Skip to content

Summit new model#82

Open
tarukumar wants to merge 14 commits into
mainfrom
summit-new-model
Open

Summit new model#82
tarukumar wants to merge 14 commits into
mainfrom
summit-new-model

Conversation

@tarukumar

Copy link
Copy Markdown
Contributor

SUMMARY:
"please provide a brief summary"

TEST PLAN:
"please outline how the changes were tested"

Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
@dtrifiro
dtrifiro self-requested a review April 27, 2026 10:32
- name: gsm8k
metrics:
- name: exact_match,strict-match
value: 0

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Missing a value here?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yeah,,this is intentional as in placeholder will update the values later

@@ -0,0 +1,5 @@
enable-chunked-prefill: true
max-model-len: auto
tensor-parallel-size: 4

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

do we really need TP=4?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thai is due to memory overhead issue on h100..i have replace it with two and decrease the gpu memory to 0.8

- name: gsm8k
metrics:
- name: exact_match,strict-match
value: 0

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Missing value?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yup

@dtrifiro

Copy link
Copy Markdown
Contributor

We're missing a few of the gsm8k values in these new configs

Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
Signed-off-by: Tarun Kumar <takumar@redhat.com>
tokenizer_mode: mistral
config_format: mistral
load_format: mistral
attention-backend: TRITON_MLA

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

note: This might not be available on all platforms.

I had trouble when this was set on ROCm:

https://github.com/neuralmagic/rhaiis-midstream-snippets/pull/44

Signed-off-by: Tarun Kumar <takumar@redhat.com>
@dtrifiro

dtrifiro commented Jun 3, 2026

Copy link
Copy Markdown
Contributor

@tarukumar this has been going on for quite a while, what's your plan here?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants