zhangyiss/gguf-tools
1
0
Fork 0
mirror of https://github.com/antirez/gguf-tools.git synced 2025-09-18 12:48:08 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Introduce --diffable flag

Browse Source 
Sometimes it's useful to get an overview of how tensors changes when
using different quantization formats. For example:

  diff -u <(gguf-tools show --diffable ggml-model-bf16.gguf) \
          <(gguf-tools show --diffable ggml-model-Q6_K.gguf) | less

Is now able to produces nice clean output. Without this change, every
line would have been different due to the file offsets and byte sizes
which means `diff -u` would produce one gigantic unreadable chunk.
This commit is contained in:
Justine Tunney
2024-05-26 00:23:41 -07:00
parent ede59bb742
commit 6deab767f9
1 changed files with 16 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
gguf-tools.c
View File

@@ -14,6 +14,7 @@
/* Global options that can could be used for all the subcommands. */ /* Global options that can could be used for all the subcommands. */
struct { struct {
int verbose; // --verbose option int verbose; // --verbose option
int diffable; // --diffable option
} Opt = {0}; } Opt = {0};
/* ========================== Utility functions ============================ */ /* ========================== Utility functions ============================ */
@@ -167,16 +168,20 @@ void gguf_tools_show(const char *filename) {
gguf_tensor tensor; gguf_tensor tensor;
uint64_t params = 0; uint64_t params = 0;
while (gguf_get_tensor(ctx,&tensor)) { while (gguf_get_tensor(ctx,&tensor)) {
printf("%s tensor %.*s @%" PRIu64 ", %" PRIu64 " weights, dims ", printf("%s tensor %.*s",
gguf_get_tensor_type_name(tensor.type), gguf_get_tensor_type_name(tensor.type),
(int)tensor.namelen, (int)tensor.namelen,
tensor.name, tensor.name);
tensor.offset, if (!Opt.diffable)
tensor.num_weights); printf(" @%" PRIu64, tensor.offset);
printf(", %" PRIu64 " weights, dims ", tensor.num_weights);
for (uint32_t j = 0; j < tensor.ndim; j++) { for (uint32_t j = 0; j < tensor.ndim; j++) {
printf("%s%" PRIu64 "",(j == 0) ? "[" : ",", tensor.dim[j]); printf("%s%" PRIu64 "",(j == 0) ? "[" : ",", tensor.dim[j]);
} }
printf("], %" PRIu64 " bytes\n", tensor.bsize); printf("]");
if (!Opt.diffable)
printf(", %" PRIu64 " bytes", tensor.bsize);
printf("\n");
params += tensor.num_weights; params += tensor.num_weights;
} }
@@ -507,6 +512,7 @@ void gguf_tools_usage(const char *progname) {
" split-mixtral <ids...> mixtral.gguf out.gguf -- extract expert.\n" " split-mixtral <ids...> mixtral.gguf out.gguf -- extract expert.\n"
"Options:\n" "Options:\n"
" --verbose :With 'show', print full arrays (e.g. token lists)\n" " --verbose :With 'show', print full arrays (e.g. token lists)\n"
" --diffable :Don't show tensor file offsets and sizes\n"
"Example:\n" "Example:\n"
" split-mixtral 65230776370407150546470161412165 mixtral.gguf out.gguf\n" " split-mixtral 65230776370407150546470161412165 mixtral.gguf out.gguf\n"
, progname); , progname);
@@ -527,6 +533,11 @@ int main(int argc, char **argv) {
argc--; argc--;
Opt.verbose = 1; Opt.verbose = 1;
} }
if (!strcmp(argv[j],"--diffable")) {
argv[j] = NULL;
argc--;
Opt.diffable = 1;
}
} }
/* Strip empty elements. */ /* Strip empty elements. */