model : Add support for Metis-1.6 models (#31134)

Assisted-by: Claude
This commit is contained in:
Giulianno authored and GitHub committed 2026-11-03 16:24:45 +01:00
1 parent d1044bc827
commit 9309bf1bcf
20 files changed
+1112 -2

No files matched your search

+5 -1
View File
@@ -242,6 +242,8 @@ static llama_model * llama_model_mapping(llm_arch arch, const llama_model_params
return new llama_model_nemotron_h(params);
case LLM_ARCH_NEMOTRON_H_MOE:
return new llama_model_nemotron_h_moe(params);
case LLM_ARCH_METIS16:
return new llama_model_metis16(params);
case LLM_ARCH_EXAONE:
return new llama_model_exaone(params);
case LLM_ARCH_EXAONE4:
@@ -515,7 +517,7 @@ struct ggml_backend_meta_split_state llama_meta_device_get_split_state(const str
};
auto get_tensor_config = [&]() -> tensor_config {
if (ud->model->arch == LLM_ARCH_HRM_TEXT) {
if (ud->model->arch == LLM_ARCH_HRM_TEXT || ud->model->arch == LLM_ARCH_METIS16) {
// aliased cache slots cannot satisfy the meta-split invariants, so replicate all tensors
return {GGML_BACKEND_SPLIT_AXIS_MIRRORED, tensor, 0, 0};
}
@@ -1049,6 +1051,7 @@ const char * llm_type_name(llm_type type) {
case LLM_TYPE_7B: return "7B";
case LLM_TYPE_8B: return "8B";
case LLM_TYPE_9B: return "9B";
case LLM_TYPE_10B: return "10B";
case LLM_TYPE_11B: return "11B";
case LLM_TYPE_12B: return "12B";
case LLM_TYPE_13B: return "13B";
@@ -3618,6 +3621,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_WAVTOKENIZER_DEC:
case LLM_ARCH_NEMOTRON_H:
case LLM_ARCH_NEMOTRON_H_MOE:
case LLM_ARCH_METIS16:
case LLM_ARCH_KIMI_LINEAR:
case LLM_ARCH_KIMI_K3:
case LLM_ARCH_GLM5_NEXT: