zhangyiss/gguf-tools
1
0
Fork 0
mirror of https://github.com/antirez/gguf-tools.git synced 2025-09-17 10:48:06 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add BF16 support and fix warnings

Browse Source 
This change updates the data type definitions to be the same as the
latest source code. Support for the bfloat16 data type is available
however it can't interpret the IQ quantization formats yet. Cleanup
of compiler warnings and other nits have been fixed, but behavioral
changes have been avoided, and no new features are as of yet added.
This commit is contained in:
Justine Tunney
2024-05-25 22:48:18 -07:00
parent 3e5c0a464d
commit ede59bb742
6 changed files with 201 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
gguflib.h
View File

@@ -27,16 +27,27 @@ enum gguf_tensor_type {
GGUF_TYPE_Q5_1 = 7,
GGUF_TYPE_Q8_0 = 8,
GGUF_TYPE_Q8_1 = 9,
// k-quantizations
GGUF_TYPE_Q2_K = 10,
GGUF_TYPE_Q3_K = 11,
GGUF_TYPE_Q4_K = 12,
GGUF_TYPE_Q5_K = 13,
GGUF_TYPE_Q6_K = 14,
GGUF_TYPE_Q8_K = 15,
GGUF_TYPE_I8,
GGUF_TYPE_I16,
GGUF_TYPE_I32,
GGUF_TYPE_IQ2_XXS = 16,
GGUF_TYPE_IQ2_XS = 17,
GGUF_TYPE_IQ3_XXS = 18,
GGUF_TYPE_IQ1_S = 19,
GGUF_TYPE_IQ4_NL = 20,
GGUF_TYPE_IQ3_S = 21,
GGUF_TYPE_IQ2_S = 22,
GGUF_TYPE_IQ4_XS = 23,
GGUF_TYPE_I8 = 24,
GGUF_TYPE_I16 = 25,
GGUF_TYPE_I32 = 26,
GGUF_TYPE_I64 = 27,
GGUF_TYPE_F64 = 28,
GGUF_TYPE_IQ1_M = 29,
GGUF_TYPE_BF16 = 30,
GGUF_TYPE_COUNT,
};
@@ -185,5 +196,6 @@ uint64_t gguf_get_alignment_padding(uint64_t alignment, uint64_t offset);
void gguf_skip_key_values_section(gguf_ctx *ctx);
float *gguf_tensor_to_float(gguf_tensor *tensor);
int16_t *gguf_tensor_to_f16(gguf_tensor *tensor);
int16_t *gguf_tensor_to_bf16(gguf_tensor *tensor);
#endif