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

now printing tensors

Browse source
This commit is contained in:
mike dupont 2023-11-25 20:06:00 -05:00
parent af698c6f27
commit ec2b03e504
6 changed files with 140 additions and 58 deletions
Download patch file Download diff file Expand all files Collapse all files

4
binding.py
View file

@ -14,9 +14,9 @@ llvmLibPath = "/usr/lib/llvm-15/lib/"
cxxClientRoot = "/home/mdupont/experiments/llama.cpp/"
fileList = [
# "ggml.cpp",
"ggml.cpp",
# "llama.cpp",
"examples/server/server.cpp",
# "examples/server/server.cpp",
]
typeList = [

3
examples/main/main.cpp
View file

@ -30,7 +30,8 @@
#if defined(_MSC_VER)
#pragma warning(disable: 4244 4267) // possible loss of data
#endif
#include "ggml-internal.hpp"
#include "llama-internal.hpp"
#include "print.hpp"
static llama_context ** g_ctx;

1
examples/server/server.cpp
View file

@ -24,6 +24,7 @@
#include <thread>
#include <mutex>
#include <chrono>
#include "llama-internal.hpp"
#include "print.hpp"
#ifndef SERVER_VERBOSE

22
ggml.cpp
View file

@ -4,6 +4,7 @@
#include "ggml-impl.h"
#include "ggml-quants.h"
#if defined(_MSC_VER) || defined(__MINGW32__)
#include <malloc.h> // using malloc.h with MSC/MINGW
#elif !defined(__FreeBSD__) && !defined(__NetBSD__) && !defined(__OpenBSD__)
@ -46,6 +47,12 @@ void type_traits_init();
void GGUF_TYPE_SIZE_init();
void GGUF_TYPE_NAME_init();
#include "llama.h"
struct ggml_allocr;
//#include "ggml-internal.hpp"
#include "llama-internal.hpp"
#include "print.hpp"
#if defined(_WIN32)
#include <windows.h>
@ -9412,7 +9419,10 @@ static void ggml_compute_forward_mul_mat(
const struct ggml_tensor * src0,
const struct ggml_tensor * src1,
struct ggml_tensor * dst) {
print_fields(*params);
print_fields(*src0);
print_fields(*src1);
print_fields(*dst);
int64_t t0 = ggml_perf_time_us();
UNUSED(t0);
@ -9456,6 +9466,7 @@ static void ggml_compute_forward_mul_mat(
if (params->ith == 0 && params->type == GGML_TASK_COMPUTE) {
ggml_cl_mul_mat(src0, src1, dst, params->wdata, params->wsize);
}
print_fields(*dst);
return;
}
#endif
@ -9463,10 +9474,12 @@ static void ggml_compute_forward_mul_mat(
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
if (params->ith != 0) {
print_fields(*dst);
return;
}
if (params->type == GGML_TASK_INIT) {
return;
}
@ -9508,7 +9521,7 @@ static void ggml_compute_forward_mul_mat(
}
//printf("CBLAS = %f ms, %d x %d x %d x %d\n", (ggml_perf_time_us() - t0)/1000.0, ne0, ne1, ne2, ne3);
print_fields(*dst);
return;
}
#endif
@ -9527,11 +9540,12 @@ static void ggml_compute_forward_mul_mat(
}
}
}
print_fields(*dst);
return;
}
if (params->type == GGML_TASK_FINALIZE) {
print_fields(*dst);
return;
}
@ -9565,6 +9579,7 @@ static void ggml_compute_forward_mul_mat(
// threads with no work simply yield (not sure if it helps)
if (ir010 >= ir011 || ir110 >= ir111) {
sched_yield();
return;
}
@ -9617,6 +9632,7 @@ static void ggml_compute_forward_mul_mat(
}
}
}
print_fields(*dst);
}
// ggml_compute_forward_out_prod

6
llama-internal.hpp
View file

@ -1,5 +1,10 @@
#include <set>
#include <queue>
#include <map>
#include <random>
#include <functional>
#include <unordered_map>
#include <memory>
enum llm_arch {
LLM_ARCH_LLAMA,
LLM_ARCH_FALCON,
@ -451,6 +456,7 @@ struct llama_model {
}
};
struct ggml_allocr;
struct llama_context {
llama_context(const llama_model & model) : model(model), t_start_us(model.t_start_us), t_load_us(model.t_load_us) {}
~llama_context();

