Comparing ISTA-DASLab's DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF 2.25bpw against the official API
I've been running ISTA-DASLab's 2.25bpw quant of DSv4.1-1017 for a few days, and it's holding up well. But the fact that it's only 2.25bpw average did make me wonder how good the quant actually is against the full model. Local setup: Project-maya v1.1.27. ISTA-DASLab's 2.25bpw quant.