Midm-2.0-Base-Instruct

K-intelligence/Midm-2.0-Base-Instruct

Midm-2.0-Base-Instruct is a text generation model from K-intelligence. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as mit.

<p align="center" <br <span style="font-size: 60px; font-weight: bold;"Mi:dm 2.0 Base</span </br </p

Downloads · 30 days

26.3K

6% of all-time downloads

All-time downloads

464K

Public

Parameters

11.5B

23.1 GB on disk

Likes

141

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors23.1 GB · 100%

At a glance

Task: Text Generation
Library: transformers
License: mit
Model type: llama
Access: Public
Created: Jul 3, 2025
Updated: Oct 29, 2025
SHA: 35479c5f

Try a prompt

Spaces

Model card

From the Hugging Face model README

Mi:dm 2.0 Base 🤗 <a href="https://huggingface.co/collections/K-intelligence/mi-dm-20-6866406c301e5f45a6926af8">Mi:dm 2.0 Models</a> | 📜 <a href="https://github.com/K-intelligence-Midm/Midm-2.0/blob/main/Mi_dm2_0__technical_report.pdf">Mi:dm 2.0 Technical Report</a> | 📕 <a href="https://kode.kt.com/blog/article/3935">Mi:dm 2.0 Technical Blog</a>

News 📢

🔧2025/10/29: Added support for function calling on vLLM with Mi:dm 2.0 parser.
📕2025/08/08: Published a technical blog article about Mi:dm 2.0 Model.
⚡️2025/07/04: Released Mi:dm 2.0 Model collection on Hugging Face🤗.

# Table of Contents

Overview
Usage
More Information

Overview

Mi:dm 2.0

Mi:dm 2.0 is a "Korea-centric AI" model developed using KT's proprietary technology. The term "Korea-centric AI" refers to a model that deeply internalizes the unique values, cognitive frameworks, and commonsense reasoning inherent to Korean society. It goes beyond simply processing or generating Korean text—it reflects a deeper understanding of the socio-cultural norms and values that define Korean society.

Mi:dm 2.0 is released in two versions:

Mi:dm 2.0 Base
An 11.5B parameter dense model designed to balance model size and performance.
It extends an 8B-scale model by applying the Depth-up Scaling (DuS) method, making it suitable for real-world applications that require both performance and versatility.
Mi:dm 2.0 Mini
A lightweight 2.3B parameter dense model optimized for on-device environments and systems with limited GPU resources.
It was derived from the Base model through pruning and distillation to enable compact deployment.

[!Note] Neither the pre-training nor the post-training data includes KT users' data.

Quickstart

Here is the code snippet to run conversational inference with the model:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig

model_name = "K-intelligence/Midm-2.0-Base-Instruct"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
generation_config = GenerationConfig.from_pretrained(model_name)

prompt = "KT에 대해 소개해줘"

# message for inference
messages = [
    {"role": "system", 
     "content": "Mi:dm(믿:음)은 KT에서 개발한 AI 기반 어시스턴트이다."},
    {"role": "user", "content": prompt}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
)

output = model.generate(
    input_ids.to("cuda"),
    generation_config=generation_config,
    eos_token_id=tokenizer.eos_token_id,
    max_new_tokens=128,
    do_sample=False,
)
print(tokenizer.decode(output[0]))

[!NOTE] The transformers library should be version 4.45.0 or higher.

