nvidia/GLM-5.2-NVFP4

Model Overview

Description:

The NVIDIA GLM-5.2 NVFP4 model is the quantized version of ZAI’s GLM-5.2 model, which is an auto-regressive language model that uses an optimized transformer architecture. GLM-5.2 is a Mixture-of-Experts (MoE) model for reasoning and coding that uses sparse attention (with an IndexShare indexer) to support a long context. For more information, please check here. The NVIDIA GLM-5.2 NVFP4 model is quantized with Model Optimizer.

This model is ready for commercial or non-commercial use.

License/Terms of Use:

GOVERNING TERMS: Use of the model is governed by the MIT License, same as the base model.

Deployment Geography:

Global

Use Case:

Developers looking to take off-the-shelf, pre-quantized models for deployment in AI Agent systems, chatbots, RAG systems, and other AI-powered applications.

Release Date:

Hugging Face 06/25/2026 via https://huggingface.co/nvidia/GLM-5.2-NVFP4

References

Nvidia Model Optimizer: https://github.com/NVIDIA/Model-Optimizer

Model Architecture:

Architecture Type: Transformers
Network Architecture: GLM-5.2 (GlmMoeDsaForCausalLM)
Number of Model Parameters: 753B in total and 40B activated

Input:

Input Type(s): Text
Input Format(s): String
Input Parameters: One-Dimensional (1D)
Other Properties Related to Input: Context length up to 1M

Output:

Output Type(s): Text
Output Format: String
Output Parameters: 1D (One-Dimensional): Sequences
Other Properties Related to Output: None

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

Software Integration:

Supported Runtime Engine(s):

• SGLang
  
• vLLM
  

Supported Hardware Microarchitecture Compatibility:

• NVIDIA Blackwell
  

Preferred Operating System(s):

• Linux
  

The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.

Model Version(s):

The model version is NVFP4 1.0 version and is quantized with nvidia-modelopt v0.46.0

Training, Testing, and Evaluation Datasets:

We calibrated the model using the dataset noted below, and performed evaluation using the benchmarks noted under Evaluation Datasets. We did not perform training or testing for this Model Optimizer release. The methods noted under Training and Testing Datasets below represent the data collection and labeling methods used by the third-party to train and test the underlying model.

Calibration Dataset:

Link: Nemotron-SFT-Instruction-Following-Chat-v2, Nemotron-Science-v1, Nemotron-Competitive-Programming-v1, Nemotron-SFT-Agentic-v2, Nemotron-Math-v2, Nemotron-SFT-SWE-v2, Nemotron-SFT-Multilingual-v1
Data Collection Method by dataset: Hybrid: Human, Synthetic, Automated.
Labeling method: Hybrid: Human, Automated.
Properties: Nemotron-SFT-Instruction-Following-Chat-v2 contains ~2M synthetic chat samples designed to strengthen open-ended chat and precise instruction following capabilities. Nemotron-Science-v1 is a synthetic science reasoning dataset with ~226K samples covering GPQA-style science questions and chemistry problems to enhance LLM reasoning in scientific domains. Nemotron-Competitive-Programming-v1 is a large-scale synthetic coding dataset with 2M+ Python and 1M+ C++ samples spanning 34K+ competitive programming questions for code completion and critique. Nemotron-SFT-Agentic-v2 contains ~992K samples of tool-calling trajectories, customer service conversations, and web-search trajectories to train interactive, tool-using agents. Nemotron-Math-v2 is a large-scale mathematical reasoning dataset with ~347K problems and 7M model-generated reasoning trajectories across multiple reasoning modes and tool-use configurations. Nemotron-SFT-SWE-v2 contains ~256K software engineering samples including agentic SWE trajectories and agentless code localization, repair, and test generation samples for SWE-Bench style tasks. Nemotron-SFT-Multilingual-v1 contains ~3M multilingual reasoning samples translated from math, code, and STEM data into German, French, Japanese, Italian, Chinese, and Spanish.

Training Dataset:

Data Modality: Undisclosed
Data Collection Method by dataset: Undisclosed
Labeling Method by dataset: Undisclosed
Properties: Undisclosed

Testing Dataset:

Data Collection Method by dataset: Undisclosed
Labeling Method by dataset: Undisclosed
Properties: Undisclosed

Evaluation Dataset:

Datasets: GPQA Diamond, SciCode, IFBench, AA-LCR, τ²-Bench Telecom
Data Collection Method by dataset: Hybrid: Automated, Human
Labeling Method by dataset: Hybrid: Human, Automated
Properties: We evaluated the model on text-based reasoning, coding, long-context recall, and agentic tool-use benchmarks: GPQA Diamond contains 448 graduate-level multiple-choice questions written by domain experts in biology, physics, and chemistry; SciCode evaluates scientific coding capabilities; IFBench is a benchmark for evaluating instruction-following capabilities across diverse and structured task constraints; AA-LCR (Artificial Analysis Long Context Recall) evaluates a model's ability to accurately retrieve and recall information from long input contexts; τ²-Bench Telecom evaluates agentic tool-use and policy-adherence capabilities in dual-control telecom customer-service scenarios where the model interacts with a simulated user and external tools to resolve account issues.

