| 7622 | } |
| 7623 | |
| 7624 | ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const { |
| 7625 | std::unique_ptr<llm_graph_context> llm; |
| 7626 | |
| 7627 | switch (arch) { |
| 7628 | case LLM_ARCH_LLAMA: |
| 7629 | { |
| 7630 | llm = std::make_unique<llm_build_llama<false>>(*this, params); |
| 7631 | } break; |
| 7632 | case LLM_ARCH_LLAMA4: |
| 7633 | { |
| 7634 | if (hparams.swa_type == LLAMA_SWA_TYPE_NONE) { |
| 7635 | llm = std::make_unique<llm_build_llama<false>>(*this, params); |
| 7636 | } else { |
| 7637 | llm = std::make_unique<llm_build_llama_iswa>(*this, params); |
| 7638 | } |
| 7639 | } break; |
| 7640 | case LLM_ARCH_LLAMA_EMBED: |
| 7641 | { |
| 7642 | llm = std::make_unique<llm_build_llama<true>>(*this, params); |
| 7643 | } break; |
| 7644 | case LLM_ARCH_MAINCODER: |
| 7645 | { |
| 7646 | llm = std::make_unique<llm_build_maincoder>(*this, params); |
| 7647 | } break; |
| 7648 | case LLM_ARCH_DECI: |
| 7649 | { |
| 7650 | llm = std::make_unique<llm_build_deci>(*this, params); |
| 7651 | } break; |
| 7652 | case LLM_ARCH_BAICHUAN: |
| 7653 | { |
| 7654 | llm = std::make_unique<llm_build_baichuan>(*this, params); |
| 7655 | } break; |
| 7656 | case LLM_ARCH_FALCON: |
| 7657 | { |
| 7658 | llm = std::make_unique<llm_build_falcon>(*this, params); |
| 7659 | } break; |
| 7660 | case LLM_ARCH_GROK: |
| 7661 | { |
| 7662 | llm = std::make_unique<llm_build_grok>(*this, params); |
| 7663 | } break; |
| 7664 | case LLM_ARCH_STARCODER: |
| 7665 | { |
| 7666 | llm = std::make_unique<llm_build_starcoder>(*this, params); |
| 7667 | } break; |
| 7668 | case LLM_ARCH_REFACT: |
| 7669 | { |
| 7670 | llm = std::make_unique<llm_build_refact>(*this, params); |
| 7671 | } break; |
| 7672 | case LLM_ARCH_BERT: |
| 7673 | case LLM_ARCH_JINA_BERT_V2: |
| 7674 | case LLM_ARCH_JINA_BERT_V3: |
| 7675 | case LLM_ARCH_NOMIC_BERT: |
| 7676 | case LLM_ARCH_NOMIC_BERT_MOE: |
| 7677 | { |
| 7678 | llm = std::make_unique<llm_build_bert>(*this, params); |
| 7679 | } break; |
| 7680 | case LLM_ARCH_MODERN_BERT: |
| 7681 | { |
no test coverage detected