vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-12 09:59:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

Merge branch 'master' into concedo_experimental

Browse source 
# Conflicts:
#	.devops/nix/sif.nix
#	.github/workflows/build.yml
#	.github/workflows/python-check-requirements.yml
#	README-sycl.md
#	README.md
#	flake.lock
#	flake.nix
#	requirements/requirements-convert-hf-to-gguf.txt
#	scripts/compare-llama-bench.py
This commit is contained in:
Concedo 2024-03-04 15:33:33 +08:00
commit 7c64845dea
41 changed files with 3325 additions and 2053 deletions
Download patch file Download diff file Expand all files Collapse all files

21
llama.h
View file

@ -129,6 +129,7 @@ extern "C" {
};
enum llama_pooling_type {
LLAMA_POOLING_TYPE_UNSPECIFIED = -1,
LLAMA_POOLING_TYPE_NONE = 0,
LLAMA_POOLING_TYPE_MEAN = 1,
LLAMA_POOLING_TYPE_CLS = 2,
@ -236,7 +237,10 @@ extern "C" {
uint32_t n_batch; // prompt processing maximum batch size
uint32_t n_threads; // number of threads to use for generation
uint32_t n_threads_batch; // number of threads to use for batch processing
int32_t rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
enum llama_rope_scaling_type rope_scaling_type; // RoPE scaling type, from `enum llama_rope_scaling_type`
enum llama_pooling_type pooling_type; // whether to pool (sum) embedding results by sequence id
// (ignored if no pooling layer)
// ref: https://github.com/ggerganov/llama.cpp/pull/2054
float rope_freq_base; // RoPE base frequency, 0 = from model
@ -255,11 +259,15 @@ extern "C" {
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 logits_all; // the llama_eval() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
bool logits_all; // the llama_decode() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
bool embedding; // embedding mode only
bool offload_kqv; // whether to offload the KQV ops (including the KV cache) to GPU
bool do_pooling; // whether to pool (sum) embedding results by sequence id (ignored if no pooling layer)
// Abort callback
// if it returns true, execution of llama_decode() will be aborted
// currently works only with CPU execution
ggml_abort_callback abort_callback;
void * abort_callback_data;
};
// model quantization parameters
@ -635,7 +643,10 @@ extern "C" {
// n_threads_batch is the number of threads used for prompt and batch processing (multiple tokens)
LLAMA_API void llama_set_n_threads(struct llama_context * ctx, uint32_t n_threads, uint32_t n_threads_batch);
// Token logits obtained from the last call to llama_eval()
// Set abort callback
LLAMA_API void llama_set_abort_callback(struct llama_context * ctx, ggml_abort_callback abort_callback, void * abort_callback_data);
// Token logits obtained from the last call to llama_decode()
// The logits for the last token are stored in the last row
// Logits for which llama_batch.logits[i] == 0 are undefined
// Rows: n_tokens provided with llama_batch