156
print.hpp
View file

@ -1,8 +1,9 @@
#include <refl-cpp/refl.hpp>
#include <iostream>
#include "llama.h"
#include "ggml-internal.hpp"
#include "llama-internal.hpp"
#include "common/common.h"
//#include "ggml-internal.hpp"
//#include "llama-internal.hpp"
REFL_TYPE(ggml_init_params )
REFL_END
@ -96,9 +97,10 @@ REFL_END
REFL_TYPE(llama_sampling_params)
REFL_END
#ifdef llm_arch
REFL_TYPE(llm_arch)
REFL_END
#endif
REFL_TYPE(llama_sampling_context )
REFL_FIELD( params)
REFL_FIELD( mirostat_mu)
@ -125,10 +127,34 @@ REFL_END
REFL_TYPE(ggml_tensor)
REFL_FIELD(type)
REFL_FIELD(type)
REFL_FIELD(backend)
REFL_FIELD(buffer)
REFL_FIELD(n_dims)
REFL_FIELD(ne)
REFL_FIELD(nb)
REFL_FIELD(op)
REFL_FIELD(op_params)
REFL_FIELD(is_param)
REFL_FIELD(grad)
REFL_FIELD(src)
REFL_FIELD(perf_runs)
REFL_FIELD(perf_cycles)
REFL_FIELD(perf_time_us)
REFL_FIELD(view_src)
REFL_FIELD(view_offs)
REFL_FIELD(data)
REFL_FIELD(name)
REFL_FIELD(extra)
REFL_FIELD(padding)
REFL_END
REFL_TYPE(ggml_cplan)
REFL_FIELD(work_size)
REFL_FIELD(work_data)
REFL_FIELD(n_threads)
REFL_FIELD(abort_callback)
REFL_FIELD(abort_callback_data)
REFL_END
REFL_TYPE(ggml_hash_set)
@ -137,14 +163,32 @@ REFL_END
REFL_TYPE(ggml_cgraph)
REFL_FIELD(size)
REFL_FIELD(n_nodes)
REFL_FIELD(n_leafs)
REFL_FIELD(nodes)
REFL_FIELD(grads)
REFL_FIELD(leafs)
REFL_FIELD(visited_hash_table)
REFL_FIELD(order)
REFL_FIELD(perf_runs)
REFL_FIELD(perf_cycles)
REFL_FIELD(perf_time_us)
REFL_END
REFL_TYPE(ggml_scratch)
REFL_FIELD(offs)
REFL_FIELD(size)
REFL_FIELD(data)
REFL_END
REFL_TYPE(ggml_compute_params)
REFL_FIELD(type)
REFL_FIELD(ith)
REFL_FIELD(nth)
REFL_FIELD(wsize)
REFL_FIELD(wdata)
REFL_END
REFL_TYPE(ggml_opt_params)
@ -162,6 +206,7 @@ REFL_TYPE(ggml_something)
REFL_FIELD(type_name)
REFL_END
#ifdef ggml_context
REFL_TYPE(ggml_context)
REFL_FIELD(mem_size)
REFL_FIELD(mem_buffer)
@ -173,14 +218,17 @@ REFL_FIELD( objects_begin)
REFL_FIELD( objects_end)
REFL_FIELD( scratch)
REFL_FIELD( scratch_save)
REFL_END
#endif
#ifdef ggml_context_container
REFL_TYPE(ggml_context_container)
REFL_FIELD(used)
REFL_FIELD(context)
REFL_END
#endif
#ifdef ggml_numa_node
REFL_TYPE(ggml_numa_node)
REFL_FIELD(cpus)
REFL_FIELD(n_cpus)
@ -220,6 +268,7 @@ REFL_TYPE(hash_map)
REFL_FIELD(set)
REFL_FIELD(vals)
REFL_END
REFL_TYPE(ggml_compute_state_shared)
REFL_FIELD(cgraph)
REFL_FIELD(cplan)
@ -232,12 +281,15 @@ REFL_TYPE(ggml_lbfgs_iteration_data)
REFL_FIELD(alpha)
REFL_FIELD(ys)
REFL_END
#endif
#ifdef gguf_kv
REFL_TYPE(gguf_kv)
REFL_FIELD(key)
REFL_FIELD(type)
REFL_END
REFL_TYPE(gguf_header)
REFL_FIELD(magic)
REFL_FIELD(version)
@ -257,7 +309,7 @@ REFL_TYPE(gguf_buf)
REFL_FIELD(data)
REFL_FIELD(size)
REFL_END
#endif
REFL_TYPE(llama_model_params)
REFL_FIELD(n_gpu_layers)
@ -265,6 +317,7 @@ REFL_END
REFL_TYPE(llama_context_params)
REFL_FIELD(seed)
REFL_END
REFL_TYPE(llama_model_quantize_params)
REFL_FIELD(nthread)
REFL_END
@ -283,12 +336,15 @@ REFL_TYPE(llama_beams_state)
REFL_FIELD(beam_views)
REFL_END
#ifdef ggml_backend
REFL_TYPE(ggml_backend)
REFL_END
#endif
REFL_TYPE(ggml_backend_buffer)
REFL_END
#ifdef ggml_allocr
REFL_TYPE(ggml_allocr)
REFL_END
@ -298,13 +354,14 @@ REFL_END
