root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-27 03:44:35 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

add sanity check

Browse Source
This commit is contained in:
ngxson 2024-07-08 21:36:35 +02:00
parent 0e16188985
commit 6c617e20ef
1 changed files with 12 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
src/llama.cpp
View File

@ -18521,7 +18521,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
static void llama_lora_adapter_init_internal(struct llama_model * model, const char * path_lora, struct llama_lora_adapter & adapter) { static void llama_lora_adapter_init_internal(struct llama_model * model, const char * path_lora, struct llama_lora_adapter & adapter) {
static const int n_inp_tensors = 5; // see llama_model static const int n_inp_tensors = 5; // see llama_model
static const int n_out_tensors = 5; // see llama_model static const int n_out_tensors = 5; // see llama_model
LLAMA_LOG_INFO("%s: applying lora adapter from '%s' - please wait ...\n", __func__, path_lora); LLAMA_LOG_INFO("%s: applying lora adapter from '%s' ...\n", __func__, path_lora);
ggml_context * ctx = nullptr; ggml_context * ctx = nullptr;
struct gguf_init_params meta_gguf_params = { struct gguf_init_params meta_gguf_params = {
@ -18530,8 +18530,7 @@ static void llama_lora_adapter_init_internal(struct llama_model * model, const c
}; };
struct gguf_context * ctx_gguf = gguf_init_from_file(path_lora, meta_gguf_params); struct gguf_context * ctx_gguf = gguf_init_from_file(path_lora, meta_gguf_params);
if (!ctx_gguf) { if (!ctx_gguf) {
LLAMA_LOG_ERROR("%s: failed to load lora adapter file from %s\n", __func__, path_lora); throw std::runtime_error("failed to load lora adapter file from " + std::string(path_lora));
throw std::exception();
} }
// check metadata // check metadata
@ -18631,11 +18630,17 @@ static void llama_lora_adapter_init_internal(struct llama_model * model, const c
if (!model_tensor) { if (!model_tensor) {
gguf_free(ctx_gguf); gguf_free(ctx_gguf);
ggml_free(ctx); ggml_free(ctx);
throw std::runtime_error("LoRA tensor '" + name + "' does not exist in base model\n"); throw std::runtime_error("LoRA tensor '" + name + "' does not exist in base model");
} }
struct ggml_context * dev_ctx = ctx_map.at(ggml_backend_buffer_get_type(model_tensor->buffer)); struct ggml_context * dev_ctx = ctx_map.at(ggml_backend_buffer_get_type(model_tensor->buffer));
// TODO: validate tensor shape // validate tensor shape
// LLAMA_LOG_INFO("%s %p %p\n", cname, w.a, w.b); if (model_tensor->ne[0] != w.a->ne[0] || model_tensor->ne[1] != w.b->ne[1]) {
throw std::runtime_error("tensor '" + name + "' has incorrect shape");
}
if (w.a->ne[1] != w.b->ne[0]) {
throw std::runtime_error("lora_a tensor is not transposed (hint: adapter from \"finetune\" example is no longer supported)");
}
// save tensor to adapter
struct ggml_tensor * tensor_a = ggml_dup_tensor(dev_ctx, w.a); struct ggml_tensor * tensor_a = ggml_dup_tensor(dev_ctx, w.a);
struct ggml_tensor * tensor_b = ggml_dup_tensor(dev_ctx, w.b); struct ggml_tensor * tensor_b = ggml_dup_tensor(dev_ctx, w.b);
ggml_set_name(tensor_a, w.a->name); ggml_set_name(tensor_a, w.a->name);