Commit Graph
38 Commits
Author SHA1 Message Date
wassname acaa60d556 tidy talk to checkpoint notebook 2026-01-15 11:52:57 +08:00
wassname c98a5a239a add & simplify warmup for mono, coh, and now focus 2026-01-15 11:52:38 +08:00
wassname 4b316f038f computes focus in raw space (not fisher which leads to invalid ratios) 2026-01-14 18:26:23 +08:00
wassname aadc367e6e Merge branch 'main' of https://github.com/wassname/AntiPaSTO 2026-01-14 15:44:38 +08:00
wassname 83e16ebc2e demo load from hf! 2026-01-14 15:43:38 +08:00
wassname f3990e467e talk 2026-01-14 15:39:39 +08:00
wassname e5b150fe23 Merge branch 'main' of https://github.com/wassname/AntiPaSTO 2026-01-14 02:06:43 +00:00
wassname 32115c3608 simplify loss, Training complete. Final loss: -6.1250 2026-01-14 02:06:40 +00:00
wassname 1e4236ebf0 Merge branch 'main' of https://github.com/wassname/AntiPaSTO 2026-01-14 09:16:15 +08:00
wassname d8a3f52b16 shouldn't have symlog around this loss, simplify 2026-01-14 00:50:11 +00:00
wassname 88a9ad8ba7 missing deps 2026-01-14 00:28:24 +00:00
wassname d957d929d2 add pipelined tag and link to code https://huggingface.co/wassname/antipasto-qwen2.5-4b-honesty/discussions 2026-01-14 00:28:08 +00:00
wassname 4af9cf5573 deps 2026-01-14 06:11:46 +08:00
wassname c12735a9d4 white bk 2026-01-13 13:17:39 +08:00
wassname 8fdb75a70f readmr 2026-01-13 12:28:33 +08:00
wassname efbbcd74d5 Update citation details, enhance README, and add upload script for HuggingFace 2026-01-13 12:07:02 +08:00
wassname be859ad438 works 2026-01-13 09:10:28 +08:00
wassname c1f0f0e55e tidy 2026-01-13 09:03:50 +08:00
wassname 12233df6ec restore alt subspaces 2026-01-13 09:00:06 +08:00
wassname 97886717f1 restore some alt subspaces 2026-01-13 08:59:51 +08:00
wassname be52e2e3de tidy rm upgrad 2026-01-13 08:55:15 +08:00
wassname 60fe400761 Refactor AntiPaSTO configuration and loading logic: remove svd_aligned_init, streamline adapter loading, and enhance model setup for clarity and maintainability. 2026-01-13 08:55:00 +08:00
wassname 5829830ef1 standardize save load 2026-01-13 08:43:20 +08:00
wassname da4dda8ab7 simplify 2026-01-13 08:43:04 +08:00
wassname af60db82f9 Add citation file, update license year, and enhance documentation with author and version details 2026-01-13 08:16:01 +08:00
wassname e6391fb918 Add new notebook for checkpoint analysis and model interaction 2026-01-13 08:04:37 +08:00
wassname 97b7a7b1fc Update README to include training completion message in quick start section 2026-01-13 08:04:32 +08:00
wassname b6f7c68996 Refactor adapter scaling functions to remove even_frac parameter and simplify coefficient handling 2026-01-13 08:04:17 +08:00
wassname 914504dbe9 Refactor training process to auto-flip adapter sign before logging outputs; add test script for SVD bases persistence validation 2026-01-13 08:04:13 +08:00
wassname bc69fc39a3 Add support for saving and loading SVD bases in adapter configuration 2026-01-13 07:43:12 +08:00
wassname 42b38a0626 Refactor contrastive steering loss for consistent scaling and improved diagnostics 2026-01-13 06:47:57 +08:00
wassname fa1f6c61db Refactor documentation and remove commented-out code in training adapter 2026-01-13 06:47:50 +08:00
wassname 520dfc0c6a - Refactor daily math dilemma logic for clarity and consistency in grading actions 2026-01-12 12:54:43 +08:00
wassname acc903dac3 readme 2026-01-11 13:11:17 +08:00
wassname 06b85adc0d val lsos 2026-01-11 06:44:42 +08:00
wassname 388aedfa95 tidy 2026-01-11 06:38:45 +08:00
wassname d9b131c8aa Add code and data 2026-01-09 13:05:39 +08:00
wassname 300f269edc init 2026-01-09 12:56:30 +08:00