Quantized checkpoints for GLM-5.2 have been created by the Red Hat AI team!

https://huggingface.co/RedHatAI/GLM-5.2-NVFP4-FP8…

This model was calibrated quantized using DDP + disk offloading in under 2 hours. The full precision model requires 1.6T of VRAM, but NVFP4 quantization of MoE layers and FP8 quantization of attention layers reduces the model size by >70% while maintaining state-of-the-art accuracy recovery on GPQA.

Pair it with the new DSpark speculator for additional throughput: https://huggingface.co/RedHatAI/GLM-5.2-speculator.dspark…

RedHatAI/GLM-5.2-NVFP4-FP8 · Hugging Face

Source: https://huggingface.co/RedHatAI/GLM-5.2-NVFP4-FP8 This is a quantized version ofzai\-org/GLM\-5\.2with MoE layers quantized to NVFP4 and attention layers quantized to FP8 block

https://huggingface.co/RedHatAI/GLM-5.2-NVFP4-FP8#usageUsage

This model is intended for deployment with vLLM and requires the following fix:https://github.com/vllm-project/vllm/pull/47780. You can serve the model using

vllm serve RedHatAI/GLM-5.2-NVFP4-FP8 \
    --tensor_parallel_size 4 \
    --reasoning-parser glm45 \
    --tool-call-parser glm47 \
    --enable-auto-tool-choice \
    --kv_cache_dtype=fp8 \

https://huggingface.co/RedHatAI/GLM-5.2-NVFP4-FP8#creation-processCreation Process

This model was created usingLLM Compressor. The example script can be found inexamples/quantizing\_moe/glm5\_example\.py[Example] GLM5.2 Example. Quantizing the model with data parallelism and 6xA100 takes about 3 hours.

LLM Compressor Creation Script``` import torch from compressed_tensors.offload import init_dist from compressed_tensors.quantization.quant_scheme import ( FP8_BLOCK, NVFP4, QuantizationScheme, ) from datasets import load_dataset from transformers import AutoModelForCausalLM, AutoTokenizer

from llmcompressor import oneshot from llmcompressor.datasets.utils import get_rank_partition from llmcompressor.modifiers.quantization import QuantizationModifier from llmcompressor.utils import load_context

Load the model

init_dist() model_id = “zai-org/GLM-5.2” with load_context(): model = AutoModelForCausalLM.from_pretrained( model_id, device_map=“auto_offload”, max_memory={}, offload_folder=“/mnt/nvme-data/engine/kylesayrs/offload_folder”, ) tokenizer = AutoTokenizer.from_pretrained(model_id)

Select calibration dataset.

DATASET_ID = “HuggingFaceH4/ultrachat_200k” DATASET_SPLIT = “train_sft”

Select number of samples. 512 samples is a good place to start.

Increasing the number of samples can improve accuracy.

NUM_CALIBRATION_SAMPLES = 512 MAX_SEQUENCE_LENGTH = 2048

Load dataset and preprocess.

ds = load_dataset( DATASET_ID, split=get_rank_partition(DATASET_SPLIT, NUM_CALIBRATION_SAMPLES) ) ds = ds.shuffle(seed=42)

def preprocess(example): return { “text”: tokenizer.apply_chat_template( example[“messages”], tokenize=False, ) }

ds = ds.map(preprocess)

Tokenize inputs.

def tokenize(sample): return tokenizer( sample[“text”], padding=False, max_length=MAX_SEQUENCE_LENGTH, truncation=True, add_special_tokens=False, )

ds = ds.map(tokenize, remove_columns=ds.column_names)

Configure the quantization algorithm to run.

recipe = QuantizationModifier( config_groups={ “attention_shared_experts”: QuantizationScheme( targets=[r“re:.self_attn..“], **FP8_BLOCK, ), “mlp”: QuantizationScheme( targets=[r“re:.mlp..“], **NVFP4, ), }, ignore=[ r“re:^model.layers.[0-2]..*” r“re:.mlp.gate.“, # not technically necessary r“re:.*indexer.weights_proj$”, # sensitive to quantization r“lm_head“, ], )

Apply algorithms.

oneshot( model=model, dataset=ds, batch_size=4, recipe=recipe, shuffle_calibration_samples=False, )

Save to disk compressed.

Note: base checkpoint generation_config needs fixing for newer transformers versions

model.generation_config.top_p = None SAVE_DIR = ( “/mnt/nvme-data/engine/kylesayrs/” + model_id.rstrip(“/”).split(“/”)[-1] + “-NVFP4-FP8” ) model.save_pretrained(SAVE_DIR, save_compressed=True) tokenizer.save_pretrained(SAVE_DIR)

torch.distributed.destroy_process_group()

## [https://huggingface.co/RedHatAI/GLM-5.2-NVFP4-FP8#evaluation](https://huggingface.co/RedHatAI/GLM-5.2-NVFP4-FP8#evaluation)Evaluation

Benchmark`zai\-org/GLM\-5\.2``RedHatAI/GLM\-5\.2\-NVFP4\-FP8`GPQA\-Diamond91\.289\.1