Restore clip's cb() to its rightful glory - extract common debugging elements in llama (#17914)
* Extract common debugging functions; plug eval-callback and mtmd's MTMD_DEBUG_GRAPH with same functionality * Move to common * Remove unneeded header * Unlink from common * chore: update webui build output * Cleanup; properly pass params to mtmd without depending on common; factorize debug.cpp to use common debug code. * Revert change to webapp * Post-merge adjust * Apply suggestions from code review Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> * Apply code review changes * Remove changes to server-context * Remove mtmd.h include * Remove utility functions from header * Apply suggestions from code review Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> * Rename functions * Update tools/mtmd/clip.cpp Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> * Update tools/mtmd/clip.cpp Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> * Update tools/mtmd/clip.cpp Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com> --------- Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
This commit is contained in:
Piotr Wilkin (ilintar)
GitHub
parent
8fb7175576
commit
d98b548120
12 changed files with 259 additions and 396 deletions
|
|
@ -1,11 +1,9 @@
|
|||
#include "debug.h"
|
||||
#include "arg.h"
|
||||
#include "common.h"
|
||||
#include "log.h"
|
||||
#include "llama.h"
|
||||
#include "ggml.h"
|
||||
|
||||
#include <cmath>
|
||||
#include <cstdint>
|
||||
#include <cstdlib>
|
||||
#include <string>
|
||||
#include <vector>
|
||||
|
|
@ -13,7 +11,7 @@
|
|||
#include <fstream>
|
||||
#include <regex>
|
||||
|
||||
static void print_usage(int, char ** argv) {
|
||||
static void print_usage(int /*argc*/, char ** argv) {
|
||||
const std::string usage_template = R"(
|
||||
example usage:
|
||||
|
||||
|
|
@ -35,28 +33,6 @@ static void print_usage(int, char ** argv) {
|
|||
LOG("%s\n", usage.c_str());
|
||||
}
|
||||
|
||||
static bool ggml_debug(struct ggml_tensor * t, bool ask, void * user_data);
|
||||
|
||||
struct callback_data {
|
||||
std::vector<uint8_t> data;
|
||||
std::vector<std::regex> tensor_filters;
|
||||
|
||||
callback_data() = default;
|
||||
|
||||
callback_data(common_params & params, const std::vector<std::string> & filter_patterns) {
|
||||
for (const auto & pattern : filter_patterns) {
|
||||
try {
|
||||
std::string anchored_pattern = "^" + pattern;
|
||||
tensor_filters.emplace_back(anchored_pattern, std::regex::optimize);
|
||||
} catch (const std::regex_error & e) {
|
||||
throw std::runtime_error("Invalid regex pattern '" + pattern + "': " + e.what());
|
||||
}
|
||||
}
|
||||
params.cb_eval = ggml_debug;
|
||||
params.cb_eval_user_data = this;
|
||||
}
|
||||
};
|
||||
|
||||
static bool has_pooling(llama_context * ctx) {
|
||||
switch (llama_pooling_type(ctx)) {
|
||||
case LLAMA_POOLING_TYPE_NONE:
|
||||
|
|
@ -120,168 +96,6 @@ struct output_data {
|
|||
}
|
||||
};
|
||||
|
||||
static std::string ggml_ne_string(const ggml_tensor * t) {
|
||||
std::string str;
|
||||
for (int i = 0; i < GGML_MAX_DIMS; ++i) {
|
||||
str += std::to_string(t->ne[i]);
|
||||
if (i + 1 < GGML_MAX_DIMS) {
|
||||
str += ", ";
|
||||
}
|
||||
}
|
||||
return str;
|
||||
}
|
||||
|
||||
static inline float ggml_compute_bf16_to_fp32(ggml_bf16_t h) {
|
||||
union {
|
||||
float f;
|
||||
uint32_t i;
|
||||
} u;
|
||||
u.i = (uint32_t)h.bits << 16;
|
||||
return u.f;
|
||||
}
|
||||
|
||||
static float ggml_get_float_value(const uint8_t * data, ggml_type type,
|
||||
const size_t * nb, size_t i0, size_t i1, size_t i2, size_t i3) {
|
||||
size_t i = i3 * nb[3] + i2 * nb[2] + i1 * nb[1] + i0 * nb[0];
|
||||
switch (type) {
|
||||
case GGML_TYPE_F16:
|
||||
return ggml_fp16_to_fp32(*(const ggml_fp16_t *) &data[i]);
|
||||
case GGML_TYPE_F32:
|
||||
return *(const float *) &data[i];
|
||||
case GGML_TYPE_I64:
|
||||
return (float) *(const int64_t *) &data[i];
|
||||
case GGML_TYPE_I32:
|
||||
return (float) *(const int32_t *) &data[i];
|
||||
case GGML_TYPE_I16:
|
||||
return (float) *(const int16_t *) &data[i];
|
||||
case GGML_TYPE_I8:
|
||||
return (float) *(const int8_t *) &data[i];
|
||||
case GGML_TYPE_BF16:
|
||||
return ggml_compute_bf16_to_fp32(*(const ggml_bf16_t *) &data[i]);
|
||||
default:
|
||||
GGML_ABORT("fatal error");
|
||||
}
|
||||
}
|
||||
|
||||
