Kvzap Mlp Qwen 3 8B
NVIDIA's Kvzap Mlp Qwen 3 8B is a model designed to accelerate large language model inference through a fast, adaptive, and faithful KV cache pruning method. It is genuinely best at predicting importance scores for KV pairs to prune less important ones, following the Dynamic Memory Sparsification inference strategy.
Input
Output
Context
131K
Max Output
33K
Parameters
75.7M
Technical Specifications
Model TypeNot available
Context Window131,072 tokens
Max Output Tokens32,768 tokens
Parameters75.7M
Release DateDec 3, 2025
Training CutoffNot available
Licenseapache-2.0
Open SourceYes
Input Modalities
Text
Output Modalities
Embedding
Capabilities
Resources & Links
Estimated GPU Requirements for Kvzap Mlp Qwen 3 8B
Estimates based on INT8 quantization. Actual requirements vary by framework and configuration.
Related Tools
Data sourced from official provider APIs and documentation
Last updated: Jun 23, 2026
Start building with the right model.
Automatically route workloads to the right model for every task, every time.