ISTA-DASLab
Quantized Model Leaderboard Faces Reproducibility Dispute
ISTA-DASLab’s quantized-model benchmark results were challenged after Unsloth and Basalt reported better results using equally specified models.
4 articles tagged with this topic
ISTA-DASLab’s quantized-model benchmark results were challenged after Unsloth and Basalt reported better results using equally specified models.
Qwen, Unsloth, and ISTA-DASLab are making local LLMs smaller and easier to quantize, with frontier-like systems running on consumer hardware.
Strata inference engine pushes Qwen3-8B on consumer laptops from 23 t/s to 51 t/s — over 2x llama.cpp. Is local AI hitting its productivity tipping po
Austria's ISTA-DASLab squeezed a Qwen 27B to 10GB, matching original quality. Local AI advances — capable, private deployment without cloud uploads.