extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-08-09 14:45:48 +02:00
Code Issues Packages Projects Releases Wiki Activity

talk-llama : sync llama.cpp

Browse Source
This commit is contained in:
Georgi Gerganov
2024-11-15 08:41:06 +02:00
parent 5089ab2d6a
commit 24d706774d
3 changed files with 2005 additions and 1901 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
examples/talk-llama/llama.h
View File

@ -2,6 +2,7 @@
#define LLAMA_H
#include "ggml.h"
#include "ggml-cpu.h"
#include "ggml-backend.h"
#include <stddef.h>
@ -205,7 +206,7 @@ extern "C" {
enum llama_split_mode {
LLAMA_SPLIT_MODE_NONE = 0, // single GPU
LLAMA_SPLIT_MODE_LAYER = 1, // split layers and KV across GPUs
LLAMA_SPLIT_MODE_ROW = 2, // split rows across GPUs
LLAMA_SPLIT_MODE_ROW = 2, // split layers and KV across GPUs, use tensor parallelism if supported
};
// TODO: simplify (https://github.com/ggerganov/llama.cpp/pull/9294#pullrequestreview-2286561979)
@ -274,10 +275,7 @@ extern "C" {
int32_t n_gpu_layers; // number of layers to store in VRAM
enum llama_split_mode split_mode; // how to split the model across multiple GPUs
// main_gpu interpretation depends on split_mode:
// LLAMA_SPLIT_MODE_NONE: the GPU that is used for the entire model
// LLAMA_SPLIT_MODE_ROW: the GPU that is used for small tensors and intermediate results
// LLAMA_SPLIT_MODE_LAYER: ignored
// the GPU that is used for the entire model when split_mode is LLAMA_SPLIT_MODE_NONE
int32_t main_gpu;
// proportion of the model (layers or rows) to offload to each GPU, size: llama_max_devices()
@ -799,7 +797,7 @@ extern "C" {
// Processes a batch of tokens with the ecoder part of the encoder-decoder model.
// Stores the encoder output internally for later use by the decoder cross-attention layers.
// 0 - success
// < 0 - error
// < 0 - error. the KV cache state is restored to the state before this call
LLAMA_API int32_t llama_encode(
struct llama_context * ctx,
struct llama_batch batch);
@ -807,7 +805,7 @@ extern "C" {
// Positive return values does not mean a fatal error, but rather a warning.
// 0 - success
// 1 - could not find a KV slot for the batch (try reducing the size of the batch or increase the context)
// < 0 - error
// < 0 - error. the KV cache state is restored to the state before this call
LLAMA_API int32_t llama_decode(
struct llama_context * ctx,
struct llama_batch batch);
@ -1087,9 +1085,6 @@ extern "C" {
/// @details Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
LLAMA_API struct llama_sampler * llama_sampler_init_min_p (float p, size_t min_keep);
/// @details Tail Free Sampling described in https://www.trentonbricken.com/Tail-Free-Sampling/.
LLAMA_API struct llama_sampler * llama_sampler_init_tail_free (float z, size_t min_keep);
/// @details Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
LLAMA_API struct llama_sampler * llama_sampler_init_typical (float p, size_t min_keep);