Mamba-Ssm - Loader for Mamba State Space models #5228
IggoOnCode
wants to merge
25 commits into
oobabooga:dev
from
IggoOnCode:mamba-ssm
Basic Mamba-Ssm support taken from trap20:mamba-ssm and adapted to cu…
6a960e42
cleanup
bdac8722
Updated README for Mamba-Ssm
94f278fc
cleanup after running pyflakes
050c4b2f
coding style fixes after running pycodestyle
2917ad04
wip for transfer to train
a15a70ed
it trains!
a67abbdd
better loading as saving, including config
d58cbe48
turns out it can be integrated
983f15c6
Updated UI for SSM training. Cleanup. Refactoring of variable names.
f79b69f9
undid unnecessary changes
c0076c71
added the packaging module to requirements.txt because the mamba-ssm …
dbbdde45
Refactoring and cleanup.
f9286936
Merge branch 'dev' into mamba-ssm
f9f994fc
Not breaking things on platforms where mamba_ssm does not work.
5a62b2c9
Added comments for module installation order.
96bc3f7c
Checking for Linux instead of not windows in installer.
57334628
Added mamba requirements to Google Colab.
c174d2c5
Implemented differential config export. Fixed bug in model reloading.…
4520088e
Fixed "libcuda.so not found" triton error on Google Colab. Still does…
8fefb8d3
updated training ui to differentiate between LoRA settings and SSM se…
17dc42ae
[x] I have remembered the Contributing guidelines
58e9dcd8
generalizing mamba training code to full fine-tune training for all m…
728fd234
fixed first token check for RWKV
c01a1c62
full training for mamab and llama. lora training for llama
15263aab
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub