diff options
| author | cebtenzzre <cebtenzzre@gmail.com> | 2023-11-01 18:04:33 -0400 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-11-01 18:04:33 -0400 |
| commit | 898aeca90a9bb992f506234cf3b8b7f7fa28a1df (patch) | |
| tree | 125f8a9b466efd4534ecd3e64419ece001c86a7d /ggml.h | |
| parent | c43c2da8afacaddfe51c09b21dbd9922cd0ea46b (diff) | |
llama : implement YaRN RoPE scaling (#2268)
Co-authored-by: cebtenzzre <cebtenzzre@gmail.com>
Co-authored-by: Jeffrey Quesnelle <jquesnelle@gmail.com>
Diffstat (limited to 'ggml.h')
| -rw-r--r-- | ggml.h | 20 |
1 files changed, 17 insertions, 3 deletions
@@ -219,7 +219,7 @@ #define GGML_MAX_CONTEXTS 64 #define GGML_MAX_SRC 6 #define GGML_MAX_NAME 64 -#define GGML_MAX_OP_PARAMS 32 +#define GGML_MAX_OP_PARAMS 64 #define GGML_DEFAULT_N_THREADS 4 #if UINTPTR_MAX == 0xFFFFFFFF @@ -1326,8 +1326,13 @@ extern "C" { int n_dims, int mode, int n_ctx, + int n_orig_ctx, float freq_base, - float freq_scale); + float freq_scale, + float ext_factor, + float attn_factor, + float beta_fast, + float beta_slow); // in-place, returns view(a) GGML_API struct ggml_tensor * ggml_rope_custom_inplace( @@ -1337,8 +1342,17 @@ extern "C" { int n_dims, int mode, int n_ctx, + int n_orig_ctx, float freq_base, - float freq_scale); + float freq_scale, + float ext_factor, + float attn_factor, + float beta_fast, + float beta_slow); + + // compute correction dims for YaRN RoPE scaling + void ggml_rope_yarn_corr_dims( + int n_dims, int n_orig_ctx, float freq_base, float beta_fast, float beta_slow, float dims[2]); // xPos RoPE, in-place, returns view(a) GGML_API struct ggml_tensor * ggml_rope_xpos_inplace( |