---
language:
- ko
- en
license: llama3
library_name: transformers
tags:
- moe
- awq
- quantized
- w4a16
- compressed-tensors
- vllm
- llm-compressor
base_model: LGAI-EXAONE/K-EXAONE-236B-A23B
---

# K-EXAONE-236B-A23B-W4A16-G128

**W4A16 AWQ quantization** of [`LGAI-EXAONE/K-EXAONE-236B-A23B`](https://huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B), produced with [llm-compressor](https://github.com/vllm-project/llm-compressor).

This is the **first W4A16 AWQ checkpoint** for K-EXAONE-236B-A23B publicly available — the original model only has FP8 and GGUF variants on HuggingFace.

---

## Model Details

| Property | Value |
|---|---|
| Base model | LGAI-EXAONE/K-EXAONE-236B-A23B |
| Architecture | ExaoneMoeForCausalLM |
| Total parameters | ~236B |
| Active parameters | ~23B per token |
| Quantization method | AWQ (Activation-aware Weight Quantization) |
| Weight precision | INT4 (packed) |
| Activation precision | BF16 |
| Group size | 128 |
| Quantization scope | All `Linear` layers except `lm_head` and gate projections |
| Compressed-tensors version | 0.15.0 |
| Context length | 262,144 tokens |
| Languages | Korean, English |

### Architecture Highlights

- **48 transformer layers** with mixed sliding-window (`LLLG` pattern) and full attention
- **MoE layers**: 47 sparse MoE layers + 1 dense MLP (layer 0)
- **128 routed experts** + 1 shared expert per MoE layer; top-8 experts activated per token
- **Sigmoid scoring** with `norm_topk_prob=True`
- **Hidden size**: 6144, **MoE intermediate size**: 2048

---

## Quantization Details

Quantization was performed using [llm-compressor](https://github.com/vllm-project/llm-compressor) with a **MoE-aware AWQ** recipe.

**Method:** AWQ applies channel-wise scaling to minimize quantization error by protecting salient weights, using a calibration dataset to determine optimal scales.

**Recipe highlights:**
- `scheme`: W4A16 (INT4 weights, BF16 activations)
- `group_size`: 128
- `n_grid`: 20 (search resolution for AWQ scale optimization)
- `duo_scaling`: True
- Smooth mappings cover all MoE expert layers (layers 1–47) independently, plus attention and MLP projections
- Layer 0 (dense MLP) and `lm_head` are excluded from quantization
- Gate weight tensors are excluded from quantization

The full recipe is available in `recipe.yaml`.

**Calibration dataset:** [`neuralmagic/LLM_compression_calibration`](https://huggingface.co/datasets/neuralmagic/LLM_compression_calibration) (512 samples, sequence length 2048)

---

## Usage

### vLLM (Recommended)

Install vLLM (≥0.6.0 recommended for compressed-tensors support):

```bash
pip install vllm
```

```python
from vllm import LLM, SamplingParams

llm = LLM(
    model="Hyun9junn/K-EXAONE-236B-A23B-W4A16-G128",
    max_model_len=8192,
    trust_remote_code=True,   # K-EXAONE uses custom modeling code
    tensor_parallel_size=4,   # adjust to the number of GPUs available
)

sampling_params = SamplingParams(
    temperature=0.6,
    top_p=0.9,
    max_tokens=512,
)

tokenizer = llm.get_tokenizer()

prompts = [
    "What is the capital of South Korea?",
    "Explain the difference between MoE and dense transformer models.",
]

formatted_prompts = [
    tokenizer.apply_chat_template(
        [{"role": "user", "content": p}],
        tokenize=False,
        add_generation_prompt=True,
    )
    for p in prompts
]

outputs = llm.generate(formatted_prompts, sampling_params)

for prompt, output in zip(prompts, outputs):
    print(f"Prompt : {prompt}")
    print(f"Response: {output.outputs[0].text.strip()}")
```

### Transformers

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "Hyun9junn/K-EXAONE-236B-A23B-W4A16-G128"

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

messages = [{"role": "user", "content": "한국의 수도는 어디인가요?"}]
input_ids = tokenizer.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=True, return_tensors="pt"
).to(model.device)

output = model.generate(input_ids, max_new_tokens=256, temperature=0.6, top_p=0.9)
print(tokenizer.decode(output[0][input_ids.shape[-1]:], skip_special_tokens=True))
```

---

## Hardware Requirements

| Precision | Min VRAM |
|---|---|
| This model (W4A16) | ~120 GB |
| Original BF16 | ~480 GB |

Tested on: NVIDIA B200 (180 GB HBM3e).

For multi-GPU inference, set `tensor_parallel_size` in vLLM to the number of GPUs.

---

## Files

| File | Description |
|---|---|
| `model-00001-of-00003.safetensors` | Model weights shard 1/3 |
| `model-00002-of-00003.safetensors` | Model weights shard 2/3 |
| `model-00003-of-00003.safetensors` | Model weights shard 3/3 |
| `model.safetensors.index.json` | Weight shard index |
| `config.json` | Model config with quantization metadata |
| `recipe.yaml` | llm-compressor AWQ recipe used for quantization |
| `tokenizer.json` | Tokenizer |
| `tokenizer_config.json` | Tokenizer config |
| `chat_template.jinja` | Chat template |
| `generation_config.json` | Default generation config |

---

## License

This model inherits the license of the base model [`LGAI-EXAONE/K-EXAONE-236B-A23B`](https://huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B). Please refer to the original model page for license details.

---

## Citation

If you use this model, please cite the original K-EXAONE work:

```
@misc{k-exaone-236b,
  title  = {K-EXAONE-236B-A23B},
  author = {LG AI Research},
  year   = {2025},
  url    = {https://huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B}
}
```

Quantization produced by [Hyun9junn](https://huggingface.co/Hyun9junn) using [llm-compressor](https://github.com/vllm-project/llm-compressor).