thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

llama : add support for Nemotron 3 Super (#20411)

Browse source 
* llama : add support for Nemotron 3 Super

This commit adds support for the Nemotron 3 Super model (120B.A12B)
enabling this model to be converted to GGUF format and run in llama.cpp.

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Matt Clayton <156335168+mattjcly@users.noreply.github.com>
This commit is contained in:
Daniel Bevenius 2026-03-11 19:27:53 +01:00 • committed by GitHub
parent 76ea1c1c46
commit eaf1d7930c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
11 changed files with 97 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

22
src/models/nemotron-h.cpp
View file

@ -114,9 +114,18 @@ ggml_tensor * llm_build_nemotron_h::build_ffn_layer(ggml_tensor * cur, const lla
LLM_FFN_RELU_SQR, LLM_FFN_PAR, il);
cb(cur, "ffn_out", il);
} else {
ggml_tensor * ffn_inp = cur;
ggml_tensor * inp_emb = cur;
ggml_tensor * inp_latent = cur;
if (model.layers[il].ffn_latent_down) {
inp_latent = ggml_mul_mat(ctx0, model.layers[il].ffn_latent_down, cur);
}
ggml_tensor * router_logits = build_lora_mm(model.layers[il].ffn_gate_inp, cur);
cb(router_logits, "ffn_moe_logits", il);
ggml_tensor * moe_out =
build_moe_ffn(ffn_inp,
build_moe_ffn(inp_latent,
model.layers[il].ffn_gate_inp,
model.layers[il].ffn_up_exps,
nullptr, // no gate
@ -126,10 +135,15 @@ ggml_tensor * llm_build_nemotron_h::build_ffn_layer(ggml_tensor * cur, const lla
LLM_FFN_RELU_SQR, hparams.expert_weights_norm,
hparams.expert_weights_scale,
LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID,
il);
il,
router_logits);
cb(moe_out, "ffn_moe_out", il);
ggml_tensor * ffn_shexp = build_ffn(ffn_inp,
if (model.layers[il].ffn_latent_up) {
moe_out = ggml_mul_mat(ctx0, model.layers[il].ffn_latent_up, moe_out);
}
ggml_tensor * ffn_shexp = build_ffn(inp_emb,
model.layers[il].ffn_up_shexp, NULL, NULL,
NULL /* no gate */ , NULL, NULL,
model.layers[il].ffn_down_shexp, NULL, NULL,