Evaluation

Korean

<table> <tr> <th rowspan="2">Model</th> <th colspan="5" align="center">Society & Culture</th> <th colspan="3" align="center">General Knowledge</th> <th colspan="3" align="center">Instruction Following</th> </tr> <tr> <th align="center">K-Refer*</th> <th align="center">K-Refer-Hard*</th> <th align="center">Ko-Sovereign*</th> <th align="center">HAERAE</th> <th align="center">Avg.</th> <th align="center">KMMLU</th> <th align="center">Ko-Sovereign*</th> <th align="center">Avg.</th> <th align="center">Ko-IFEval</th> <th align="center">Ko-MTBench</th> <th align="center">Avg.</th> </tr>  <tr> <td>Qwen3-4B</td> <td align="center">53.6</td> <td align="center">42.9</td> <td align="center">35.8</td> <td align="center">50.6</td> <td align="center">45.7</td> <td align="center">50.6</td> <td align="center">42.5</td> <td align="center">46.5</td> <td align="center">75.9</td> <td align="center">63.0</td> <td align="center">69.4</td> </tr> <tr> <td>Exaone-3.5-2.4B-inst</td> <td align="center">64.0</td> <td align="center">67.1</td> <td align="center">44.4</td> <td align="center">61.3</td> <td align="center">59.2</td> <td align="center">43.5</td> <td align="center">42.4</td> <td align="center">43.0</td> <td align="center">65.4</td> <td align="center">74.0</td> <td align="center">68.9</td> </tr> <tr> <td>Mi:dm 2.0-Mini-inst</td> <td align="center">66.4</td> <td align="center">61.4</td> <td align="center">36.7</td> <td align="center">70.8</td> <td align="center">58.8</td> <td align="center">45.1</td> <td align="center">42.4</td> <td align="center">43.8</td> <td align="center">73.3</td> <td align="center">74.0</td> <td align="center">73.6</td> </tr>  <tr><td colspan="13"> </td></tr>  <tr> <td>Qwen3-14B</td> <td align="center">72.4</td> <td align="center">65.7</td> <td align="center">49.8</td> <td align="center">68.4</td> <td align="center">64.1</td> <td align="center">55.4</td> <td align="center">54.7</td> <td align="center">55.1</td> <td align="center">83.6</td> <td align="center">71</td> <td align="center">77.3</td> </tr> <tr> <td>Llama-3.1-8B-inst</td> <td align="center">43.2</td> <td align="center">36.4</td> <td align="center">33.8</td> <td align="center">49.5</td> <td align="center">40.7</td> <td align="center">33.0</td> <td align="center">36.7</td> <td align="center">34.8</td> <td align="center">60.1</td> <td align="center">57</td> <td align="center">58.5</td> </tr> <tr> <td>Exaone-3.5-7.8B-inst</td> <td align="center">71.6</td> <td align="center">69.3</td> <td align="center">46.9</td> <td align="center">72.9</td> <td align="center">65.2</td> <td align="center">52.6</td> <td align="center">45.6</td> <td align="center">49.1</td> <td align="center">69.1</td> <td align="center">79.6</td> <td align="center">74.4</td> </tr> <tr> <td>Mi:dm 2.0-Base-inst</td> <td align="center">89.6</td> <td align="center">86.4</td> <td align="center">56.3</td> <td align="center">81.5</td> <td align="center">78.4</td> <td align="center">57.3</td> <td align="center">58.0</td> <td align="center">57.7</td> <td align="center">82</td> <td align="center">89.7</td> <td align="center">85.9</td> </tr> </table>  <table> <tr> <th rowspan="2" align="center">Model</th> <th colspan="5" align="center">Comprehension</th> <th colspan="5" align="center">Reasoning</th> </tr> <tr> <th align="center">K-Prag*</th> <th align="center">K-Refer-Hard*</th> <th align="center">Ko-Best</th> <th align="center">Ko-Sovereign*</th> <th align="center">Avg.</th> <th align="center">Ko-Winogrande</th> <th align="center">Ko-Best</th> <th align="center">LogicKor</th> <th align="center">HRM8K</th> <th align="center">Avg.</th> </tr>  <tr> <td>Qwen3-4B</td> <td align="center">73.9</td> <td align="center">56.7</td> <td align="center">91.5</td> <td align="center">43.5</td> <td align="center">66.6</td> <td align="center">67.5</td> <td align="center">69.2</td> <td align="center">5.6</td> <td align="center">56.7</td> <td align="center">43.8</td> </tr> <tr> <td>Exaone-3.5-2.4B-inst</td> <td align="center">68.7</td> <td align="center">58.5</td> <td align="center">87.2</td> <td align="center">38.0</td> <td align="center">62.5</td> <td align="center">60.3</td> <td align="center">64.1</td> <td align="center">7.4</td> <td align="center">38.5</td> <td align="center">36.7</td> </tr> <tr> <td>Mi:dm 2.0-Mini-inst</td> <td align="center">69.5</td> <td align="center">55.4</td> <td align="center">80.5</td> <td align="center">42.5</td> <td align="center">61.9</td> <td align="center">61.7</td> <td align="center">64.5</td> <td align="center">7.7</td> <td align="center">39.9</td> <td align="center">37.4</td> </tr>  <tr><td colspan="11"> </td></tr>  <tr> <td>Qwen3-14B</td> <td align="center">86.7</td> <td align="center">74.0</td> <td align="center">93.9</td> <td align="center">52.0</td> <td align="center">76.8</td> <td align="center">77.2</td> <td align="center">75.4</td> <td align="center">6.4</td> <td align="center">64.5</td> <td align="center">48.8</td> </tr> <tr> <td>Llama-3.1-8B-inst</td> <td align="center">59.9</td> <td align="center">48.6</td> <td align="center">77.4</td> <td align="center">31.5</td> <td align="center">51.5</td> <td align="center">40.1</td> <td align="center">26.0</td> <td align="center">2.4</td> <td align="center">30.9</td> <td align="center">19.8</td> </tr> <tr> <td>Exaone-3.5-7.8B-inst</td> <td align="center">73.5</td> <td align="center">61.9</td> <td align="center">92.0</td> <td align="center">44.0</td> <td align="center">67.2</td> <td align="center">64.6</td> <td align="center">60.3</td> <td align="center">8.6</td> <td align="center">49.7</td> <td align="center">39.5</td> </tr> <tr> <td>Mi:dm 2.0-Base-inst</td> <td align="center">86.5</td> <td align="center">70.8</td> <td align="center">95.2</td> <td align="center">53.0</td> <td align="center">76.1</td> <td align="center">75.1</td> <td align="center">73.0</td> <td align="center">8.6</td> <td align="center">52.9</td> <td align="center">44.8</td> </tr> </table>

