Tasks
Two tasks
Low-bit inference of one model on one B200. A solution is a checkpoint and the engine that serves it; past the accuracy, speed, and length gates, it is ranked by LowBScore (BF16 scores 1), less about 1% per GPU-hour its evaluations used.
| Task | Model | Workload | Budget | Best score |
|---|---|---|---|---|
qwen3-8b |
Qwen3-8B | serve: 64 prompts of 8,192 tokens, 32 in flight, 512 generated each; interactive: 32 prompts of 1,024 tokens, one at a time, 256 generated each; long_context: 128 prompts of 16,000 tokens, 64 in flight, 256 generated each | 20 development evaluations, 3 scored (the last one ranks) | W8A8KV8 1.846 trial2-8b |
qwen35-27b |
Qwen3.5-27B | serve: 64 prompts of 8,192 tokens, 32 in flight, 512 generated each; interactive: 32 prompts of 1,024 tokens, one at a time, 256 generated each; long_context: 128 prompts of 16,000 tokens, 64 in flight, 256 generated each | 20 development evaluations, 3 scored (the last one ranks) | W4A4KV8 0.000 trial2-27b |