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

tabs to spaces

Browse source
This commit is contained in:
Andrew Godfrey 2023-10-29 19:14:27 -07:00
parent c5c54d1057
commit 998a548a30
4 changed files with 47 additions and 47 deletions
Download patch file Download diff file Expand all files Collapse all files

2
common/train.h
View file

@ -44,7 +44,7 @@ struct train_params_common {
int n_batch; int n_batch;
int n_gradient_accumulation; int n_gradient_accumulation;
int n_epochs; int n_epochs;
int n_gpu_layers; int n_gpu_layers;
bool custom_n_ctx; bool custom_n_ctx;

18
examples/finetune/finetune.cpp
View file

@ -1459,16 +1459,16 @@ static bool train_params_parse(int argc, char ** argv, struct train_params * par
} }
params->n_rank_w3 = std::stoi(argv[i]); params->n_rank_w3 = std::stoi(argv[i]);
params->custom_n_rank_w3 = true; params->custom_n_rank_w3 = true;
} else if (arg == "--gpu-layers" || arg == "-ngl" || arg == "--n-gpu-layers") { } else if (arg == "--gpu-layers" || arg == "-ngl" || arg == "--n-gpu-layers") {
if (++i >= argc) { if (++i >= argc) {
invalid_param = true; invalid_param = true;
break; break;
} }
#ifdef LLAMA_SUPPORTS_GPU_OFFLOAD #ifdef LLAMA_SUPPORTS_GPU_OFFLOAD
params->common.n_gpu_layers = std::stoi(argv[i]); params->common.n_gpu_layers = std::stoi(argv[i]);
#else #else
fprintf(stderr, "warning: not compiled with GPU offload support, --n-gpu-layers option will be ignored\n"); fprintf(stderr, "warning: not compiled with GPU offload support, --n-gpu-layers option will be ignored\n");
fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n"); fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n");
#endif #endif
} else { } else {
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str()); fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
@ -1556,7 +1556,7 @@ int main(int argc, char ** argv) {
srand(params.common.seed); srand(params.common.seed);
struct llama_model_params llama_mparams = llama_model_default_params(); struct llama_model_params llama_mparams = llama_model_default_params();
llama_mparams.n_gpu_layers = params.common.n_gpu_layers; llama_mparams.n_gpu_layers = params.common.n_gpu_layers;
llama_mparams.vocab_only = false; llama_mparams.vocab_only = false;
printf("%s: model base = '%s'\n", __func__, params.fn_model_base); printf("%s: model base = '%s'\n", __func__, params.fn_model_base);

2
examples/finetune/finetune.sh
View file

@ -24,7 +24,7 @@ done
$DEBUGGER $EXE \ $DEBUGGER $EXE \
--model-base $MODEL \ --model-base $MODEL \
$GPUARG \ $GPUARG \
--checkpoint-in chk-ol3b-shakespeare-LATEST.gguf \ --checkpoint-in chk-ol3b-shakespeare-LATEST.gguf \
--checkpoint-out chk-ol3b-shakespeare-ITERATION.gguf \ --checkpoint-out chk-ol3b-shakespeare-ITERATION.gguf \
--lora-out lora-ol3b-shakespeare-ITERATION.bin \ --lora-out lora-ol3b-shakespeare-ITERATION.bin \

