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

minor : clean-up

Browse source
This commit is contained in:
Georgi Gerganov 2024-05-10 15:31:05 +03:00
parent d9adb8832b
commit a1278f13da
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
2 changed files with 2 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml.c
View file

@ -13335,7 +13335,7 @@ static void ggml_compute_forward_soft_max_f32(
const float slope = (max_bias > 0.0f) ? h < n_head_log2 ? powf(m0, h + 1) : powf(m1, 2*(h - n_head_log2) + 1) : 1.0f; const float slope = (max_bias > 0.0f) ? h < n_head_log2 ? powf(m0, h + 1) : powf(m1, 2*(h - n_head_log2) + 1) : 1.0f;
float * sp = (float *)((char *) src0->data + i1*src0->nb[1]); float * sp = (float *)((char *) src0->data + i1*src0->nb[1]);
float * dp = (float *)((char *) dst->data + i1*dst->nb[1]); float * dp = (float *)((char *) dst->data + i1*dst->nb[1]);
// broadcast the mask across rows // broadcast the mask across rows
ggml_fp16_t * mp_f16 = src1 ? (ggml_fp16_t *)((char *) src1->data) + (i1%ne01)*ne00 : NULL; ggml_fp16_t * mp_f16 = src1 ? (ggml_fp16_t *)((char *) src1->data) + (i1%ne01)*ne00 : NULL;

5
llama.cpp
View file

@ -6403,7 +6403,7 @@ static struct ggml_tensor * llm_build_ffn(
llm_ffn_gate_type type_gate, llm_ffn_gate_type type_gate,
const llm_build_cb & cb, const llm_build_cb & cb,
int il) { int il) {
struct ggml_tensor * tmp = up ? ggml_mul_mat(ctx, up, cur): cur; struct ggml_tensor * tmp = up ? ggml_mul_mat(ctx, up, cur) : cur;
cb(tmp, "ffn_up", il); cb(tmp, "ffn_up", il);
if (up_b) { if (up_b) {
@ -8372,7 +8372,6 @@ struct llm_build_context {
// output layer norm // output layer norm
cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].layer_out_norm, model.layers[il].layer_out_norm_b, LLM_NORM, cb, il); cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].layer_out_norm, model.layers[il].layer_out_norm_b, LLM_NORM, cb, il);
// input for next layer // input for next layer
inpL = cur; inpL = cur;
} }
@ -12806,8 +12805,6 @@ static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab &
} }
} }
//GGML_ASSERT(vocab.special_add_eos != 1);
//TODO: Check this, why this tokenizer does not add at the end, why not leaving up to the `gguf` exporter?
if (add_special && vocab.special_add_eos == 1) { if (add_special && vocab.special_add_eos == 1) {
GGML_ASSERT(vocab.special_add_eos != -1); GGML_ASSERT(vocab.special_add_eos != -1);
output.push_back(vocab.special_eos_id); output.push_back(vocab.special_eos_id);