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

More changes

Browse source
This commit is contained in:
Howard Su 2023-07-17 09:47:31 +08:00
parent f607bd1217
commit 2c9385289e
1 changed files with 65 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

74
llama.cpp
View file

@ -191,9 +191,7 @@ struct llama_layer {
struct ggml_tensor * w1; struct ggml_tensor * w1;
struct ggml_tensor * w2; struct ggml_tensor * w2;
struct ggml_tensor * w3; struct ggml_tensor * w3;
};
struct llama_lora_layers {
// LoRA optional // LoRA optional
struct ggml_tensor * wq_a; struct ggml_tensor * wq_a;
struct ggml_tensor * wq_b; struct ggml_tensor * wq_b;
@ -1375,12 +1373,37 @@ static bool llama_eval_internal(
// self-attention // self-attention
{ {
struct ggml_tensor * wq = model.layers[il].wq;
struct ggml_tensor * wk = model.layers[il].wk;
if (model.layers[il].wq_a != nullptr) {
// apply lora
ggml_tensor * BA = ggml_mul_mat(ctx0, model.layers[il].wq_a, model.layers[il].wq_b);
offload_func(BA);
ggml_set_name(BA, "BA");
#if 0
if (scaling != 1.0f) {
ggml_tensor * scale_tensor = ggml_new_f32(lora_ctx, scaling);
ggml_set_name(scale_tensor, "scale_tensor");
BA = ggml_scale_inplace(lora_ctx, BA, scale_tensor);
offload_func(BA);
ggml_set_name(BA, "BA_scaled");
}
#endif
wq = ggml_add(ctx0, wq, BA);
offload_func(wq);
ggml_set_name(wq, "lora_wq");
}
// compute Q and K and RoPE them // compute Q and K and RoPE them
struct ggml_tensor * tmpk = ggml_mul_mat(ctx0, model.layers[il].wk, cur); struct ggml_tensor * tmpk = ggml_mul_mat(ctx0, wk, cur);
offload_func_kq(tmpk); offload_func_kq(tmpk);
ggml_set_name(tmpk, "tmpk"); ggml_set_name(tmpk, "tmpk");
struct ggml_tensor * tmpq = ggml_mul_mat(ctx0, model.layers[il].wq, cur); struct ggml_tensor * tmpq = ggml_mul_mat(ctx0, wq, cur);
offload_func_kq(tmpq); offload_func_kq(tmpq);
ggml_set_name(tmpq, "tmpq"); ggml_set_name(tmpq, "tmpq");
@ -1395,8 +1418,30 @@ static bool llama_eval_internal(
// store key and value to memory // store key and value to memory
{ {
// compute the transposed [N, n_embd] V matrix // compute the transposed [N, n_embd] V matrix
struct ggml_tensor * wv = model.layers[il].wv;
if (model.layers[il].wv_a != nullptr) {
// apply lora
ggml_tensor * BA = ggml_mul_mat(ctx0, model.layers[il].wv_a, model.layers[il].wv_b);
offload_func(BA);
ggml_set_name(BA, "BA");
struct ggml_tensor * tmpv = ggml_mul_mat(ctx0, model.layers[il].wv, cur); #if 0
if (scaling != 1.0f) {
ggml_tensor * scale_tensor = ggml_new_f32(lora_ctx, scaling);
ggml_set_name(scale_tensor, "scale_tensor");
BA = ggml_scale_inplace(lora_ctx, BA, scale_tensor);
offload_func(BA);
ggml_set_name(BA, "BA_scaled");
}
#endif
wv = ggml_add(ctx0, wv, BA);
offload_func(wv);
ggml_set_name(wv, "lora_wv");
}
struct ggml_tensor * tmpv = ggml_mul_mat(ctx0, wv, cur);
offload_func_v(tmpv); offload_func_v(tmpv);
ggml_set_name(tmpv, "tmpv"); ggml_set_name(tmpv, "tmpv");
@ -3536,8 +3581,7 @@ const std::vector<std::pair<std::string, struct ggml_tensor *>>& llama_internal_
} }
// finetune related code // finetune related code
int llama_enable_finetune(struct llama_context * ctx, enum llama_finetune_type flags, int n_lora) { int llama_enable_finetune(struct llama_model * model, enum llama_finetune_type flags, int n_lora) {
auto model = &ctx->model;
const auto& hparams = model->hparams; const auto& hparams = model->hparams;
const uint32_t n_layer = hparams.n_layer; const uint32_t n_layer = hparams.n_layer;
@ -3565,20 +3609,32 @@ int llama_enable_finetune(struct llama_context * ctx, enum llama_finetune_type f
} else if (flags & LLAMA_FINETUNE_LORA) { } else if (flags & LLAMA_FINETUNE_LORA) {
// create AB tensor if they are not present // create AB tensor if they are not present
for (uint32_t i = 0; i < n_layer; ++i) { for (uint32_t i = 0; i < n_layer; ++i) {
llama_lora_layers layer = {0}; auto & layer = model->layers[i];
if (flags & LLAMA_FINETUNE_LORA_Q) { if (flags & LLAMA_FINETUNE_LORA_Q) {
if (layer.wq_a == nullptr || layer.wq_b == nullptr) { if (layer.wq_a == nullptr || layer.wq_b == nullptr) {
layer.wq_a = ggml_new_tensor_2d(ctx0, GGML_TYPE_F16, n_lora, n_embd); layer.wq_a = ggml_new_tensor_2d(ctx0, GGML_TYPE_F16, n_lora, n_embd);
layer.wq_b = ggml_new_tensor_2d(ctx0, GGML_TYPE_F16, n_embd, n_lora); layer.wq_b = ggml_new_tensor_2d(ctx0, GGML_TYPE_F16, n_embd, n_lora);
// initialize // initialize
// offload
} }
ggml_set_param(ctx0, layer.wq_a); ggml_set_param(ctx0, layer.wq_a);
ggml_set_param(ctx0, layer.wq_b); ggml_set_param(ctx0, layer.wq_b);
} }
if (flags & LLAMA_FINETUNE_LORA_Q) { if (flags & LLAMA_FINETUNE_LORA_V) {
if (layer.wv_a == nullptr || layer.wv_b == nullptr) {
layer.wv_a = ggml_new_tensor_2d(ctx0, GGML_TYPE_F16, n_lora, n_embd);
layer.wv_b = ggml_new_tensor_2d(ctx0, GGML_TYPE_F16, n_embd, n_lora);
// initialize
// offload
}
ggml_set_param(ctx0, layer.wv_a);
ggml_set_param(ctx0, layer.wv_b);
} }
} }
} }