* indicates KT proprietary evaluation resources.

English

<table> <tr> <th rowspan="2" align="center">Model</th> <th align="center">Instruction</th> <th colspan="4" align="center">Reasoning</th> <th align="center">Math</th> <th align="center">Coding</th> <th colspan="3" align="center">General Knowledge</th> </tr> <tr> <th align="center">IFEval</th> <th align="center">BBH</th> <th align="center">GPQA</th> <th align="center">MuSR</th> <th align="center">Avg.</th> <th align="center">GSM8K</th> <th align="center">MBPP+</th> <th align="center">MMLU-pro</th> <th align="center">MMLU</th> <th align="center">Avg.</th> </tr>  <tr> <td>Qwen3-4B</td> <td align="center">79.7</td> <td align="center">79.0</td> <td align="center">39.8</td> <td align="center">58.5</td> <td align="center">59.1</td> <td align="center">90.4</td> <td align="center">62.4</td> <td align="center">-</td> <td align="center">73.3</td> <td align="center">73.3</td> </tr> <tr> <td>Exaone-3.5-2.4B-inst</td> <td align="center">81.1</td> <td align="center">46.4</td> <td align="center">28.1</td> <td align="center">49.7</td> <td align="center">41.4</td> <td align="center">82.5</td> <td align="center">59.8</td> <td align="center">-</td> <td align="center">59.5</td> <td align="center">59.5</td> </tr> <tr> <td>Mi:dm 2.0-Mini-inst</td> <td align="center">73.6</td> <td align="center">44.5</td> <td align="center">26.6</td> <td align="center">51.7</td> <td align="center">40.9</td> <td align="center">83.1</td> <td align="center">60.9</td> <td align="center">-</td> <td align="center">56.5</td> <td align="center">56.5</td> </tr> <tr><td colspan="11"> </td></tr>  <tr> <td>Qwen3-14B</td> <td align="center">83.9</td> <td align="center">83.4</td> <td align="center">49.8</td> <td align="center">57.7</td> <td align="center">63.6</td> <td align="center">88.0</td> <td align="center">73.4</td> <td align="center">70.5</td> <td align="center">82.7</td> <td align="center">76.6</td> </tr> <tr> <td>Llama-3.1-8B-inst</td> <td align="center">79.9</td> <td align="center">60.3</td> <td align="center">21.6</td> <td align="center">50.3</td> <td align="center">44.1</td> <td align="center">81.2</td> <td align="center">81.8</td> <td align="center">47.6</td> <td align="center">70.7</td> <td align="center">59.2</td> </tr> <tr> <td>Exaone-3.5-7.8B-inst</td> <td align="center">83.6</td> <td align="center">50.1</td> <td align="center">33.1</td> <td align="center">51.2</td> <td align="center">44.8</td> <td align="center">81.1</td> <td align="center">79.4</td> <td align="center">40.7</td> <td align="center">69.0</td> <td align="center">54.8</td> </tr> <tr> <td>Mi:dm 2.0-Base-inst</td> <td align="center">84.0</td> <td align="center">77.7</td> <td align="center">33.5</td> <td align="center">51.9</td> <td align="center">54.4</td> <td align="center">91.6</td> <td align="center">77.5</td> <td align="center">53.3</td> <td align="center">73.7</td> <td align="center">63.5</td> </tr> </table>

