GLM-5.2-MXFP4xMXFP8
Model Overview
- Model Architecture: GlmMoeDsaForCausalLM
- Input: Text
- Output: Text
- Model Optimizations:
- Weight quantization: MXFP4
- Activation quantization: MXFP8
- Release Date: 2026-09-15
- Version: 1.0
- Model Developers: RedHatAI
This model is a quantized version of zai-org/GLM-5.2.
Model Optimizations
This model was obtained by quantizing the MoE linears (experts and shared experts) of zai-org/GLM-5.2 to 4-bit MXFP4 weights with dynamic 8-bit MXFP8 activation scaling, ready for inference with vLLM.
This optimization reduces the number of bits per parameter from 16 to an average of ~4.5 bits, shrinking the on-disk model size from ~1.47 TB (BF16) to ~424 GB and reducing disk and GPU memory requirements by approximately 72%.
Only the linear operators within transformer blocks are quantized using LLM Compressor.
Note: this quantized checkpoint is experimentally supported in vLLM.
Deployment
Use with vLLM
vllm serve RedHatAI/GLM-5.2-MXFP4xMXFP8 \
--tensor-parallel-size 8 \
--kv-cache-dtype fp8_e4m3 \
--speculative-config '{"method": "mtp", "num_speculative_tokens": 5}' \
--tool-call-parser glm47 \
--reasoning-parser glm45 \
--enable-auto-tool-choice
Creation
This model was created by applying LLM Compressor with the quantization scheme described above (RTN, no calibration data), exported in compressed-tensors format.