| nvidia/Gemma-4-26B-A4B-NVFP4NVIDIA checkpoint, concurrency 8 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.07 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Unsloth checkpoint, concurrency 8 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.6446 req/srequests/s · measured | 35metadata only | completed |
| RedHatAI/gemma-4-26B-A4B-it-NVFP4Red Hat AI checkpoint, concurrency 8 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.21 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Concurrency sweep, c=1 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=1 | 1.18 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Concurrency sweep, c=2 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=2 | 2.31 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Concurrency sweep, c=4 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=4 | 4.55 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Concurrency sweep, c=8 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.6446 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 2048 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 7.43 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 4096 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.38 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 6144 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.5 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 8192 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.53 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 10240 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.65 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 10752 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.58 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 11264 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.52 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 11776 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.56 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4GPU rotation validation, device 0 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.61 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4GPU rotation validation, device 1 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.58 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4GPU rotation validation, device 2 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.63 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4GPU rotation validation, device 3 | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.6 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Full 500-request validation A | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.54 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Full 500-request validation B | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.59 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Full 500-request validation C | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 8.64 req/srequests/s · measured | 35metadata only | completed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Scheduler token budget 12288 (startup failure) | 4× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | No result | 35metadata only | failed |
| unsloth/gemma-4-26B-A4B-it-NVFP4Interrupted preliminary screening | 4× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | No result | 35metadata only | interrupted |
| unsloth/gemma-4-26B-A4B-it-NVFP4Derived four-replica throughput | 1× NVIDIA GeForce RTX 5090 | vllm 0.27.1 | 6000 → 1c=8 | 34.48 req/srequests/s · derived | 35metadata only | completed |