vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-14 02:49:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

model : gemma3n text-only (#14400)

Browse source 
* gemma3n

* add llm_graph_input_one
This commit is contained in:
Xuan-Son Nguyen 2025-06-26 19:34:02 +02:00 • committed by GitHub
parent a01047b041
commit 8846aace49
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 960 additions and 15 deletions
Download patch file Download diff file Expand all files Collapse all files

9
src/llama-quant.cpp
View file

@ -223,7 +223,7 @@ static ggml_type llama_tensor_get_type(quantize_state_impl & qs, ggml_type new_t
new_type = GGML_TYPE_Q6_K;
}
}
} else if (name == "token_embd.weight") {
} else if (name == "token_embd.weight" || name == "per_layer_token_embd.weight") {
if (qs.params->token_embedding_type < GGML_TYPE_COUNT) {
new_type = qs.params->token_embedding_type;
} else {
@ -830,6 +830,13 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
// NOTE: can't use LLM_TN here because the layer number is not known
quantize &= name.find("ffn_gate_inp.weight") == std::string::npos;
// these are very small (e.g. 4x4)
quantize &= name.find("altup") == std::string::npos;
quantize &= name.find("laurel") == std::string::npos;
// these are not too big so keep them as it is
quantize &= name.find("per_layer_model_proj") == std::string::npos;
// do not quantize positional embeddings and token types (BERT)
quantize &= name != LLM_TN(model.arch)(LLM_TENSOR_POS_EMBD, "weight");
quantize &= name != LLM_TN(model.arch)(LLM_TENSOR_TOKEN_TYPES, "weight");