happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Minor cosmetic changes

This commit is contained in:
Ed Addario 2025-08-03 16:57:27 +01:00
parent 2117c4e54b
commit 90cb1be99d
No known key found for this signature in database
GPG Key ID: E7875815A3230993
1 changed files with 7 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
tools/imatrix/imatrix.cpp
View File

@ -311,52 +311,43 @@ static void compute_tensor_statistics(std::vector<tensor_statistics> & tstats) {
static void compute_layer_statistics(const std::vector<tensor_statistics> & tstats, static void compute_layer_statistics(const std::vector<tensor_statistics> & tstats,
std::map<int, float> & layer_cossim, std::map<int, float> & layer_cossim,
const std::unordered_map<std::string, Stats> & stats_map) { const std::unordered_map<std::string, Stats> & stats_map) {
struct layer_aggregation { struct layer_aggregation {
std::vector<float> curr_avg; std::vector<float> curr_avg;
std::vector<float> prev_avg; std::vector<float> prev_avg;
}; };
static const std::regex pattern(R"(blk\.(\d+)\.)"); static const std::regex pattern(R"(blk\.(\d+)\.)");
// index tensor stats by name for quick lookup
std::unordered_map<std::string, const tensor_statistics*> tidx; std::unordered_map<std::string, const tensor_statistics*> tidx;
tidx.reserve(tstats.size()); tidx.reserve(tstats.size());
for (const auto & ts : tstats) tidx[ts.tensor] = &ts; for (const auto & ts : tstats) tidx[ts.tensor] = &ts;
std::map<int, layer_aggregation> taggr;
// concatenate per-layer
std::map<int, layer_aggregation> taggr; // ordered by layer
for (const auto & ts : tstats) { for (const auto & ts : tstats) {
std::smatch match; std::smatch match;
if (!std::regex_search(ts.tensor, match, pattern)) continue; if (!std::regex_search(ts.tensor, match, pattern)) continue;
const int blk = std::stoi(match[1]); const int blk = std::stoi(match[1]);
if (blk <= 0) continue; if (blk <= 0) continue;
std::string prev_lyr(ts.tensor); std::string prev_lyr(ts.tensor);
prev_lyr.replace(match.position(1), match.length(1), std::to_string(blk-1)); prev_lyr.replace(match.position(1), match.length(1), std::to_string(blk-1));
if (auto it_prev = tidx.find(prev_lyr); it_prev == tidx.end()) continue; if (auto it_prev = tidx.find(prev_lyr); it_prev == tidx.end()) continue;
// use stored Stats to rebuild averages
const auto curr_avg = compute_tensor_averages(stats_map.at(ts.tensor)); const auto curr_avg = compute_tensor_averages(stats_map.at(ts.tensor));
const auto prev_avg = compute_tensor_averages(stats_map.at(prev_lyr)); const auto prev_avg = compute_tensor_averages(stats_map.at(prev_lyr));
if (curr_avg.empty() || prev_avg.empty() || curr_avg.size() != prev_avg.size()) continue; if (curr_avg.empty() || prev_avg.empty() || curr_avg.size() != prev_avg.size()) continue;
auto & [curr, prev] = taggr[blk]; auto & [curr, prev] = taggr[blk];
curr.insert(curr.end(), curr_avg.begin(), curr_avg.end()); curr.insert(curr.end(), curr_avg.begin(), curr_avg.end());
prev.insert(prev.end(), prev_avg.begin(), prev_avg.end()); prev.insert(prev.end(), prev_avg.begin(), prev_avg.end());
} }
// compute cosine per layer // compute the cosine similarity between consecutive layers
for (auto & kv : taggr) { for (auto & kv : taggr) {
const auto & curr = kv.second.curr_avg; const auto & curr = kv.second.curr_avg;
const auto & prev = kv.second.prev_avg; const auto & prev = kv.second.prev_avg;
if (curr.size() != prev.size() || curr.empty()) continue; if (curr.size() != prev.size() || curr.empty()) continue;
float dot_prod = 0.0, lyr1 = 0.0, lyr2 = 0.0; float dot_prod = 0.0, lyr1 = 0.0, lyr2 = 0.0;
for (size_t i = 0; i < curr.size(); ++i) { for (size_t i = 0; i < curr.size(); ++i) {
const double a = curr[i], b = prev[i]; float crr = curr[i], prv = prev[i];
dot_prod += a*b; dot_prod += crr * prv;
lyr1 += a*a; lyr1 += crr * crr;
lyr2 += b*b; lyr2 += prv * prv;
} }
float cossim = 0.0f; float cossim = 0.0f;
if (lyr1 > 0.0 && lyr2 > 0.0) cossim = dot_prod / (std::sqrt(lyr1) * std::sqrt(lyr2)); if (lyr1 > 0.0 && lyr2 > 0.0) cossim = dot_prod / (std::sqrt(lyr1) * std::sqrt(lyr2));