fix: correct Gemma3 rope settings and vram limit propagation (#1583)

This commit is contained in:
leejet 2026-05-30 22:23:49 +08:00 committed by GitHub
parent d3b2cb047e
commit d2797b8667
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 9 additions and 4 deletions

View File

@ -2290,6 +2290,11 @@ struct LTXAVEmbedder : public Conditioner {
projector->set_flash_attention_enabled(enabled);
}
void set_max_graph_vram_bytes(size_t max_vram_bytes) override {
llm->set_max_graph_vram_bytes(max_vram_bytes);
projector->set_max_graph_vram_bytes(max_vram_bytes);
}
void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) override {
llm->set_weight_adapter(adapter);
projector->set_weight_adapter(adapter);

View File

@ -879,8 +879,8 @@ namespace LLM {
input_pos,
nullptr,
head_dim,
GGML_ROPE_TYPE_NORMAL,
0,
GGML_ROPE_TYPE_NEOX,
131072,
rope_theta,
freq_scale,
0.f,
@ -892,8 +892,8 @@ namespace LLM {
input_pos,
nullptr,
head_dim,
GGML_ROPE_TYPE_NORMAL,
0,
GGML_ROPE_TYPE_NEOX,
131072,
rope_theta,
freq_scale,
0.f,