Skip to content

Granite4 2b & 3b support #496

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.

Sign up for GitHub

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

Open
wants to merge 12 commits into
base: main
Choose a base branch
from
Open

Granite4 2b & 3b support #496

Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
84cfeeb
consume fms dev branch
yannicks1 Oct 1, 2025
3c2e870
adapt kv cache size
yannicks1 Oct 1, 2025
662c656
Merge branch 'main' into ysc-support-granite-2b-3b
yannicks1 Oct 1, 2025
039c59a
fix fmt
yannicks1 Oct 1, 2025
e46ff48
load granite model types correctly too
yannicks1 Oct 1, 2025
f03d61b
remove print
yannicks1 Oct 1, 2025
ecd957f
exclude llama from head dim expansion hack
yannicks1 Oct 2, 2025
8f40bf2
grab new head_dim arg directly from fms
yannicks1 Oct 7, 2025
29dc43e
support llama again
yannicks1 Oct 7, 2025
083c55b
remove fms feature branch as merged
yannicks1 Oct 15, 2025
f792c39
Merge branch 'main' into ysc-support-granite-2b-3b
yannicks1 Oct 15, 2025
049956d
establish backward compatibility, simplify
yannicks1 Oct 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 4 additions & 3 deletions vllm_spyre/model_executor/model_loader/spyre.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -339,10 +339,11 @@ def __init__(
'num_kv_heads'] = self.model_config.get_num_kv_heads(
self.parallel_config)

if self.config.model_type in {'llama', 'granite'}:
if self.config.model_type in {'llama', 'granite', 'granitemoehybrid'}:
self.kv_cache_specs['num_layers'] = self.config.num_hidden_layers
self.kv_cache_specs['head_dim'] = self.config.hidden_size // \
self.config.num_attention_heads
self.kv_cache_specs['head_dim'] = getattr(
self.model.config, "head_dim",
self.config.hidden_size // self.config.num_attention_heads)
elif self.config.model_type == 'gpt_bigcode':
self.kv_cache_specs['num_layers'] = self.config.n_layer
self.kv_cache_specs[
Expand Down