Skip to content

Update Gemma4 weight conversion script #45328

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

Merged
Cyrilvallez merged 4 commits into from
Apr 22, 2026
Merged

Update Gemma4 weight conversion script #45328

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
bf50b4a
fix: KV cache sharing
RyanMullins Apr 8, 2026
a09b070
fix: duplicate shared_kv_states in function params
RyanMullins Apr 9, 2026
7d863e5
fix: module style
RyanMullins Apr 9, 2026
2685bc7
revert: modeling changes
RyanMullins Apr 9, 2026
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
15 changes: 9 additions & 6 deletions src/transformers/models/gemma4/convert_gemma4_weights.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -706,13 +706,15 @@ def convert_transformer_weights(

converted_paths: list[str] = []
converted_weights: list[Any] = []
first_kv_shared_layer_idx = config.num_hidden_layers - getattr(config, "num_kv_shared_layers", 0)

# Handle new checkpoint format: transformer/layer_N/...
# TODO(philculliton):Direct handling for unstacked checkpoint type, needs to be merged to allow for unified tensor handling
if path.startswith(f"{_TRANSFORMER_PARAMETER}/layer_"):
# Extract layer number from path like "transformer/layer_0/attn/q_einsum"
layer_str = path.split("/")[1] # "layer_0"
layer_idx = int(layer_str.replace("layer_", "")) # 0
is_kv_shared_layer = layer_idx >= first_kv_shared_layer_idx > 0
base_path = f"layers.{layer_idx}"

# Determine head_dim from actual checkpoint weight dimensions
Expand All @@ -738,7 +740,7 @@ def convert_transformer_weights(
converted_weights.append(
matrix.transpose(2, 0, 1).reshape(config.hidden_size, config.num_attention_heads * head_dim)
)
elif path.endswith("attn/kv_einsum"):
elif path.endswith("attn/kv_einsum") and not is_kv_shared_layer:
converted_paths.extend(
[
f"{base_path}.self_attn.k_proj.weight",
Expand All @@ -753,7 +755,7 @@ def convert_transformer_weights(
v_proj_weights.reshape(kv_proj_shape).transpose(),
]
)
elif path.endswith("attn/k_einsum"):
elif path.endswith("attn/k_einsum") and not is_kv_shared_layer:
converted_paths.append(f"{base_path}.self_attn.k_proj.weight")
converted_weights.append(
matrix.transpose(1, 0, 2)
Expand All @@ -770,7 +772,7 @@ def convert_transformer_weights(
elif path.endswith("attn/query_norm"):
converted_paths.append(f"{base_path}.self_attn.q_norm.weight")
converted_weights.append(matrix)
elif path.endswith("attn/key_norm"):
elif path.endswith("attn/key_norm") and not is_kv_shared_layer:
converted_paths.append(f"{base_path}.self_attn.k_norm.weight")
converted_weights.append(matrix)
elif path.endswith("mlp/gating_einsum"):
Expand Down Expand Up @@ -816,6 +818,7 @@ def convert_transformer_weights(

for i, matrix in enumerate(weights):
layer_idx = _SLIDING_WINDOW_PATTERN * i + attention_type_index
is_kv_shared_layer = layer_idx >= first_kv_shared_layer_idx > 0
base_path = f"layers.{layer_idx}"
head_dim = (
config.global_head_dim
Expand All @@ -831,7 +834,7 @@ def convert_transformer_weights(
converted_weights.append(
matrix.transpose(2, 0, 1).reshape(config.hidden_size, config.num_attention_heads * head_dim)
)
elif path.endswith("attn/kv_einsum"):
elif path.endswith("attn/kv_einsum") and not is_kv_shared_layer:
converted_paths.extend(
[
f"{base_path}.self_attn.k_proj.weight",
Expand All @@ -846,7 +849,7 @@ def convert_transformer_weights(
v_proj_weights.reshape(kv_proj_shape).transpose(),
]
)
elif path.endswith("attn/k_einsum"):
elif path.endswith("attn/k_einsum") and not is_kv_shared_layer:
converted_paths.append(f"{base_path}.self_attn.k_proj.weight")
converted_weights.append(
matrix.transpose(1, 0, 2)
Expand All @@ -863,7 +866,7 @@ def convert_transformer_weights(
elif path.endswith("attn/query_norm"):
converted_paths.append(f"{base_path}.self_attn.q_norm.weight")
converted_weights.append(matrix)
elif path.endswith("attn/key_norm"):
elif path.endswith("attn/key_norm") and not is_kv_shared_layer:
converted_paths.append(f"{base_path}.self_attn.k_norm.weight")
converted_weights.append(matrix)
elif path.endswith("mlp/gating_einsum"):
Expand Down
Loading