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

Fix conversion of unnormalized BF16->BF16 weights (#7843)

Browse source 
* add truncate_bf16

* truncate intermediate fp32 if converting bf16 to bf16

* fix masking in __compute_fp32_to_bf16

* np.int16 no longer used

* missing cast and additional numpy 2.x fix

* ggml-impl : do not flush bf16 subnormals to zero

* ggml : add reference fp32 to bf16 conversion

The fast version is no longer equivalent for all platforms
because of the handling of subnormal values.

* gguf-py : remove flush to zero for bf16 subnormals

* gguf-py : remove float32 truncation to bf16

Rounding achieves the same thing in the cases where this was used.

* missed prototype update in merge

* merge cleanup

---------

Co-authored-by: Francis Couture-Harpin <git@compilade.net>
This commit is contained in:
Sigbjørn Skjæret 2024-08-02 21:11:39 +02:00 • committed by GitHub
parent e09a800f9a
commit b72c20b85c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 20 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

11
ggml/src/ggml.c
View file

@ -480,9 +480,16 @@ void ggml_bf16_to_fp32_row(const ggml_bf16_t * x, float * y, int64_t n) {
}
}
void ggml_fp32_to_bf16_row_ref(const float * x, ggml_bf16_t * y, int64_t n) {
for (int i = 0; i < n; i++) {
y[i] = ggml_compute_fp32_to_bf16(x[i]);
}
}
void ggml_fp32_to_bf16_row(const float * x, ggml_bf16_t * y, int64_t n) {
int i = 0;
#if defined(__AVX512BF16__)
// subnormals are flushed to zero on this platform
for (; i + 32 <= n; i += 32) {
_mm512_storeu_si512(
(__m512i *)(y + i),
@ -962,7 +969,7 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
.is_quantized = false,
.to_float = (ggml_to_float_t) ggml_bf16_to_fp32_row,
.from_float = (ggml_from_float_t) ggml_fp32_to_bf16_row,
.from_float_ref = (ggml_from_float_t) ggml_fp32_to_bf16_row,
.from_float_ref = (ggml_from_float_t) ggml_fp32_to_bf16_row_ref,
.vec_dot = (ggml_vec_dot_t) ggml_vec_dot_bf16,
.vec_dot_type = GGML_TYPE_BF16,
.nrows = 1,
@ -20650,7 +20657,7 @@ size_t ggml_quantize_chunk(
case GGML_TYPE_BF16:
{
size_t elemsize = sizeof(ggml_bf16_t);
ggml_fp32_to_bf16_row(src + start, (ggml_bf16_t *)dst + start, n);
ggml_fp32_to_bf16_row_ref(src + start, (ggml_bf16_t *)dst + start, n);
result = n * elemsize;
} break;
case GGML_TYPE_F32: