-
Notifications
You must be signed in to change notification settings - Fork 146
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
aqlm #233
Conversation
@tgaddair |
+1 for the AQLM requirement |
@@ -234,7 +234,7 @@ def get_model( | |||
trust_remote_code=trust_remote_code, | |||
) | |||
|
|||
if model_type == "llama": | |||
if model_type in ["llama", "llama_aqlm"]: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
llama_aqlm
and similar custom model types are deprecated. We moved AQLM from custom code to transformers integration with transformers
4.38.0
. Correct me if I'm wrong and it serves different purpose here.
@@ -38,13 +38,14 @@ boto3 = "^1.28.34" | |||
urllib3 = "<=1.26.18" | |||
hqq = { version = "^0.1.2", optional = true } | |||
stanford-stk = { version = "^0.7.0", markers = "sys_platform == 'linux'" } | |||
aqlm = { version = "^1.0.0"} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Proper backprop was added in aqlm==1.1.0
and bf16 support was added in aqlm==1.1.2
.
What does this PR do?
Fixes # (issue)
Before submitting
to it if that's the case.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.