Skip to content

fix: validate vision projector output dim against LLM hidden size #1918

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
fszontagh wants to merge 2 commits into
base: master
Choose a base branch
from
Open

fix: validate vision projector output dim against LLM hidden size #1918

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
66e2352
fix: validate vision projector output dim against LLM hidden size
fszontagh Aug 27, 2026
916c514
Merge remote-tracking branch 'upstream/master' into fix/llm-vision-di…
fszontagh Aug 27, 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
13 changes: 12 additions & 1 deletion src/model/te/llm.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,6 +73,7 @@ namespace LLM {
int num_heads = 16;
int64_t in_channels = 3;
int64_t out_hidden_size = 3584;
bool out_hidden_size_detected = false;
int temporal_patch_size = 2;
int patch_size = 14;
int spatial_merge_size = 2;
Expand Down Expand Up @@ -238,7 +239,8 @@ namespace LLM {
}
if (contains(name, "visual.merger.linear_fc2.weight") ||
contains(name, "visual.merger.mlp.2.weight")) {
config.vision.out_hidden_size = tensor_storage.ne[1];
config.vision.out_hidden_size = tensor_storage.ne[1];
config.vision.out_hidden_size_detected = true;
}
continue;
}
Expand Down Expand Up @@ -1768,6 +1770,15 @@ namespace LLM {
LOG_WARN("no vision weights detected, vision disabled");
enable_vision = false;
}
// The default would reject valid models, so only compare a detected dim.
if (enable_vision && config.vision.out_hidden_size_detected &&
config.vision.out_hidden_size != config.hidden_size) {
LOG_ERROR("vision projector output size (%" PRId64 ") does not match LLM hidden size (%" PRId64 "), "
"the vision weights (mmproj) likely belong to a different LLM variant, vision disabled",
config.vision.out_hidden_size,
config.hidden_size);
enable_vision = false;
}
if (enable_vision) {
LOG_DEBUG("enable llm vision");
if (config.llama_cpp_style) {
Expand Down
Loading