) | Q5_K_M | 5 | 6.26 GB| large, very low quality loss - recommended |
| [Nemotron-Mini-4B-Instruct-Q5_K_S.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q5_K_S.gguf) | Q5_K_S | 5 | 6.11 GB| large, low quality loss - recommended |
| [Nemotron-Mini-4B-Instruct-Q6_K.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q6_K.gguf)     | Q6_K   | 6 | 7.25 GB| very large, extremely low quality loss |
| [Nemotron-Mini-4B-Instruct-Q8_0.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-Q8_0.gguf)     | Q8_0   | 8 | 9.38 GB| very large, extremely low quality loss - not recommended |
| [Nemotron-Mini-4B-Instruct-f16.gguf](https://huggingface.co/second-state/Nemotron-Mini-4B-Instruct-GGUF/blob/main/Nemotron-Mini-4B-Instruct-f16.gguf)       | f16   | 16 | 17.7 GB| |

*Quantized with llama.cpp b3751*