WIP: Structured training data and output with Langchain + RELLM #11

james · 2025-01-17T00:36:21Z

james commented

2025-01-17 00:36:21 +00:00

Hopefully this will be the update that makes Miku smart.

Note to self: accidentally used 'vinso' instead of 'vinso1445' in data prep script, so trained model may not be aware of Vin's real name, but we can still use the corrected mapping at runtime for now until it gets retrained.

Hopefully this will be the update that makes Miku smart. ~~Note to self: accidentally used 'vinso' instead of 'vinso1445' in data prep script, so trained model may not be aware of Vin's real name, but we can still use the corrected mapping at runtime for now until it gets retrained.~~

james added 3 commits 2025-01-17 00:36:21 +00:00

Langchain version aa8057c417

Make prompts consistent across files f755a03eb2

New training data prep scripts 8159b11f4f

james added 1 commit 2025-01-18 09:58:58 +00:00

Langchain server fb78222a3e

james added 2 commits 2025-01-19 02:46:59 +00:00

Fix assistant messages not being authored by Miku db7fb0f001

Fix output parsing and add back sys_prompt option 627787d49b

james commented

2025-01-20 02:02:00 +00:00

Closes #9.