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

Remove split metadata when quantize model shards

Browse source
This commit is contained in:
z5269887 2024-04-11 00:39:15 +08:00
parent 65c64dc36f
commit 502d069b14
3 changed files with 18 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

11
ggml.c
View file

@ -21148,6 +21148,17 @@ static int gguf_get_or_add_key(struct gguf_context * ctx, const char * key) {
return n_kv;
}
void gguf_remove_key(struct gguf_context * ctx, const char * key) {
const int idx = gguf_find_key(ctx, key);
if (idx >= 0) {
const int n_kv = gguf_get_n_kv(ctx);
for (int i = idx; i < n_kv; ++i)
ctx->kv[i] = ctx->kv[i+1];
ctx->kv = realloc(ctx->kv, (n_kv - 1) * sizeof(struct gguf_kv));
ctx->header.n_kv--;
}
}
void gguf_set_val_u8(struct gguf_context * ctx, const char * key, uint8_t val) {
const int idx = gguf_get_or_add_key(ctx, key);

3
ggml.h
View file

@ -2289,6 +2289,9 @@ extern "C" {
GGML_API char * gguf_get_tensor_name (const struct gguf_context * ctx, int i);
GGML_API enum ggml_type gguf_get_tensor_type (const struct gguf_context * ctx, int i);
// removes key if it exists
GGML_API void gguf_remove_key(struct gguf_context * ctx, const char * key);
// overrides existing values or adds a new one
GGML_API void gguf_set_val_u8 (struct gguf_context * ctx, const char * key, uint8_t val);
GGML_API void gguf_set_val_i8 (struct gguf_context * ctx, const char * key, int8_t val);

4
llama.cpp
View file

@ -13514,6 +13514,10 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
gguf_set_kv (ctx_out, ml.meta);
gguf_set_val_u32(ctx_out, "general.quantization_version", GGML_QNT_VERSION);
gguf_set_val_u32(ctx_out, "general.file_type", ftype);
// Remove split metadata
gguf_remove_key(ctx_out, "split.no");
gguf_remove_key(ctx_out, "split.count");
gguf_remove_key(ctx_out, "split.tensors.count");
if (params->kv_overrides) {
const std::vector<llama_model_kv_override> & overrides = *(const std::vector<llama_model_kv_override> *)params->kv_overrides;