vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge commit 'bcc0eb4591' into ceb/nomic-vulkan

Browse source
This commit is contained in:
Jared Van Bortel 2024-01-08 15:50:18 -05:00
commit 8b65f4c5e5
11 changed files with 755 additions and 291 deletions
Download patch file Download diff file Expand all files Collapse all files

13
llama.h
View file

@ -42,7 +42,7 @@
#define LLAMA_FILE_MAGIC_GGSN 0x6767736eu // 'ggsn'
#define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
#define LLAMA_SESSION_VERSION 2
#define LLAMA_SESSION_VERSION 3
#if defined(GGML_USE_CUBLAS) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_METAL) || defined(GGML_USE_KOMPUTE)
// Defined when llama.cpp is compiled with support for offloading model layers to GPU.
@ -211,11 +211,14 @@ extern "C" {
float yarn_beta_slow; // YaRN high correction dim
uint32_t yarn_orig_ctx; // YaRN original context size
enum ggml_type type_k; // data type for K cache
enum ggml_type type_v; // data type for V cache
// Keep the booleans together to avoid misalignment during copy-by-value.
bool mul_mat_q; // if true, use experimental mul_mat_q kernels (DEPRECATED - always true)
bool f16_kv; // use fp16 for KV cache, fp32 otherwise
bool logits_all; // the llama_eval() call computes all logits, not just the last one
bool embedding; // embedding mode only
bool mul_mat_q; // if true, use experimental mul_mat_q kernels (DEPRECATED - always true)
bool logits_all; // the llama_eval() call computes all logits, not just the last one
bool embedding; // embedding mode only
bool offload_kqv; // whether to offload the KQV ops (including the KV cache) to GPU
};
// model quantization parameters