tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 12:25:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

imatrix : support 3d tensors with MUL_MAT

Browse Source
This commit is contained in:
Francis Couture-Harpin
2025-06-23 11:50:54 -04:00
parent 1a9454a3d2
commit 43cd2b3eb5
2 changed files with 53 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

85
tools/imatrix/imatrix.cpp
View File

@ -4,6 +4,7 @@
#include "llama.h" #include "llama.h"
#include "gguf.h" #include "gguf.h"
#include <algorithm>
#include <chrono> #include <chrono>
#include <cmath> #include <cmath>
#include <cstdio> #include <cstdio>
@ -15,7 +16,6 @@
#include <fstream> #include <fstream>
#include <unordered_map> #include <unordered_map>
#include <map> #include <map>
#include <algorithm>
#if defined(_MSC_VER) #if defined(_MSC_VER)
#pragma warning(disable: 4244 4267) // possible loss of data #pragma warning(disable: 4244 4267) // possible loss of data
@ -124,14 +124,21 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
const char * data = is_host ? (const char *) src1->data : m_src1_data.data(); const char * data = is_host ? (const char *) src1->data : m_src1_data.data();
GGML_ASSERT(src1->nb[0] == ggml_element_size(src1)); GGML_ASSERT(src1->nb[0] == ggml_element_size(src1));
// TODO: 4d? (is that even used in practice?)
// the extra dimension would need to be stored somewhere to be reflected in the imatrix file
if (ggml_nrows(src1) != src1->ne[1] * src1->ne[2]) {
LOG_ERR("%s: tensor has more than 3 dimensions: %s", __func__, wname.c_str());
GGML_ASSERT(false);
}
// this has been adapted to the new format of storing merged experts in a single 3d tensor // this has been adapted to the new format of storing merged experts in a single 3d tensor
// ref: https://github.com/ggml-org/llama.cpp/pull/6387 // ref: https://github.com/ggml-org/llama.cpp/pull/6387
if (t->op == GGML_OP_MUL_MAT_ID) { if (t->op == GGML_OP_MUL_MAT_ID) {
// ids -> [n_experts_used, n_tokens] // ids -> [n_experts_used, n_tokens]
// src1 -> [cols, n_expert_used, n_tokens] // src1 -> [cols, n_expert_used, n_tokens]
const ggml_tensor * ids = t->src[2]; const ggml_tensor * ids = t->src[2];
const int n_as = src0->ne[2]; const int64_t n_as = src0->ne[2];
const int n_ids = ids->ne[0]; const int64_t n_ids = ids->ne[0];
// the top-k selected expert ids are stored in the ids tensor // the top-k selected expert ids are stored in the ids tensor
// for simplicity, always copy ids to host, because it is small // for simplicity, always copy ids to host, because it is small
@ -153,7 +160,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
e.counts.resize(n_as, 0); e.counts.resize(n_as, 0);
} }
else if (e.values.size() != (size_t)src1->ne[0]*n_as) { else if (e.values.size() != (size_t)src1->ne[0]*n_as) {
LOG_ERR("%s: inconsistent size for %s (%d vs %d)\n", __func__, wname.c_str(), (int)e.values.size(), (int)src1->ne[0]*n_as); LOG_ERR("%s: inconsistent size for %s (%d vs %d)\n", __func__, wname.c_str(), (int)e.values.size(), (int)(src1->ne[0]*n_as));
exit(1); //GGML_ABORT("fatal error"); exit(1); //GGML_ABORT("fatal error");
} }
else if (e.counts.size() != (size_t)n_as) { else if (e.counts.size() != (size_t)n_as) {
@ -162,11 +169,11 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
} }
LOG_DBGV(2, "%s[%d]: %32s, %s, %5d x %5d, %d\n", __func__, m_last_chunk, wname.c_str(), ggml_op_name(t->op), (int)src1->ne[0], (int)src1->ne[2], (int)src1->type); LOG_DBGV(2, "%s[%d]: %32s, %s, %5d x %5d, %d\n", __func__, m_last_chunk, wname.c_str(), ggml_op_name(t->op), (int)src1->ne[0], (int)src1->ne[2], (int)src1->type);
// loop over all possible experts, regardless if they are used or not in the batch // loop over all possible experts, regardless if they are used or not in the batch
for (int ex = 0; ex < n_as; ++ex) { for (int64_t ex = 0; ex < n_as; ++ex) {
size_t e_start = ex*src1->ne[0]; size_t e_start = ex*src1->ne[0];
for (int idx = 0; idx < n_ids; ++idx) { for (int64_t idx = 0; idx < n_ids; ++idx) {
for (int row = 0; row < (int)src1->ne[2]; ++row) { for (int64_t row = 0; row < src1->ne[2]; ++row) {
const int excur = *(const int32_t *) (m_ids.data() + row*ids->nb[1] + idx*ids->nb[0]); const int excur = *(const int32_t *) (m_ids.data() + row*ids->nb[1] + idx*ids->nb[0]);
