Swift efficient reasoning + HyperQwen serving: three checkpoints compared on RTX 3090 task time, token use, throughput and quality.
🔄 In a Training Loop
David Hauser
daavidhauser
AI & ML interests
ML 4 Science, LLMs and more.
Recent Activity
updated a collection 2 days ago
Swift for HyperQwen — RTX 3090 benchmarks updated a collection 2 days ago
Swift for HyperQwen — RTX 3090 benchmarks updated a model 2 days ago
daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwenOrganizations
None yet
models 5
daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen
Text Generation • 28B • Updated • 395 • 2
daavidhauser/Swift-1.0-Qwen3.8-27B-W4A16-HyperQwen
Text Generation • 28B • Updated • 380
daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4
Text Generation • 28B • Updated • 62
daavidhauser/chess-bot-3000-250m
0.2B • Updated • 30 • 4
daavidhauser/chess-bot-3000-100m
0.1B • Updated • 48
datasets 0
None public yet