Usage

Run on Friendli.AI

You can try our model immediately via Friendli.AI. Simply click Deploy and then Friendli Endpoints.

[!Note] Please note that a login to Friendli.AI is required after your fifth chat interaction.

Run on Your Local Machine

We provide a detailed description about running Mi:dm 2.0 on your local machine using llama.cpp, LM Studio, and Ollama. Please check our github for more information

Deployment

Basic Serving

To serve Mi:dm 2.0 using vLLM(>=0.8.0) with an OpenAI-compatible API:

vllm serve K-intelligence/Midm-2.0-Base-Instruct

With Function Calling

For advanced function calling tasks, you can serve Mi:dm 2.0 with our own tool parser:

Download and place Mi:dm 2.0 parser file in your working directory.

Run the following Docker command to launch the vLLM server with our custom parser file:

docker run --rm -it --gpus all -p 8000:8000 \
    -e HUGGING_FACE_HUB_TOKEN="<YOUR_HUGGINGFACE_TOKEN>" \
    -v "$(pwd)/midm_parser.py:/custom/midm_parser.py" \
    vllm/vllm-openai:v0.11.0 \
    --model K-intelligence/Midm-2.0-Base-Instruct \
    --enable-auto-tool-choice \
    --tool-parser-plugin /custom/midm_parser.py \
    --tool-call-parser midm-parser \
    --host 0.0.0.0

[!Note] This setup is compatible with vllm/vllm-openai:v0.8.0 and later, but we strongly recommend using v0.11.0 for optimal stability and compatibility with our parser.

Tutorials

To help our end-users easily use Mi:dm 2.0, we have provided comprehensive tutorials on github.

More Information

Limitation

The training data for both Mi:dm 2.0 models consists primarily of English and Korean. Understanding and generation in other languages are not guaranteed.
The model is not guaranteed to provide reliable advice in fields that require professional expertise, such as law, medicine, or finance.
Researchers have made efforts to exclude unethical content from the training data — such as profanity, slurs, bias, and discriminatory language. However, despite these efforts, the model may still produce inappropriate expressions or factual inaccuracies.

License

Mi:dm 2.0 is licensed under the MIT License.

Contact

Mi:dm 2.0 Technical Inquiries: midm-llm@kt.com

Task: Text Generation
Library: transformers
Type: llama
License: mit
Languages: en, ko
Created: Jul 3, 2025
Updated: Oct 29, 2025