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

Refactor last budget pass

This commit is contained in:
Ed Addario 2025-09-21 13:43:09 +01:00
parent c466c53808
commit b433fd9547
No known key found for this signature in database
GPG Key ID: E7875815A3230993
1 changed files with 3 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
src/llama-quant.cpp
View File

@ -1433,19 +1433,16 @@ static std::unordered_map<std::string, ggml_type> target_bpw_type(
double best_ratio = -1.0;
size_t best_delta = 0;
for (int i = 0; i < (int)all.size(); ++i) {
const auto & ti = all[i];
if (ti.choice >= (int)ti.candidate.size() - 1) {
continue;
}
const auto &ti = all[i];
int j = ti.choice + 1;
// skip same-bytes entries
while (j < (int)ti.candidate.size() && ti.candidate[j].bytes == ti.candidate[ti.choice].bytes) { ++j; }
if (j >= (int)ti.candidate.size()) { continue; }
size_t delta = ti.candidate[j].bytes - ti.candidate[ti.choice].bytes;
if (cur_bytes + delta > budget_bytes) { continue; }
double err_gain = std::max(0.0, (double)ti.candidate[ti.choice].error - (double)ti.candidate[j].error);
double err_gain = std::max(0.0, ti.candidate[ti.choice].error - ti.candidate[j].error);
double ratio = err_gain / (double)(delta * 8);
if (ratio > best_ratio + epsilon || (std::abs(ratio - best_ratio) <= epsilon && delta < best_delta)) {
best_ratio = ratio;
@ -1454,7 +1451,6 @@ static std::unordered_map<std::string, ggml_type> target_bpw_type(
best_j = j;
}
}
if (best_i < 0) { break; }
all[best_i].choice = best_j;
cur_bytes += best_delta;