vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

gguf : support big endian platform (#3552)

Browse source 
* check whether platform is 390x if yes->do not import immintrin.h

* support s390x big endian

* support --bigendian option for s390x
1. verified with baichuan7b-chat with float 16 on s390x
2. verified with baichuan7b-chat
3. verified with chinese-alpaca-2-13b-f16

* update format based on editor-config checker result

* Update convert-baichuan-hf-to-gguf.py

* 1. check in ggml.c if endianess is not match
2. update GGUF version
3. change get_pack_prefix to property
4. update information log

* always use "GGUF" as beginng of GGUF file

* Compare "GGUF" with file header char by char
1.  Set GGUF_MAGIC to "GGUF" string instead of int value
2. Compare "GGUF" char by char to ensure its byte order
3. Move bytes swap code from convert.py to gguf.py write_tensor_data

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Qin Yue Chen 2023-10-20 06:19:40 -05:00 • committed by GitHub
parent a0edf73bda
commit 8cf19d60dc
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
9 changed files with 84 additions and 49 deletions
Download patch file Download diff file Expand all files Collapse all files

19
ggml.c
View file

@ -20845,7 +20845,7 @@ struct gguf_kv {
};
struct gguf_header {
uint32_t magic;
char magic[4];
uint32_t version;
uint64_t n_tensors; // GGUFv2
uint64_t n_kv; // GGUFv2
@ -20915,7 +20915,7 @@ static bool gguf_fread_str_v1(FILE * file, struct gguf_str * p, size_t * offset)
struct gguf_context * gguf_init_empty(void) {
struct gguf_context * ctx = GGML_ALIGNED_MALLOC(sizeof(struct gguf_context));
ctx->header.magic = GGUF_MAGIC;
memcpy(ctx->header.magic, GGUF_MAGIC, sizeof(ctx->header.magic));
ctx->header.version = GGUF_VERSION;
ctx->header.n_tensors = 0;
ctx->header.n_kv = 0;
@ -20941,16 +20941,18 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
// offset from start of file
size_t offset = 0;
uint32_t magic = 0;
char magic[4];
// check the magic before making allocations
{
gguf_fread_el(file, &magic, sizeof(magic), &offset);
if (magic != GGUF_MAGIC) {
fprintf(stderr, "%s: invalid magic number %08x\n", __func__, magic);
fclose(file);
return NULL;
for (uint32_t i = 0; i < sizeof(magic); i++) {
if (magic[i] != GGUF_MAGIC[i]) {
fprintf(stderr, "%s: invalid magic characters %s.\n", __func__, magic);
fclose(file);
return NULL;
}
}
}
@ -20960,7 +20962,8 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
// read the header
{
ctx->header.magic = magic;
strncpy(ctx->header.magic, magic, 4);
ctx->kv = NULL;
ctx->infos = NULL;