mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-04-14 10:36:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

gguf-split : --merge now respects --dry-run option (#12681)

Browse Source 
* gguf-split now respects dry-run option

* removing trailing space
This commit is contained in:
nick huang 2025-04-04 22:09:12 +08:00 committed by GitHub
parent 94148ba330
commit 23106f94ea
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 15 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
examples/gguf-split/gguf-split.cpp
View File

@ -408,8 +408,6 @@ static void gguf_merge(const split_params & split_params) {
exit(EXIT_FAILURE);
}
std::ofstream fout(split_params.output.c_str(), std::ios::binary);
fout.exceptions(std::ofstream::failbit); // fail fast on write errors
auto * ctx_out = gguf_init_empty();
@ -453,7 +451,6 @@ static void gguf_merge(const split_params & split_params) {
gguf_free(ctx_gguf);
ggml_free(ctx_meta);
gguf_free(ctx_out);
fout.close();
exit(EXIT_FAILURE);
}
@ -466,7 +463,6 @@ static void gguf_merge(const split_params & split_params) {
gguf_free(ctx_gguf);
ggml_free(ctx_meta);
gguf_free(ctx_out);
fout.close();
exit(EXIT_FAILURE);
}
@ -479,7 +475,6 @@ static void gguf_merge(const split_params & split_params) {
gguf_free(ctx_gguf);
ggml_free(ctx_meta);
gguf_free(ctx_out);
fout.close();
exit(EXIT_FAILURE);
}
@ -500,9 +495,11 @@ static void gguf_merge(const split_params & split_params) {
fprintf(stderr, "\033[3Ddone\n");
}
// placeholder for the meta data
{
std::ofstream fout;
if (!split_params.dry_run) {
fout.open(split_params.output.c_str(), std::ios::binary);
fout.exceptions(std::ofstream::failbit); // fail fast on write errors
// placeholder for the meta data
auto meta_size = gguf_get_meta_size(ctx_out);
::zeros(fout, meta_size);
}
@ -518,7 +515,9 @@ static void gguf_merge(const split_params & split_params) {
ggml_free(ctx_metas[i]);
}
gguf_free(ctx_out);
fout.close();
if (!split_params.dry_run) {
fout.close();
}
exit(EXIT_FAILURE);
}
fprintf(stderr, "%s: writing tensors %s ...", __func__, split_path);
@ -540,10 +539,11 @@ static void gguf_merge(const split_params & split_params) {
auto offset = gguf_get_data_offset(ctx_gguf) + gguf_get_tensor_offset(ctx_gguf, i_tensor);
f_input.seekg(offset);
f_input.read((char *)read_data.data(), n_bytes);
// write tensor data + padding
fout.write((const char *)read_data.data(), n_bytes);
zeros(fout, GGML_PAD(n_bytes, GGUF_DEFAULT_ALIGNMENT) - n_bytes);
if (!split_params.dry_run) {
// write tensor data + padding
fout.write((const char *)read_data.data(), n_bytes);
zeros(fout, GGML_PAD(n_bytes, GGUF_DEFAULT_ALIGNMENT) - n_bytes);
}
}
gguf_free(ctx_gguf);
@ -552,16 +552,15 @@ static void gguf_merge(const split_params & split_params) {
fprintf(stderr, "\033[3Ddone\n");
}
{
if (!split_params.dry_run) {
// go back to beginning of file and write the updated metadata
fout.seekp(0);
std::vector<uint8_t> data(gguf_get_meta_size(ctx_out));
gguf_get_meta_data(ctx_out, data.data());
fout.write((const char *)data.data(), data.size());
fout.close();
gguf_free(ctx_out);
}
gguf_free(ctx_out);
fprintf(stderr, "%s: %s merged from %d split with %d tensors.\n",
__func__, split_params.output.c_str(), n_split, total_tensors);