James S james
0 Followers · 1 Following
Repositories
18
 Projects Packages Public Activity Starred Repositories
james pushed to langchain at james/FemScoreboard 2025-01-29 11:25:46 +00:00
8ef7a03895 with every commit, Miku grows stronger.
james pushed to langchain at james/FemScoreboard 2025-01-29 09:40:06 +00:00
21a2b1d4d0 Huggingface api mode, support for editing system prompts, config refactor
a361f110ec Add min_new_tokens flag
Compare 2 commits »
james commented on pull request james/MikuAI#11 2025-01-20 02:02:00 +00:00
WIP: Structured training data and output with Langchain + RELLM

Closes #9.

james pushed to langchain at james/FemScoreboard 2025-01-20 00:56:00 +00:00
411d458549 Add back repetition penalty
james created branch langchain in james/FemScoreboard 2025-01-19 04:31:56 +00:00
james pushed to langchain at james/FemScoreboard 2025-01-19 04:31:56 +00:00
8c3c68f384 Update for compatibility with Langchain server
james pushed to langchain at james/MikuAI 2025-01-19 02:46:58 +00:00
627787d49b Fix output parsing and add back sys_prompt option
db7fb0f001 Fix assistant messages not being authored by Miku
Compare 2 commits »
james pushed to langchain at james/MikuAI 2025-01-18 09:58:58 +00:00
fb78222a3e Langchain server
james created pull request james/MikuAI#11 2025-01-17 00:36:22 +00:00
WIP: Structured training data and output with Langchain + RELLM
james created branch langchain in james/MikuAI 2025-01-17 00:33:08 +00:00
james pushed to langchain at james/MikuAI 2025-01-17 00:33:08 +00:00
8159b11f4f New training data prep scripts
f755a03eb2 Make prompts consistent across files
aa8057c417 Langchain version
Compare 3 commits »
james created pull request james/MikuAI#10 2025-01-17 00:20:43 +00:00
WIP: Vision model
james created branch vision in james/MikuAI 2025-01-17 00:20:10 +00:00
james pushed to vision at james/MikuAI 2025-01-17 00:20:10 +00:00
247477edc8 Vision training playground
james commented on issue james/MikuAI#6 2025-01-03 09:23:23 +00:00
Multimodal Model

when a message is sent to miku with an image, we can just have it infer an image description and offer it as context in the langchain prompt; #9

james commented on issue james/MikuAI#2 2025-01-03 04:22:46 +00:00
New Training Methods

probably could reduce the amount of training data coming from our chat relative to other datasets, or training for fewer steps - might be some overfitting going on

james commented on issue james/MikuAI#7 2025-01-03 04:21:14 +00:00
Inclusion of Other Datasets

get the schizodev guy on youtube to let me see his cunnyposting dataset lol

james commented on issue james/MikuAI#6 2025-01-03 04:21:03 +00:00
Multimodal Model

figure out if second model entirely is needed for vision, or if it’s possible to train vision models on datasets without images

james opened issue james/MikuAI#9 2025-01-03 04:20:35 +00:00
Langchain Refactor
james commented on issue james/MikuAI#7 2025-01-03 04:18:51 +00:00
Inclusion of Other Datasets

hatsune miku's funny tweets https://youtu.be/i8GiGyP39vY?feature=shared