Quantized Qwen3-8B from the sm120-quant-bench series: W4A16 (GPTQ, AWQ) and NVFP4, with per-capability damage tables and kernel receipts.
Rodney Woods
Rodder5
AI & ML interests
Quantization and inference efficiency on consumer GPUs, evaluation methodology (capability probes over aggregates), speech AI for low-resource languages, on-device ML.
Recent Activity
updated a collection 2 days ago
sm120-quant-bench checkpoints updated a collection 2 days ago
sm120-quant-bench checkpoints updated a collection 2 days ago
sm120-quant-bench checkpointsOrganizations
None yet