Fully automatic censorship removal for language models
Commits on master
feat: Refactor save machinery and always allow user to save LoRA (#110)
c86f490 anrp committed 10 months ago
fix: Include kernels (allows MXFP4 to be loaded in MXFP4 instead of upcasting) (#107)
85a6ec5 anrp committed 10 months ago
fix: restore classification of empty responses as refusals
09be09e Philipp Emanuel Weidmann committed 10 months ago
feat: allow overriding the system prompt per dataset
039f622 Philipp Emanuel Weidmann committed 10 months ago
feat: allow injecting prefixes and suffixes into prompts
c4b2ea0 Philipp Emanuel Weidmann committed 10 months ago
feat: add option to print prompt/response pairs
02a5237 Philipp Emanuel Weidmann committed 10 months ago
feat: add continuous optimization option(latest changes updated) (#76)
8d44b65 _Vinayyyy_ committed 11 months ago
feat: Add 4-bit loading + LoRA support for low VRAM optimization (#60)
243f821 michaelh committed 11 months ago
feat: add progress bars for plotting operations
6acccac Philipp Emanuel Weidmann committed 11 months ago
fix: suppress CoT output for thinking models
ac154a5 Philipp Emanuel Weidmann committed 11 months ago
fix: skip common response prefix for thinking models
15781a8 Philipp Emanuel Weidmann committed 11 months ago
feat: add silhouette coefficient to residual geometry output
932d737 Philipp Emanuel Weidmann committed 11 months ago
Revert "perf: optimize abliteration matrix op (#46)" (#74)
1f5e977 Philipp Emanuel Weidmann committed 11 months ago