Skip to content

minicpm: make embedding_scale residual_scale logit_scale optional with legacy defaults. #16273

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 2 commits into from
Sep 26, 2025
Merged

minicpm: make embedding_scale residual_scale logit_scale optional with legacy defaults. #16273

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6337679
minicpm: make GGUF scaling keys optional with legacy defaults
vinkal-chudgar Sep 25, 2025
28a4100
Update src/llama-model.cpp
vinkal-chudgar Sep 26, 2025
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
17 changes: 14 additions & 3 deletions src/llama-model.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -675,10 +675,21 @@ void llama_model::load_hparams(llama_model_loader & ml) {
} break;
case LLM_ARCH_MINICPM:
{
// Backward-compatible defaults for older MiniCPM GGUFs:
// historical MiniCPM graph constants:
// n_embd_base = 256, scale_embd = 12.0f, scale_depth = 1.4f

hparams.f_embedding_scale = 12.0f;
const float scale_depth = 1.4f;
hparams.f_residual_scale = scale_depth / sqrtf((float) hparams.n_layer);
const float n_embd_base = 256.0f;
hparams.f_logit_scale = hparams.n_embd ? (n_embd_base / float(hparams.n_embd)) : 1.0f;
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale);
ml.get_key(LLM_KV_RESIDUAL_SCALE, hparams.f_residual_scale);
ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale);

// Optional KV reads, override defaults if present in newer GGUF exports
ml.get_key(LLM_KV_EMBEDDING_SCALE, hparams.f_embedding_scale, /*required=*/false);
ml.get_key(LLM_KV_RESIDUAL_SCALE, hparams.f_residual_scale, /*required=*/false);
ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale, /*required=*/false);

// MiniCPM uses rope by default, unlike Granite which uses it as a switch
hparams.rope_finetuned = true;
Expand Down