Skip to content

model: support T5xxl with llama.cpp naming convention #654

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
merged 1 commit into from
Jul 2, 2025
+58 −0
Merged

model: support T5xxl with llama.cpp naming convention #654

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
58 changes: 58 additions & 0 deletions model.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -177,6 +177,64 @@ std::unordered_map<std::string, std::string> pmid_v2_name_map = {
std::string convert_open_clip_to_hf_clip(const std::string& name) {
std::string new_name = name;
std::string prefix;
if (contains(new_name, ".enc.")) {
// llama.cpp naming convention for T5
size_t pos = new_name.find(".enc.");
if (pos != std::string::npos) {
new_name.replace(pos, 5, ".encoder.");
}
pos = new_name.find("blk.");
if (pos != std::string::npos) {
new_name.replace(pos, 4, "block.");
}
pos = new_name.find("output_norm.");
if (pos != std::string::npos) {
new_name.replace(pos, 12, "final_layer_norm.");
}
pos = new_name.find("attn_k.");
if (pos != std::string::npos) {
new_name.replace(pos, 7, "layer.0.SelfAttention.k.");
}
pos = new_name.find("attn_v.");
if (pos != std::string::npos) {
new_name.replace(pos, 7, "layer.0.SelfAttention.v.");
}
pos = new_name.find("attn_o.");
if (pos != std::string::npos) {
new_name.replace(pos, 7, "layer.0.SelfAttention.o.");
}
pos = new_name.find("attn_q.");
if (pos != std::string::npos) {
new_name.replace(pos, 7, "layer.0.SelfAttention.q.");
}
pos = new_name.find("attn_norm.");
if (pos != std::string::npos) {
new_name.replace(pos, 10, "layer.0.layer_norm.");
}
pos = new_name.find("ffn_norm.");
if (pos != std::string::npos) {
new_name.replace(pos, 9, "layer.1.layer_norm.");
}
pos = new_name.find("ffn_up.");
if (pos != std::string::npos) {
new_name.replace(pos, 7, "layer.1.DenseReluDense.wi_1.");
}
pos = new_name.find("ffn_down.");
if (pos != std::string::npos) {
new_name.replace(pos, 9, "layer.1.DenseReluDense.wo.");
}
pos = new_name.find("ffn_gate.");
if (pos != std::string::npos) {
new_name.replace(pos, 9, "layer.1.DenseReluDense.wi_0.");
}
pos = new_name.find("attn_rel_b.");
if (pos != std::string::npos) {
new_name.replace(pos, 11, "layer.0.SelfAttention.relative_attention_bias.");
}
} else if (name == "text_encoders.t5xxl.transformer.token_embd.weight") {
new_name = "text_encoders.t5xxl.transformer.shared.weight";
}

if (starts_with(new_name, "conditioner.embedders.0.open_clip.")) {
prefix = "cond_stage_model.";
new_name = new_name.substr(strlen("conditioner.embedders.0.open_clip."));
Expand Down
Loading