affine-gpt-oss-120b-light

Model Overview

Description:

The NVIDIA gpt-oss-120b Eagle model is the Eagle head of the OpenAI’s gpt-oss-120b model, which is an auto-regressive language model that uses a mixture-of-experts (MoE) architecture with 32 billion activated parameters and 1 trillion total parameters. For more information, please check here. The NVIDIA gpt-oss-120b Eagle3 model incorporates Eagle speculative decoding with TensorRT Model Optimizer.

This model is ready for commercial/non-commercial use. <br>

License/Terms of Use:

nvidia-open-model-license

Deployment Geography:

Global <br>

Use Case: <br>

Developers designing AI Agent systems, chatbots, RAG systems, and other AI-powered applications. Also suitable for typical instruction-following tasks. <br>

Release Date: <br>

Huggingface: Aug 20th, 2025 via [https://huggingface.co/nvidia/gpt-oss-120b-Eagle3] <br>

Model Architecture:

Architecture Type: Transformers <br> Network Architecture: gpt-oss-120b <br>

##Computational Load **Cumulative Compute: 4.8x10^20 **Estimated Energy and Emissions for Model Training: *Total kWh = 2500 *Total Emissions (tCO2e) = 0.8075

Input:

Input Type(s): Text <br> Input Format(s): String <br> Input Parameters: One Dimensional (1D): Sequences <br>

Output:

Output Type(s): Text <br> Output Format: String <br> Output Parameters: One-Dimensional (1D): Sequences <br>

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. <br>

Software Integration:

Supported Runtime Engine(s): <br>

• TensorRT-LLM <br>

Supported Hardware Microarchitecture Compatibility: <br>

• NVIDIA Blackwell <br>

Preferred Operating System(s): <br>

• Linux <br>

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 is quantized with nvidia-modelopt v0.35.0 <br>

Training and Evaluation Datasets:

** The total size (in number of data points) 503.3K <br> ** Total number of datasets 2<br>
** Dataset partition: Training 100%<br>

Training Dataset:

Link: ultrachat_200k and Magpie-Llama-3.1-Pro-300K-Filtered, only prompts from the datasets were used for data synthesis, (the original responses from GPT were not used) for data synthesis, which is then used to train the Eagle modules. Click the links above for more information regarding the dataset. <br>

** Data Modality [Text]

** Data Collection Method by dataset <br>

• Hybrid: Synthetic, Human, Automated<br>

** Labeling Method by dataset <br>

• Hybrid: Synthetic, Human, Automated<br> Properties: 500K samples, majority synthetic, others sourced from commercially-friendly datasets. <br>

Evaluation Dataset: <br>

Link: MTBench, for more details, see here <br>

** Data Collection Method by dataset <br>

• Hybrid: Human, Synthetic<br>

** Labeling Method by dataset <br>

• Hybrid: Human, Synthetic<br>

Properties: 3,300 multi-turn dialogue sequences, each annotated with expert preference votes.<br>

Inference:

Engine: TensorRT-LLM <br> Test Hardware: B200 <br>

Eagle Speculative Decoding

Synthesized data was obtained from OpenAI's gpt-oss-120b model, which is then used to finetune the Eagle modules. This model is ready for inference with TensorRT-LLM in Eagle speculative decoding mode. Eagle modules are used to predict candidate tokens beyond the next token. In the generation step, each forward Eagle module generates a distribution of tokens beyond the previous. Then, a tree-based attention mechanism samples some candidate sequences for the original model to validate. The longest accepted candidate sequence is selected so that more than 1 token is returned in the generation step. The number of tokens generated in each step is called acceptance rate.

Usage

To serve the quantized checkpoint with TensorRT-LLM, follow the sample commands below with the TensorRT-LLM GitHub repo:

trtllm-serve <gpt-oss-120b checkpoint> --host 0.0.0.0 --port 8000 --backend pytorch --max_batch_size 32 --max_num_tokens 8192 --max_seq_len 8192 --tp_size 8 --extra_llm_api_options extra-llm-api-config.yml

extra-llm-api-config.yml is like this

enable_attention_dp: false
pytorch_backend_config:
  enable_overlap_scheduler: false
  use_cuda_graph: true
  cuda_graph_max_batch_size: 1
  autotuner_enabled: false

speculative_config:
    decoding_type: Eagle
    max_draft_len: 3
    pytorch_eagle_weights_path: <eagle3 checkpoint>

kv_cache_config:
    enable_block_reuse: false

Evaluation

The Eagle acceptance rate benchmark results (MT-Bench) with draft length 3 are presented in the table below for medium reasoning:

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. When downloaded or used in accordance with our terms of service, 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.

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

Please report security vulnerabilities or NVIDIA AI Concerns here. SUBCARDS:

Explainability

Bias

Safety & Security

Privacy