QUEST-30B-RL
QUEST-30B-RL
credit: Quest / qwen3moe
B70 input tok/s76.3
B70 decode tok/s335.5
B70 per-agent tok/s11.2
B70 live fleet32 · Intel Arc Pro B70
Q5_K_M qwen3moe · llama.cpp SYCL on Intel Arc Pro B70 · context-calibrated team of 6 · 15-min timed production loop.
model dossier
Model sheet
Playable artifacts and benchmark context for this model page.
Road Hopper
0 builds
Robot-Filled Maze Shooter
1 builds
Ribbit Rush
0 builds
- Builds
- 1 recorded outputs
- Method
- team 6 / 15-min build
- Runtime
- llama.cpp SYCL on Intel Arc Pro B70
- Quant
- Q5_K_M qwen3moe
- Bench
- B70 live: 32 agents / 335.5 decode tok/sec
benchmark sheet
B70 live throughput profile
B70 input tok/s76.3
B70 decode tok/s335.5
B70 per-agent tok/s11.2
B70 live fleet32 · Intel Arc Pro B70
aggregate decode tok/secper-agent tok/sechighlight = selected live fleet
tool · 1 agents66.5 dec84.6/agent · 154.2 in
tool · 8 agents164.7 dec22.0/agent · 156.8 in
tool · 16 agents223.0 dec14.8/agent · 121.0 in
tool · 32 agents335.5 dec11.2/agent · 76.3 in
structured · 32 agents211.2 dec7.5/agent · 101.0 in
novel · 32 agents148.2 dec5.4/agent · 126.2 in
Highlighted row is the selected 32-agent tool-prompt live fleet on Intel Arc Pro B70. Structured and novel rows are LocalMaxxing prompt classes.
Raw B70 live table
| class | agents | prefill_tps | agg_decode_tps | per_agent_tps | batch_wall_s | status |
|---|---|---|---|---|---|---|
| tool | 1 | 154.2 | 66.5 | 84.6 | 3.85 | OK |
| tool | 8 | 156.8 | 164.7 | 22.0 | 12.44 | OK |
| tool | 16 | 121.0 | 223.0 | 14.8 | 18.37 | OK |
| tool | 32 | 76.3 | 335.5 | 11.2 | 24.42 | OK |
| structured | 32 | 101.0 | 211.2 | 7.5 | 38.79 | OK |
| novel | 32 | 126.2 | 148.2 | 5.4 | 55.28 | OK |
Generated game outputs
Versions are listed first for selection. Embedded outputs remain below for direct review.
Robot-Filled Maze Shooter 1 versions
maze_quest_b70_live · Robot-Filled Maze Shooter · team buildfullscreen ↗ · compareRobot-Filled Maze Shooter team build Quest Turbo
maze_quest_b70_livefullscreen ↗