@HuggingPapers: NVIDIA just released an optimized GLM-5.2 on Hugging Face A 753B parameter MoE with 1M context, quantized to NVFP4 for …
Summary
NVIDIA released an optimized GLM-5.2 MoE model on Hugging Face with 753B parameters and 1M context, quantized to NVFP4 for Blackwell GPUs while nearly matching FP8 accuracy.
View Cached Full Text
Cached at: 06/27/26, 05:53 AM
NVIDIA just released an optimized GLM-5.2 on Hugging Face
A 753B parameter MoE with 1M context, quantized to NVFP4 for Blackwell GPUs— nearly matching FP8 accuracy. https://t.co/tjtk0dVPEW
Similar Articles
nvidia/GLM-5.2-NVFP4
NVIDIA released GLM-5.2-NVFP4, a quantized version of ZAI's GLM-5.2 MoE language model optimized for inference on NVIDIA Blackwell GPUs using Model Optimizer.
@mr_r0b0t: Official @NVIDIAAI GLM5.1-NVFP4 spotted on @huggingface
NVIDIA releases GLM-5.1-NVFP4, a quantized version of ZAI's GLM-5.1 model with 754B total parameters (40B activated), available on Hugging Face under MIT license.
@lmsysorg: NVIDIA just released an NVFP4 checkpoint of GLM-5.2 from @Zai_org, a 744B MoE (40B active) for reasoning & coding. Day-…
NVIDIA released an NVFP4 quantized checkpoint of GLM-5.2, a 744B MoE model (40B active) optimized for reasoning and coding, with day-0 support in SGLang.
@0xSero: We found a way to run GLM-5.2 with full context in vLLM without pruning. - top 32 experts NVFP4 - rest fp3 - intel auto…
A community researcher enabled running GLM-5.2 (753B parameters, all 256 experts) in vLLM without pruning via a hybrid quantization (NVFP4, NF3, MXFP8), fitting on 4×96GB GPUs with ~307k KV cache and near-FP8 accuracy.
@tolak_eth: I wanted to share how we avoided spending roughly $160k/year to host GLM-5.2 with its full 1M context. When GLM-5.2 lau…
Phala avoided $160k/year hosting costs for GLM-5.2 with full 1M context by quantizing MoE experts to 4-bit and keeping critical parts in FP8/BF16, achieving the same benchmark results on a single 8×H200 node and releasing the optimized model GLM-5.2-W4AFP8 on Hugging Face.