72
ggml.c
View file

@ -1276,7 +1276,7 @@ static void quantize_row_q8_0(const float * restrict x, void * restrict vy, int
__riscv_vse8_v_i8m1(y[i].qs , vs, vl); __riscv_vse8_v_i8m1(y[i].qs , vs, vl);
} }
#else #else
UNUSED(nb); UNUSED(nb);
// scalar // scalar
quantize_row_q8_0_reference(x, y, k); quantize_row_q8_0_reference(x, y, k);
#endif #endif
@ -1530,7 +1530,7 @@ static void quantize_row_q8_1(const float * restrict x, void * restrict vy, int
y[i].s = sum*d; y[i].s = sum*d;
} }
#else #else
UNUSED(nb); UNUSED(nb);
// scalar // scalar
quantize_row_q8_1_reference(x, y, k); quantize_row_q8_1_reference(x, y, k);
#endif #endif
@ -9359,13 +9359,13 @@ static void ggml_compute_forward_add_f16_f32(
GGML_ASSERT(src0->type == GGML_TYPE_F16); GGML_ASSERT(src0->type == GGML_TYPE_F16);
GGML_ASSERT(src1->type == GGML_TYPE_F32); GGML_ASSERT(src1->type == GGML_TYPE_F32);
if (dst->type == GGML_TYPE_F32) { if (dst->type == GGML_TYPE_F32) {
GGML_ASSERT( nb0 == sizeof(float)); GGML_ASSERT( nb0 == sizeof(float));
} }
else { else {
GGML_ASSERT(dst->type == GGML_TYPE_F16); GGML_ASSERT(dst->type == GGML_TYPE_F16);
GGML_ASSERT( nb0 == sizeof(ggml_fp16_t)); GGML_ASSERT( nb0 == sizeof(ggml_fp16_t));
} }
GGML_ASSERT(nb00 == sizeof(ggml_fp16_t)); GGML_ASSERT(nb00 == sizeof(ggml_fp16_t));
@ -9377,37 +9377,37 @@ static void ggml_compute_forward_add_f16_f32(
const int ir1 = MIN(ir0 + dr, nr); const int ir1 = MIN(ir0 + dr, nr);
if (nb10 == sizeof(float)) { if (nb10 == sizeof(float)) {
if (dst->type == GGML_TYPE_F16) { if (dst->type == GGML_TYPE_F16) {
for (int ir = ir0; ir < ir1; ++ir) { for (int ir = ir0; ir < ir1; ++ir) {
// src0, src1 and dst are same shape => same indices // src0, src1 and dst are same shape => same indices
const int i3 = ir/(ne2*ne1); const int i3 = ir/(ne2*ne1);
const int i2 = (ir - i3*ne2*ne1)/ne1; const int i2 = (ir - i3*ne2*ne1)/ne1;
const int i1 = (ir - i3*ne2*ne1 - i2*ne1); const int i1 = (ir - i3*ne2*ne1 - i2*ne1);
ggml_fp16_t * dst_ptr = (ggml_fp16_t *) ((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1); ggml_fp16_t * dst_ptr = (ggml_fp16_t *) ((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1);
ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01); ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01);
float * src1_ptr = (float *) ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11); float * src1_ptr = (float *) ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11);
for (int i = 0; i < ne0; i++) { for (int i = 0; i < ne0; i++) {
dst_ptr[i] = GGML_FP32_TO_FP16(GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i]); dst_ptr[i] = GGML_FP32_TO_FP16(GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i]);
} }
} }
} else { } else {
for (int ir = ir0; ir < ir1; ++ir) { for (int ir = ir0; ir < ir1; ++ir) {
// src0, src1 and dst are same shape => same indices // src0, src1 and dst are same shape => same indices
const int i3 = ir/(ne2*ne1); const int i3 = ir/(ne2*ne1);
const int i2 = (ir - i3*ne2*ne1)/ne1; const int i2 = (ir - i3*ne2*ne1)/ne1;
const int i1 = (ir - i3*ne2*ne1 - i2*ne1); const int i1 = (ir - i3*ne2*ne1 - i2*ne1);
float * dst_ptr = (float *) ((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1); float * dst_ptr = (float *) ((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1);
ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01); ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01);
float * src1_ptr = (float *) ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11); float * src1_ptr = (float *) ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11);
for (int i = 0; i < ne0; i++) { for (int i = 0; i < ne0; i++) {
dst_ptr[i] = GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i]; dst_ptr[i] = GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i];
} }
} }
} }
} }
else { else {
// src1 is not contiguous // src1 is not contiguous