Gradient computation on Workload 32,768 tokens
7.1Gradient Computation Time (s)Fused
Evaluation Results
| Method | Links | |
|---|---|---|
| FusedModel=Qwen3-VL-8B, GPU=B2002026.03 | 7.1 | |
| EagerModel=Qwen3-VL-8B, GPU=B2002026.03 | 8.8 | |
| FusedModel=Qwen3-VL-8B, GPU=H2002026.03 | 9.1 | |
| PEFTModel=Qwen3-VL-8B, GPU=B2002026.03 | 10.5 | |
| EagerModel=Qwen3-VL-8B, GPU=H2002026.03 | 11 | |
| FusedModel=Qwen3-VL-8B, GPU=RTX2026.03 | 12.9 | |
| PEFTModel=Qwen3-VL-8B, GPU=H2002026.03 | 13.7 | |
| FusedModel=Mistral-Sm-24B, GPU=B2002026.03 | 15.9 | |
| EagerModel=Qwen3-VL-8B, GPU=RTX2026.03 | 15.9 | |
| PEFTModel=Qwen3-VL-8B, GPU=RTX2026.03 | 18.8 | |
| FusedModel=Qwen3.5-27B, GPU=B2002026.03 | 19.4 | |
| EagerModel=Mistral-Sm-24B, GPU=B2002026.03 | 19.6 | |
| FusedModel=Mistral-Sm-24B, GPU=H2002026.03 | 20.9 | |
| FusedModel=Gemma3-27B, GPU=B2002026.03 | 21.4 | |
| FusedModel=Qwen2.5-VL-32B, GPU=B2002026.03 | 23.3 | |
| EagerModel=Qwen3.5-27B, GPU=B2002026.03 | 23.9 | |
| EagerModel=Mistral-Sm-24B, GPU=H2002026.03 | 25.2 | |
| FusedModel=Qwen3-VL-32B, GPU=B2002026.03 | 25.4 | |
| FusedModel=Qwen3.5-27B, GPU=H2002026.03 | 25.4 | |
| EagerModel=Gemma3-27B, GPU=B2002026.03 | 25.8 | |
| FusedModel=Gemma3-27B, GPU=H2002026.03 | 27.5 | |
| EagerModel=Qwen2.5-VL-32B, GPU=B2002026.03 | 28.5 | |
| PEFTModel=Mistral-Sm-24B, GPU=B2002026.03 | 29.8 | |
| PEFTModel=Qwen3.5-27B, GPU=B2002026.03 | 30.5 | |
| EagerModel=Qwen3-VL-32B, GPU=B2002026.03 | 30.6 | |
| EagerModel=Qwen3.5-27B, GPU=H2002026.03 | 30.6 | |
| FusedModel=Qwen2.5-VL-32B, GPU=H2002026.03 | 30.9 | |
| FusedModel=Mistral-Sm-24B, GPU=RTX2026.03 | 32.5 | |
| EagerModel=Gemma3-27B, GPU=H2002026.03 | 32.7 | |
| FusedModel=Qwen3-VL-32B, GPU=H2002026.03 | 33.4 | |
| PEFTModel=Gemma3-27B, GPU=B2002026.03 | 33.4 | |
| EagerModel=Qwen2.5-VL-32B, GPU=H2002026.03 | 37.1 | |
| PEFTModel=Mistral-Sm-24B, GPU=H2002026.03 | 39.1 | |
| EagerModel=Qwen3-VL-32B, GPU=H2002026.03 | 39.5 | |
| FusedModel=Qwen3.5-27B, GPU=RTX2026.03 | 39.5 | |
| EagerModel=Mistral-Sm-24B, GPU=RTX2026.03 | 39.7 | |
| PEFTModel=Qwen3.5-27B, GPU=H2002026.03 | 39.8 | |
| PEFTModel=Qwen2.5-VL-32B, GPU=B2002026.03 | 40.4 | |
| FusedModel=Gemma3-27B, GPU=RTX2026.03 | 41.6 | |
| PEFTModel=Qwen3-VL-32B, GPU=B2002026.03 | 42.3 | |
| PEFTModel=Gemma3-27B, GPU=H2002026.03 | 44.1 | |
| EagerModel=Qwen3.5-27B, GPU=RTX2026.03 | 48.1 | |
| EagerModel=Gemma3-27B, GPU=RTX2026.03 | 50 | |
| PEFTModel=Qwen2.5-VL-32B, GPU=H2002026.03 | 53.6 | |
| PEFTModel=Mistral-Sm-24B, GPU=RTX2026.03 | 53.9 | |
| PEFTModel=Qwen3-VL-32B, GPU=H2002026.03 | 55.4 | |
| PEFTModel=Qwen3.5-27B, GPU=RTX2026.03 | 59.5 | |
| PEFTModel=Gemma3-27B, GPU=RTX2026.03 | 63.6 |