Qwen-3-4B

The 'Qwen3-4B' dataset/benchmark is used to evaluate the performance of LLM-guided CUDA inference compilers by measuring speedup in end-to-end autoregressive generation tasks.

Papers using Qwen-3-4B (3)

AgentCompile: An LLM-Guided Compiler for Direct CUDA Inference2026From Human Guidance to Autonomy: Agent Skill System for End-to-End LLM Deployment on Spatial NPUs2026Real-Time Progress Prediction in Reasoning Language Models2025