Skip to content

Commit 50712b6

Browse files
converter: clarify Eagle2-VL projector layer indexing
Co-authored-by: Copilot <[email protected]>
1 parent 6711a0d commit 50712b6

File tree

1 file changed

+3
-0
lines changed

1 file changed

+3
-0
lines changed

convert_hf_to_gguf.py

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3965,6 +3965,9 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
39653965
return [(new_name, data_torch)]
39663966
return [(new_name, data_torch)]
39673967
# Map second Linear (mlp1.3.*) -> mm.2.*
3968+
# Note: In the HuggingFace model, the projector is implemented as a Sequential module:
3969+
# [0]: LayerNorm, [1]: Linear, [2]: GELU, [3]: Linear
3970+
# So the second Linear layer is at index 3 (mlp1.3.*), not 2, because GELU occupies index 2.
39683971
if mlp_suffix.startswith("3."):
39693972
new_name = "mm.2." + mlp_suffix[2:]
39703973
if new_name.endswith(".weight"):

0 commit comments

Comments
 (0)