Kishin Nishik's picture

5

Kishin Nishik

Malum0x

·

AI & ML interests

None yet

Recent Activity

reacted to s3nh's post with ❤️ 3 days ago

Existing methods — GPTQ, AWQ, llama.cpp's k-quants — minimize empirical loss heuristically. None of them prove they are optimal in any information-theoretic sense. ICRB-Q builds a quantization scheme that is provably optimal via the Cramér-Rao lower bound (CRB): no unbiased estimator of a weight can have lower variance than [F(θ)]⁻¹, where F is the Fisher information matrix.

liked a model 29 days ago

Malum0x/mlp-surgery-restored-top30

updated a collection about 1 month ago

View all activity

Organizations

Malum0x 's datasets 1

Malum0x/openhermes2.5-Perplexity_filtered_top30

Viewer • Updated Mar 8 • 300k • 11 • 1