GGML_ASSERT(excur >= 0 && excur < n_as); // sanity check GGML_ASSERT(excur >= 0 && excur < n_as); // sanity check
@ -179,7 +186,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
e.counts[ex]++; e.counts[ex]++;
for (int j = 0; j < (int)src1->ne[0]; ++j) { for (int64_t j = 0; j < src1->ne[0]; ++j) {
e.values[e_start + j] += x[j] * x[j]; e.values[e_start + j] += x[j] * x[j];
if (!std::isfinite((float)e.values[e_start + j])) { if (!std::isfinite((float)e.values[e_start + j])) {
LOG_ERR("%f detected in %s\n", (float)e.values[e_start + j], wname.c_str()); LOG_ERR("%f detected in %s\n", (float)e.values[e_start + j], wname.c_str());
@ -202,40 +209,48 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
} }
} else { } else {
auto & e = m_stats[wname]; auto & e = m_stats[wname];
const int64_t n_mat = src1->ne[2] * src1->ne[3];
if (e.values.empty()) { if (e.values.empty()) {
e.values.resize(src1->ne[0], 0); e.values.resize(src1->ne[0] * n_mat, 0);
e.counts.resize(1, 0); e.counts.resize(n_mat, 0);
} }
else if (e.values.size() != (size_t)src1->ne[0]) { else if (e.values.size() != (size_t)(src1->ne[0] * n_mat)) {
LOG_ERR("%s: inconsistent size for %s (%d vs %d)\n", __func__, wname.c_str(), (int)e.values.size(), (int)src1->ne[0]); LOG_ERR("%s: inconsistent size for %s (%d vs %d)\n", __func__, wname.c_str(), (int)e.values.size(), (int)(src1->ne[0] * n_mat));
exit(1); //GGML_ABORT("fatal error"); exit(1); //GGML_ABORT("fatal error");
} }
else if (e.counts.size() != 1) { else if (e.counts.size() != (size_t)n_mat) {
LOG_ERR("%s: inconsistent expert count for %s (%d vs %d)\n", __func__, wname.c_str(), (int)e.counts.size(), 1); LOG_ERR("%s: inconsistent expert count for %s (%d vs %d)\n", __func__, wname.c_str(), (int)e.counts.size(), (int)n_mat);
exit(1); //GGML_ABORT("fatal error"); exit(1); //GGML_ABORT("fatal error");
} }
LOG_DBGV(2, "%s[%d]: %32s, %s, %5d x %5d, %d\n", __func__, m_last_chunk, wname.c_str(), ggml_op_name(t->op), (int)src1->ne[0], (int)src1->ne[1], (int)src1->type); LOG_DBGV(2, "%s[%d]: %32s, %s, %5d x %5d x %5d, %d\n", __func__, m_last_chunk, wname.c_str(), ggml_op_name(t->op), (int)src1->ne[0], (int)src1->ne[1], (int)src1->ne[2], (int)src1->type);
// TODO: higher dimensions for (int64_t i3 = 0; i3 < src1->ne[3]; ++i3) {
for (int row = 0; row < (int)src1->ne[1]; ++row) { for (int64_t i2 = 0; i2 < src1->ne[2]; ++i2) {
const float * x = (const float *) (data + row * src1->nb[1]); const int64_t mat_id = i3 * src1->ne[2] + i2;
e.counts[0]++; const int64_t mat_start = mat_id * src1->ne[0];
for (int j = 0; j < (int)src1->ne[0]; ++j) {
e.values[j] += x[j] * x[j]; for (int64_t row = 0; row < src1->ne[1]; ++row) {
if (!std::isfinite((float)e.values[j])) { const float * x = (const float *) (data + row * src1->nb[1] + i2 * src1->nb[2] + i3 * src1->ne[3]);
LOG_ERR("%f detected in %s\n", (float)e.values[j], wname.c_str()); e.counts[mat_id]++;
exit(1); for (int64_t j = 0; j < src1->ne[0]; ++j) {
e.values[mat_start + j] += x[j] * x[j];
if (!std::isfinite((float)e.values[j])) {
LOG_ERR("%f detected in %s\n", (float)e.values[j], wname.c_str());
exit(1);
}
}
}
const int32_t n_chunk = e.counts[mat_id] / chunk_size;
if (n_chunk > m_last_chunk) {
const int32_t chunk_step = n_chunk - m_last_chunk;
m_last_chunk = n_chunk;
if ((m_last_chunk % m_params.n_out_freq) / chunk_step == 0) {
save_imatrix();
}
if (m_params.n_save_freq > 0 && (m_last_chunk % m_params.n_save_freq) / chunk_step == 0) {
save_imatrix(m_last_chunk);
}
} }
}
}
const int32_t n_chunk = e.counts[0] / chunk_size;
if (n_chunk > m_last_chunk) {
const int32_t chunk_step = n_chunk - m_last_chunk;
m_last_chunk = n_chunk;
if ((m_last_chunk % m_params.n_out_freq) / chunk_step == 0) {
save_imatrix();
}
if (m_params.n_save_freq > 0 && (m_last_chunk % m_params.n_save_freq) / chunk_step == 0) {
save_imatrix(m_last_chunk);
} }
} }
} }

4
tools/quantize/quantize.cpp
View File

@ -196,7 +196,9 @@ static int load_legacy_imatrix(const std::string & imatrix_file, std::vector<std
exit(1); exit(1);
} }
if (ncall > 0) { if (ncall > 0) {
for (auto& v : e) v /= ncall; for (auto & v : e) {
v /= ncall;
}
} }
if (getenv("LLAMA_TRACE")) { if (getenv("LLAMA_TRACE")) {