Skip to content

gemma4/gguf: accept a scalar attention.head_count_kv #64

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
Rhycecomley wants to merge 1 commit into
base: main
Choose a base branch
from
Open

gemma4/gguf: accept a scalar attention.head_count_kv #64

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 6 additions & 1 deletion python/freetoken/models/gemma4/gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,7 +60,12 @@ def g(key: str):
num_layers = int(g("block_count"))
hidden = int(g("embedding_length"))
num_qo_heads = int(g("attention.head_count"))
kv_per_layer = g("attention.head_count_kv") # per-layer list
# Per-layer list, or a single value when every layer shares it: llama.cpp collapses
# a uniform array to a scalar, which the official gemma-4 E2B QAT q4_0 GGUF hits
# (head_count_kv == 1 for all 35 blocks). Broadcast so the indexing below holds.
kv_per_layer = g("attention.head_count_kv")
if isinstance(kv_per_layer, (int, float)):
kv_per_layer = [int(kv_per_layer)] * num_layers
# True -> sliding-window (SWA) layer, False -> full attention.
swa_pattern = [bool(x) for x in g("attention.sliding_window_pattern")]
assert len(swa_pattern) == num_layers, "sliding_window_pattern length != block_count"
Expand Down