| 915 | } |
| 916 | |
| 917 | struct llama_context_params llama_context_params_from_gpt_params(const gpt_params & params) { |
| 918 | auto cparams = llama_context_default_params(); |
| 919 | |
| 920 | cparams.n_ctx = params.n_ctx; |
| 921 | cparams.n_batch = params.n_batch; |
| 922 | cparams.n_threads = params.n_threads; |
| 923 | cparams.n_threads_batch = params.n_threads_batch == -1 ? params.n_threads : params.n_threads_batch; |
| 924 | cparams.mul_mat_q = params.mul_mat_q; |
| 925 | cparams.seed = params.seed; |
| 926 | cparams.f16_kv = params.memory_f16; |
| 927 | cparams.logits_all = params.logits_all; |
| 928 | cparams.embedding = params.embedding; |
| 929 | cparams.rope_scaling_type = params.rope_scaling_type; |
| 930 | cparams.rope_freq_base = params.rope_freq_base; |
| 931 | cparams.rope_freq_scale = params.rope_freq_scale; |
| 932 | cparams.yarn_ext_factor = params.yarn_ext_factor; |
| 933 | cparams.yarn_attn_factor = params.yarn_attn_factor; |
| 934 | cparams.yarn_beta_fast = params.yarn_beta_fast; |
| 935 | cparams.yarn_beta_slow = params.yarn_beta_slow; |
| 936 | cparams.yarn_orig_ctx = params.yarn_orig_ctx; |
| 937 | |
| 938 | return cparams; |
| 939 | } |
| 940 | |
| 941 | void llama_batch_clear(struct llama_batch & batch) { |
| 942 | batch.n_tokens = 0; |
no test coverage detected