huginnfork/Qwen3.8-Flash-Next-mixed-NVFP4-FP8-PLE-NVFP4
68B • Updated • 50 • 1
Qwen3.8-Flash-Next: quantised PLE matrix
Two primitive-ai Qwen3.8-Flash-Next checkpoints x three quantised PLE tables. BF16 n-gram table dropped; one repo id per configuration.
Note 102.6 GiB. Smallest table on the faster checkpoint - the pick for a small-RAM host.
huginnfork/Qwen3.8-Flash-Next-mixed-NVFP4-FP8-PLE-INT4
68B • Updated • 50Note 105.6 GiB. The only table primitive-ai validated end to end under a memory cap.
huginnfork/Qwen3.8-Flash-Next-mixed-NVFP4-FP8-PLE-FP8
68B • Updated • 28Note 124.7 GiB. Largest table; its edge is throughput at high concurrency.
huginnfork/Qwen3.8-Flash-Next-NVFP4-PLE-NVFP4
68B • Updated • 31Note 105.1 GiB.
huginnfork/Qwen3.8-Flash-Next-NVFP4-PLE-INT4
68B • Updated • 42Note 108.1 GiB.
huginnfork/Qwen3.8-Flash-Next-NVFP4-PLE-FP8
68B • Updated • 30Note 127.2 GiB.