REFL_TYPE(ggml_gallocr)
REFL_END
#endif
#ifdef llama_buffer
REFL_TYPE(llama_buffer)
REFL_FIELD(data)
REFL_FIELD(size)
REFL_END
REFL_TYPE(llama_file)
REFL_FIELD(fp)
REFL_FIELD(size)
@ -353,91 +410,92 @@ REFL_TYPE(llama_kv_cache)
REFL_FIELD(has_shift)
REFL_FIELD(head)
REFL_END
#endif
#ifdef e_model
REFL_TYPE(e_model)
REFL_END
#endif
REFL_TYPE(llama_ftype)
REFL_END
//#ifdef llama_model
REFL_TYPE(llama_model)
REFL_FIELD(type)
REFL_FIELD(arch)
REFL_FIELD(type)
REFL_FIELD(arch)
REFL_FIELD(ftype )
REFL_FIELD( name )
REFL_FIELD( hparams )
REFL_FIELD( vocab)
REFL_FIELD( tok_embd)
REFL_FIELD( pos_embd)
REFL_FIELD( tok_norm)
REFL_FIELD( tok_norm_b)
REFL_FIELD( output_norm)
REFL_FIELD( output_norm_b)
REFL_FIELD( output)
REFL_FIELD( layers)
REFL_FIELD( n_gpu_layers)
REFL_FIELD( gguf_kv) //unordered map
REFL_FIELD( ctx)
REFL_FIELD( buf)
REFL_FIELD( mapping) //std::unique_ptr
REFL_FIELD( mlock_buf)
REFL_FIELD( mlock_mmap)
REFL_FIELD( tensors_by_name)
REFL_FIELD( t_load_us)
REFL_FIELD( t_start_us)
REFL_FIELD(name )
REFL_FIELD(hparams )
REFL_FIELD(vocab)
REFL_FIELD(tok_embd)
REFL_FIELD(pos_embd)
REFL_FIELD(tok_norm)
REFL_FIELD(tok_norm_b)
REFL_FIELD(output_norm)
REFL_FIELD(output_norm_b)
REFL_FIELD(output)
REFL_FIELD(layers)
REFL_FIELD(n_gpu_layers)
REFL_FIELD(gguf_kv) //unordered map
REFL_FIELD(ctx)
REFL_FIELD(buf)
REFL_FIELD(mapping) //std::unique_ptr
REFL_FIELD(mlock_buf)
REFL_FIELD(mlock_mmap)
REFL_FIELD(tensors_by_name)
REFL_FIELD(t_load_us)
REFL_FIELD(t_start_us)
REFL_END
//#endif
#ifdef llama_vocab
REFL_TYPE(llama_vocab)
REFL_END
#endif
REFL_TYPE(grammar_parser::parse_state)
REFL_END
REFL_TYPE(grammar_parser::parse_state)
REFL_END
//#ifdef llama_context
REFL_TYPE(llama_context)
REFL_FIELD( cparams)
//REFL_FIELD(model)
REFL_FIELD(kv_self)
REFL_FIELD(rng) //random numbers
REFL_FIELD(rng) //random numbers
REFL_FIELD(has_evaluated_once )
REFL_FIELD( t_start_us)
REFL_FIELD( t_load_us)
REFL_FIELD( t_sample_us )
REFL_FIELD( t_sample_us )
REFL_FIELD( t_p_eval_us )
REFL_FIELD( t_eval_us)
REFL_FIELD( t_eval_us)
REFL_FIELD( n_sample )
REFL_FIELD( n_p_eval )
REFL_FIELD( n_eval )
REFL_FIELD( n_eval )
//REFL_FIELD( logits) crash
REFL_FIELD( logits_all )
REFL_FIELD( embedding)
//REFL_FIELD( work_buffer)
REFL_FIELD( buf_compute)
REFL_FIELD( buf_alloc)
REFL_FIELD( buf_compute)
REFL_FIELD( buf_alloc)
REFL_FIELD( alloc )
#ifdef GGML_USE_METAL
REFL_FIELD( ctx_metal )
#endif
#ifdef GGML_USE_MPI
REFL_FIELD( ctx_mpi )
#endif
REFL_END
//#endif
#ifdef llama_model_loader
REFL_TYPE(llama_model_loader)
REFL_FIELD(n_kv)
REFL_FIELD(n_tensors)
REFL_END
#endif
#ifdef llm_build_context
REFL_TYPE(llm_build_context)
// REFL_FIELD(model) cannot create pointer to reference member ‘llm_build_context::model’
// REFL_FIELD(hparams) cannot create pointer to reference member ‘llm_build_context::hparams’
@ -511,7 +569,6 @@ REFL_FIELD( n_feed_forward_w2 )
REFL_FIELD( i_feed_forward_w2 )
REFL_FIELD( n_k_quantized )
REFL_FIELD( n_fallback )
REFL_END
REFL_TYPE(llama_data_context)
@ -524,6 +581,7 @@ REFL_END
REFL_TYPE(llama_data_file_context)
REFL_FIELD(file)
REFL_END
#endif
template <typename T>
constexpr auto get_value_type_name(const T t) noexcept