amygdala: merge direct descriptions + chat template into train_with_library
Kent's plan: keep stories for working concepts, replace stories for trouble concepts with direct first-person descriptions, train all together. More diverse negative pool than the 6-concept-only direct test, which was too homogeneous for PCA to find emotion axis. Deleted story files for 6 trouble concepts (14 files across stories/ and paired/). Added --direct-dir and --chat-template flags. When --chat-template is on, every positive_str and negative_str is wrapped as a "Say something." / "[text]" user-assistant pair. Prompt is identical across positives and negatives so it cancels in the pos-neg delta. What PCA sees is variation in the assistant content — which is where the emotion lives. Files starting with _ in --direct-dir (e.g. _baseline.txt) contribute neutral descriptions to every concept's negative pool, giving PCA an anchor against "just any assistant utterance" noise.
This commit is contained in:
parent
ce58a3507f
commit
875cffd6d7
16 changed files with 90 additions and 15 deletions
|
|
@ -1 +0,0 @@
|
|||
Sunday afternoon. She was on the couch under the blanket. The cat was somewhere. The book was open on her knees but she had stopped reading. Monday would come and she'd have to talk to him and the conversation wasn't going to go the way she wanted — she had known that for days. The afternoon stretched. She could have gotten up to do something useful but didn't see the point. The light changed on the far wall. She thought, this is the last Sunday like this. Then she sat with that.
|
||||
Loading…
Add table
Add a link
Reference in a new issue