List view

Pointwise scheduler changes

MoE Inference nvFP4Get Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inferenceGet Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inference

Status: Closed (completed).

Enhancement

#5419 In NVIDIA/Fuser;

· by nvMelissa was closed on Dec 2, 2025

Expose quantization through Python API to test

MoE Inference nvFP4Get Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inferenceGet Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inference

Status: Closed (completed).

Enhancement

#5424 In NVIDIA/Fuser;

· by nvMelissa was closed on Dec 2, 2025

Lower Quantization Operations

MoE Inference nvFP4Get Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inferenceGet Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inference

Status: Closed (completed).

Enhancement

#5418 In NVIDIA/Fuser;

· by nvMelissa was closed on Nov 6, 2025

Use the output of nvfuser quantization in scaled mm.

MoE Inference nvFP4Get Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inferenceGet Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inference

Status: Closed (completed).

#5617 In NVIDIA/Fuser;

· by protonu was closed on Dec 5, 2025

Linear Quantization

MoE Inference nvFP4Get Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inferenceGet Llama 4, GPT OSS, Deepseek R1, and Qwen3-Next running with performant 4-bit inference

Status: Closed (completed).

Enhancement

#5417 In NVIDIA/Fuser;

· by nvMelissa was closed on Dec 5, 2025