K-EXAONE-236B-A23B is a Mixture-of-Experts language model developed by LG AI Research, featuring 236B total parameters with 23B active parameters per token, 128 routed experts, and 1 shared expert per layer.

Quantization Details

Property	Value
Base Model	LGAI-EXAONE/K-EXAONE-236B-A23B (236B params, MoE)
Precision	MXFP4 (E2M1 weights + E8M0 scales, group_size=32)
Activation Quantization	MXFP4 (E2M1), Dynamic, group_size=32
Quantization Tool	quanto + AMD Quark 0.11.1 (file-to-file)
Algorithm	RTN (Round-To-Nearest)
Original Size	443 GB
Quantized Size	130 GB
Compression Ratio	3.4x

Excluded Layers

This model uses the attn-excl strategy: all attention projections and all shared expert projections are excluded from quantization, leaving only MoE routed expert weights quantized to MXFP4.

Self-Attention (all 48 layers): q_proj, k_proj, v_proj, o_proj
Shared Expert MLP (all 48 layers): gate_proj, up_proj, down_proj
MoE Router Gates: All mlp.gate layers
Standard exclusions: lm_head, embed_tokens, all norm layers

Total: 382 excluded layers.

Hardware

Quantized on: AMD MI355 (gfx950), 288 GB VRAM
Tested with: vLLM v0.18.2 (ROCm), TP=1
Compatible with: vLLM with Quark quantization support (quant_method: "quark")

Usage

vLLM

bash

vllm serve haanjack/K-EXAONE-236B-A23B-MXFP4 \
  --trust-remote-code \
  --tensor-parallel-size 1 \
  --max-model-len 4096

Quantization Reproduction

python

from quanto import UnifiedQuantizer, UnifiedConfig

NUM_LAYERS = 48
exclude = ["lm_head", "*embed*", "*norm*", "*.gate"]
for i in range(NUM_LAYERS):
    for proj in ["q_proj", "k_proj", "v_proj", "o_proj"]:
        exclude.append(f"model.layers.{i}.self_attn.{proj}")
    for proj in ["gate_proj", "up_proj", "down_proj"]:
        exclude.append(f"model.layers.{i}.mlp.shared_experts.{proj}")

config = UnifiedConfig(
    model_path="LGAI-EXAONE/K-EXAONE-236B-A23B",
    output_dir="./K-EXAONE-236B-A23B-MXFP4",
    precision="mxfp4",
    sensitivity_analysis=False,
    exclude_layers=exclude,
    trust_remote_code=True,
)
UnifiedQuantizer(config).run()

Credits

Base Model: LG AI Research — K-EXAONE
Quantization: quanto with AMD Quark
Hardware: AMD MI355 (gfx950), 288 GB VRAM

License

This model inherits the EXAONE License from the base model.