Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Training Throughput on Qwen3 32B (train)

545.29Training Throughput (128K Seq Len)

Ulysses

110.2996223.2298336.16449.0902Feb 24, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.02
545.29370.7217.04117.0259.98---
2026.02
483.29339.56204.46113.2659.5640.4229.97-
2026.02
418.39308.88194.44110.2758.45---
2026.02
286.4217.85151.9195.8855.4138.8627.66-
2026.02
127.03112.291.39-----