static void ggml_print_tensor(uint8_t * data, ggml_type type, const int64_t * ne, const size_t * nb, int64_t n) {
|
||||
GGML_ASSERT(n > 0);
|
||||
float sum = 0;
|
||||
float sum_sq = 0.0;
|
||||
for (int64_t i3 = 0; i3 < ne[3]; i3++) {
|
||||
for (int64_t i2 = 0; i2 < ne[2]; i2++) {
|
||||
for (int64_t i1 = 0; i1 < ne[1]; i1++) {
|
||||
for (int64_t i0 = 0; i0 < ne[0]; i0++) {
|
||||
const float v = ggml_get_float_value(data, type, nb, i0, i1, i2, i3);
|
||||
sum += v;
|
||||
sum_sq += v * v;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
for (int64_t i3 = 0; i3 < ne[3]; i3++) {
|
||||
LOG_DBG(" [\n");
|
||||
for (int64_t i2 = 0; i2 < ne[2]; i2++) {
|
||||
if (i2 == n && ne[2] > 2*n) {
|
||||
LOG_DBG(" ..., \n");
|
||||
i2 = ne[2] - n;
|
||||
}
|
||||
LOG_DBG(" [\n");
|
||||
for (int64_t i1 = 0; i1 < ne[1]; i1++) {
|
||||
if (i1 == n && ne[1] > 2*n) {
|
||||
LOG_DBG(" ..., \n");
|
||||
i1 = ne[1] - n;
|
||||
}
|
||||
LOG_DBG(" [");
|
||||
for (int64_t i0 = 0; i0 < ne[0]; i0++) {
|
||||
if (i0 == n && ne[0] > 2*n) {
|
||||
LOG_DBG("..., ");
|
||||
i0 = ne[0] - n;
|
||||
}
|
||||
const float v = ggml_get_float_value(data, type, nb, i0, i1, i2, i3);
|
||||
LOG_DBG("%12.4f", v);
|
||||
if (i0 < ne[0] - 1) {
|
||||
LOG_DBG(", ");
|
||||
}
|
||||
}
|
||||
LOG_DBG("],\n");
|
||||
}
|
||||
LOG_DBG(" ],\n");
|
||||
}
|
||||
LOG_DBG(" ]\n");
|
||||
LOG_DBG(" sum = %f\n", sum);
|
||||
LOG_DBG(" sum_sq = %f\n", sum_sq);
|
||||
}
|
||||
|
||||
if (std::isnan(sum)) {
|
||||
LOG_ERR("encountered NaN - aborting\n");
|
||||
exit(0);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* GGML operations callback during the graph execution.
|
||||
*
|
||||
* @param t current tensor
|
||||
* @param ask when ask is true, the scheduler wants to know if we are interested in data from this tensor
|
||||
* if we return true, a follow-up call will be made with ask=false in which we can do the actual collection.
|
||||
* see ggml_backend_sched_eval_callback
|
||||
* @param user_data user data to pass at each call back
|
||||
* @return true to receive data or continue the graph, false otherwise
|
||||
*/
|
||||
static bool ggml_debug(struct ggml_tensor * t, bool ask, void * user_data) {
|
||||
auto * cb_data = (callback_data *) user_data;
|
||||
|
||||
const struct ggml_tensor * src0 = t->src[0];
|
||||
const struct ggml_tensor * src1 = t->src[1];
|
||||
|
||||
if (ask) {
|
||||
return true; // Always retrieve data
|
||||
}
|
||||
|
||||
bool matches_filter = cb_data->tensor_filters.empty();
|
||||
|
||||
if (!matches_filter) {
|
||||
for (const auto & filter : cb_data->tensor_filters) {
|
||||
if (std::regex_search(t->name, filter)) {
|
||||
matches_filter = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
char src1_str[128] = {0};
|
||||
if (src1) {
|
||||
snprintf(src1_str, sizeof(src1_str), "%s{%s}", src1->name, ggml_ne_string(src1).c_str());
|
||||
}
|
||||
|
||||
if (matches_filter) {
|
||||
LOG_DBG("%s: %24s = (%s) %10s(%s{%s}, %s}) = {%s}\n", __func__,
|
||||
t->name,
|
||||
ggml_type_name(t->type),
|
||||
ggml_op_desc(t),
|
||||
src0->name,
|
||||
ggml_ne_string(src0).c_str(),
|
||||
src1 ? src1_str : "",
|
||||
ggml_ne_string(t).c_str());
|
||||
}
|
||||
|
||||
const bool is_host = ggml_backend_buffer_is_host(t->buffer);
|
||||
|
||||
if (!is_host) {
|
||||
auto n_bytes = ggml_nbytes(t);
|
||||
cb_data->data.resize(n_bytes);
|
||||
ggml_backend_tensor_get(t, cb_data->data.data(), 0, n_bytes);
|
||||
}
|
||||
|
||||
if (!ggml_is_quantized(t->type) && matches_filter) {
|
||||
uint8_t * data = is_host ? (uint8_t *) t->data : cb_data->data.data();
|
||||
ggml_print_tensor(data, t->type, t->ne, t->nb, 3);
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
|
||||
static void save_output_data(const output_data & output, const std::string & model_name, const std::string & output_dir) {
|
||||
std::filesystem::create_directory(output_dir);
|
||||
auto base_path = std::filesystem::path{output_dir} / ("llamacpp-" + model_name + output.type_suffix);
|
||||
|
|
@ -408,7 +222,7 @@ int main(int argc, char ** argv) {
|
|||
llama_backend_init();
|
||||
llama_numa_init(params.numa);
|
||||
|
||||
callback_data cb_data(params, params.tensor_filter);
|
||||
base_callback_data cb_data(params, params.tensor_filter);
|
||||
|
||||
auto llama_init = common_init_from_params(params);
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue