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

Add compatibility with #801

Browse source
This commit is contained in:
Slaren 2023-04-10 21:52:10 +02:00
parent 87c518bb3d
commit c920f00136
2 changed files with 10 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

3
examples/common.cpp
View file

@ -145,6 +145,7 @@ bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
break; break;
} }
params.lora_adapter = argv[i]; params.lora_adapter = argv[i];
params.use_mmap = false;
} else if (arg == "-i" || arg == "--interactive") { } else if (arg == "-i" || arg == "--interactive") {
params.interactive = true; params.interactive = true;
} else if (arg == "--embedding") { } else if (arg == "--embedding") {
@ -248,7 +249,7 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
} }
fprintf(stderr, " --mtest compute maximum memory usage\n"); fprintf(stderr, " --mtest compute maximum memory usage\n");
fprintf(stderr, " --verbose-prompt print prompt before generation\n"); fprintf(stderr, " --verbose-prompt print prompt before generation\n");
fprintf(stderr, " --lora FNAME apply LoRA adapter\n"); fprintf(stderr, " --lora FNAME apply LoRA adapter (implies --no-mmap)\n");
fprintf(stderr, " -m FNAME, --model FNAME\n"); fprintf(stderr, " -m FNAME, --model FNAME\n");
fprintf(stderr, " model path (default: %s)\n", params.model.c_str()); fprintf(stderr, " model path (default: %s)\n", params.model.c_str());
fprintf(stderr, "\n"); fprintf(stderr, "\n");

10
llama.cpp
View file

@ -1808,6 +1808,12 @@ int llama_apply_lora_from_file(struct llama_context * ctx, const char * path_lor
ggml_context* lora_ctx = ggml_init(params); ggml_context* lora_ctx = ggml_init(params);
std::unordered_map<std::string, struct ggml_tensor *> lora_tensors; std::unordered_map<std::string, struct ggml_tensor *> lora_tensors;
// create a name -> tensor map of the model to accelerate lookups
std::unordered_map<std::string, struct ggml_tensor*> model_tensors;
for (auto & kv: model.tensors_by_name) {
model_tensors.insert(kv);
}
fprintf(stderr, "%s: ", __func__); fprintf(stderr, "%s: ", __func__);
// read tensors and apply // read tensors and apply
@ -1847,7 +1853,7 @@ int llama_apply_lora_from_file(struct llama_context * ctx, const char * path_lor
base_name.erase(pos); base_name.erase(pos);
// fprintf(stderr, "%s: %s => %s (lora type %s) ", __func__, name.c_str(),base_name.c_str(), lora_type.c_str()); // fprintf(stderr, "%s: %s => %s (lora type %s) ", __func__, name.c_str(),base_name.c_str(), lora_type.c_str());
if (model.tensors.find(base_name.data()) == model.tensors.end()) { if (model_tensors.find(base_name.data()) == model_tensors.end()) {
fprintf(stderr, "%s: unknown tensor '%s' in lora adapter\n", __func__, name.data()); fprintf(stderr, "%s: unknown tensor '%s' in lora adapter\n", __func__, name.data());
return 1; return 1;
} }
@ -1886,7 +1892,7 @@ int llama_apply_lora_from_file(struct llama_context * ctx, const char * path_lor
if (lora_tensors.find(base_name + ".loraA") != lora_tensors.end() && if (lora_tensors.find(base_name + ".loraA") != lora_tensors.end() &&
lora_tensors.find(base_name + ".loraB") != lora_tensors.end()) { lora_tensors.find(base_name + ".loraB") != lora_tensors.end()) {
ggml_tensor * tensor = model.tensors[base_name]; ggml_tensor * tensor = model_tensors[base_name];
ggml_tensor * loraA = lora_tensors[base_name + ".loraA"]; ggml_tensor * loraA = lora_tensors[base_name + ".loraA"];
ggml_tensor * loraB = lora_tensors[base_name + ".loraB"]; ggml_tensor * loraB = lora_tensors[base_name + ".loraB"];