Inference:

Acceleration Engine: SGLang, vLLM
Test Hardware: NVIDIA B200
NVIDIA B300

Post Training Quantization

This model was obtained by quantizing the weights and activations of GLM-5.2 to NVFP4 data type, ready for inference with SGLang and vLLM. Only the weights and activations of the linear operators within transformer blocks in MoE experts are quantized. The shared expert is not quantized.

Usage

SGLang

This checkpoint was served with the latest SGLang image (lmsysorg/sglang:latest). GLM-5.2's glm_moe_dsa architecture requires transformers>=5.3.0, which we installed in the container before launching the server:

pip install -U "transformers>=5.3.0" && \
python3 -m sglang.launch_server \
    --model nvidia/GLM-5.2-NVFP4 \
    --tensor-parallel-size 8 \
    --quantization modelopt_fp4 \
    --tool-call-parser glm47 \
    --reasoning-parser glm45 \
    --trust-remote-code \
    --chunked-prefill-size 131072 \
    --mem-fraction-static 0.80

Evaluation

The accuracy benchmark results are presented in the table below. AA-LCR was measured with SGLang; all other benchmarks were measured with vLLM.

Precision	GPQA Diamond	SciCode	IFBench	AA-LCR	τ²-Bench Telecom
baseline (FP8)	89.52	49.85	74.95	69.38	97.9
NVFP4	89.39	49.04	75.81	70.13	98.25

Baseline: GLM-5.2-FP8. Benchmarked with temperature=1.0, top_p=0.95. GPQA Diamond used max_new_tokens=100000; all other benchmarks used max_new_tokens=64000.

Model Limitations:

The base model was trained on data that contains toxic language and societal biases originally crawled from the internet. Therefore, the model may amplify those biases and return toxic responses especially when prompted with toxic prompts. The model may generate answers that may be inaccurate, omit key information, or include irrelevant or redundant text producing socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive.

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Please make sure you have proper rights and permissions for all input image and video content; if image or video includes people, personal health information, or intellectual property, the image or video generated will not blur or maintain proportions of image subjects included.

For more detailed information on ethical considerations for this model, please see the Model Card++ Bias, Explainability, Safety & Security, and Privacy Subcards.

Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.

SUBCARDS:

Explainability

Field:	Response:
Intended Task/Domain:	Text generation, reasoning, summarization, and question answering.
Model Type:	Text and Image-to-text transformer
Intended Users:	This model is intended for developers, researchers, and customers building/utilizing LLMs, while balancing accuracy and efficiency.
Output:	Text String(s)
Describe how the model works:	Generates text by predicting the next word or token based on the context provided in the input sequence using multiple self-attention layers
Name the adversely impacted groups this has been tested to deliver comparable outcomes regardless of:	Not Applicable
Technical Limitations & Mitigation:	The model was trained on data that contains toxic language and societal biases originally crawled from the internet. Therefore, the model may amplify those biases and return toxic responses especially when prompted with toxic prompts. Therefore, before deploying any applications of this model, developers should perform safety testing and tuning tailored to their specific applications of the model.
Verified to have met prescribed quality standards?	Yes
Performance Metrics:	Accuracy, Throughput, and user-side throughput
Potential Known Risk	The model may generate answers that may be inaccurate, omit key information, or include irrelevant or redundant text producing socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive.
Licensing:	Your usage is governed by the following GOVERNING TERMS: Use of the model is governed by the MIT License, same as the base model.

Bias

Field:	Response:
Participation considerations from adversely impacted groups protected classes in model design and testing:	None
Measures taken to mitigate against unwanted bias:	None

Safety & Security

Field:	Response:
Model Application(s):	Chat, Instruction Following, Chatbot Development, Code Generation, Reasoning
Describe life critical application (if present):	Not Applicable
Use Case Restrictions:	Abide by the GOVERNING TERMS: Use of the model is governed by the MIT License, same as the base model.
Model and Dataset Restrictions:	The Principle of least privilege (PoLP) is applied limiting access for dataset generation. Restrictions enforce dataset access during training, and dataset license constraints adhered to. Model checkpoints are made available on Hugging Face, and may become available on cloud providers' model catalog.

Privacy

Field:	Response:
Generatable or Reverse engineerable personal data?	No
Personal data used to create this model?	No
Was consent obtained for any personal data used?	Not Applicable
How often is dataset reviewed?	Before Release
Was data from user interactions with the AI model (e.g. user input and prompts) used to train the model?	No
Is there provenance for all datasets used in training?	Yes
Does data labeling (annotation, metadata) comply with privacy laws?	Yes
Is data compliant with data subject requests for data correction or removal, if such a request was made?	Not Applicable
Applicable NVIDIA Privacy Policy	https://www.nvidia.com/en-us/about-nvidia/privacy-policy/