mirror of
https://github.com/ggerganov/whisper.cpp.git
synced 2024-12-18 20:27:53 +00:00
8932c2d6ce
* llama : add pipeline parallelism support for batch processing with multiple CUDA GPUs ggml-ci * server : add -ub, --ubatch-size parameter * fix server embedding test * llama : fix Mamba inference for pipeline parallelism Tested to work correctly with both `main` and `parallel` examples. * llama : limit max batch size to n_batch * add LLAMA_SCHED_MAX_COPIES to configure the number of input copies for pipeline parallelism default increase to 4 (from 2) changing this value may improve performance for some systems, but increases memory usage * fix hip build * fix sycl build (disable cpy_tensor_async) * fix hip build * llama : limit n_batch and n_ubatch to n_ctx during context creation * llama : fix norm backend * batched-bench : sync after decode * swiftui : sync after decode * ggml : allow ggml_get_rows to use multiple threads if they are available * check n_ubatch >= n_tokens with non-casual attention * llama : do not limit n_batch to n_ctx with non-casual attn * server : construct batch with size of llama_n_batch * ggml_backend_cpu_graph_compute : fix return value when alloc fails * llama : better n_batch and n_ubatch comment * fix merge * small fix * reduce default n_batch to 2048 --------- Co-authored-by: Francis Couture-Harpin <git@compilade.net> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
77 lines
2.9 KiB
C
77 lines
2.9 KiB
C
#pragma once
|
|
|
|
#include "ggml.h"
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
typedef struct ggml_backend_buffer_type * ggml_backend_buffer_type_t;
|
|
typedef struct ggml_backend_buffer * ggml_backend_buffer_t;
|
|
typedef struct ggml_backend * ggml_backend_t;
|
|
|
|
// Tensor allocator
|
|
struct ggml_tallocr {
|
|
ggml_backend_buffer_t buffer;
|
|
void * base;
|
|
size_t alignment;
|
|
size_t offset;
|
|
};
|
|
|
|
GGML_API struct ggml_tallocr ggml_tallocr_new(ggml_backend_buffer_t buffer);
|
|
GGML_API void ggml_tallocr_alloc(struct ggml_tallocr * talloc, struct ggml_tensor * tensor);
|
|
|
|
// Graph allocator
|
|
/*
|
|
Example usage:
|
|
ggml_gallocr_t galloc = ggml_gallocr_new(ggml_bacckend_cpu_buffer_type());
|
|
|
|
// optional: create a worst-case graph and reserve the buffers to avoid reallocations
|
|
ggml_gallocr_reserve(galloc, build_graph(max_batch));
|
|
|
|
// allocate the graph
|
|
struct ggml_cgraph * graph = build_graph(batch);
|
|
ggml_gallocr_alloc_graph(galloc, graph);
|
|
|
|
printf("compute buffer size: %zu bytes\n", ggml_gallocr_get_buffer_size(galloc, 0));
|
|
|
|
// evaluate the graph
|
|
ggml_backend_graph_compute(backend, graph);
|
|
*/
|
|
|
|
// special tensor flags for use with the graph allocator:
|
|
// ggml_set_input(): all input tensors are allocated at the beginning of the graph in non-overlapping addresses
|
|
// ggml_set_output(): output tensors are never freed and never overwritten
|
|
|
|
typedef struct ggml_gallocr * ggml_gallocr_t;
|
|
|
|
GGML_API ggml_gallocr_t ggml_gallocr_new(ggml_backend_buffer_type_t buft);
|
|
GGML_API ggml_gallocr_t ggml_gallocr_new_n(ggml_backend_buffer_type_t * bufts, int n_bufs);
|
|
GGML_API void ggml_gallocr_free(ggml_gallocr_t galloc);
|
|
|
|
// pre-allocate buffers from a measure graph - does not allocate or modify the graph
|
|
// call with a worst-case graph to avoid buffer reallocations
|
|
// not strictly required for single buffer usage: ggml_gallocr_alloc_graph will reallocate the buffers automatically if needed
|
|
// returns false if the buffer allocation failed
|
|
GGML_API bool ggml_gallocr_reserve(ggml_gallocr_t galloc, struct ggml_cgraph * graph);
|
|
GGML_API bool ggml_gallocr_reserve_n(
|
|
ggml_gallocr_t galloc,
|
|
struct ggml_cgraph * graph,
|
|
const int * node_buffer_ids,
|
|
const int * leaf_buffer_ids);
|
|
|
|
// automatic reallocation if the topology changes when using a single buffer
|
|
// returns false if using multiple buffers and a re-allocation is needed (call ggml_gallocr_reserve_n first to set the node buffers)
|
|
GGML_API bool ggml_gallocr_alloc_graph(ggml_gallocr_t galloc, struct ggml_cgraph * graph);
|
|
|
|
GGML_API size_t ggml_gallocr_get_buffer_size(ggml_gallocr_t galloc, int buffer_id);
|
|
|
|
// Utils
|
|
// Create a buffer and allocate all the tensors in a ggml_context
|
|
GGML_API struct ggml_backend_buffer * ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_context * ctx, ggml_backend_buffer_type_t buft);
|
|
GGML_API struct ggml_backend_buffer * ggml_backend_alloc_ctx_tensors(struct ggml_context * ctx, ggml_backend_t backend);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|