feature: add remaining LoRA functions bindings to llama.cpp

Signed-off-by: deadprogram <[email protected]>

ROADMAP.md

@@ -163,6 +163,8 @@ This is a list of all functions exposed by `llama.cpp` and the current state of
 - [x] `llama_state_set_data`
 
 ### LoRA Functions
+- [x] `llama_adapter_get_alora_invocation_tokens`
+- [x] `llama_adapter_get_alora_n_invocation_tokens`
 - [x] `llama_adapter_lora_free`
 - [x] `llama_adapter_lora_init`
 - [x] `llama_adapter_meta_count`
@@ -223,8 +225,6 @@ Note that these functions are considered by `llama.cpp` to be experimental, and
 
 ## Functions in `llama.cpp` still needing wrappers
 
-- [ ] `llama_adapter_get_alora_invocation_tokens`
-- [ ] `llama_adapter_get_alora_n_invocation_tokens`
 - [ ] `llama_apply_adapter_cvec`
 - [ ] `llama_attach_threadpool`
 - [ ] `llama_detach_threadpool`

pkg/llama/lora.go

@@ -54,6 +54,12 @@ var (
 	// Remove all LoRA adapters from given context
 	// LLAMA_API void llama_clear_adapter_lora(struct llama_context * ctx);
 	clearAdapterLoraFunc ffi.Fun
+
+	// LLAMA_API uint64_t            llama_adapter_get_alora_n_invocation_tokens(const struct llama_adapter_lora * adapter);
+	adapterGetAloraNInvocationTokensFunc ffi.Fun
+
+	// LLAMA_API const llama_token * llama_adapter_get_alora_invocation_tokens  (const struct llama_adapter_lora * adapter);
+	adapterGetAloraInvocationTokensFunc ffi.Fun
 )
 
 var (
@@ -90,13 +96,23 @@ func loadLoraFuncs(lib ffi.Lib) error {
 	if setAdapterLoraFunc, err = lib.Prep("llama_set_adapter_lora", &ffi.TypeSint32, &ffi.TypePointer, &ffi.TypePointer, &ffi.TypeFloat); err != nil {
 		return loadError("llama_set_adapter_lora", err)
 	}
+
 	if rmAdapterLoraFunc, err = lib.Prep("llama_rm_adapter_lora", &ffi.TypeSint32, &ffi.TypePointer, &ffi.TypePointer); err != nil {
 		return loadError("llama_rm_adapter_lora", err)
 	}
+
 	if clearAdapterLoraFunc, err = lib.Prep("llama_clear_adapter_lora", &ffi.TypeVoid, &ffi.TypePointer); err != nil {
 		return loadError("llama_clear_adapter_lora", err)
 	}
 
+	if adapterGetAloraNInvocationTokensFunc, err = lib.Prep("llama_adapter_get_alora_n_invocation_tokens", &ffi.TypeUint64, &ffi.TypePointer); err != nil {
+		return loadError("llama_adapter_get_alora_n_invocation_tokens", err)
+	}
+
+	if adapterGetAloraInvocationTokensFunc, err = lib.Prep("llama_adapter_get_alora_invocation_tokens", &ffi.TypePointer, &ffi.TypePointer); err != nil {
+		return loadError("llama_adapter_get_alora_invocation_tokens", err)
+	}
+
 	return nil
 }
 
@@ -263,3 +279,29 @@ func ClearAdapterLora(ctx Context) {
 	}
 	clearAdapterLoraFunc.Call(nil, unsafe.Pointer(&ctx))
 }
+
+// AdapterGetAloraNInvocationTokens returns the number of invocation tokens for the adapter.
+func AdapterGetAloraNInvocationTokens(adapter AdapterLora) uint64 {
+	if adapter == 0 {
+		return 0
+	}
+	var result ffi.Arg
+	adapterGetAloraNInvocationTokensFunc.Call(unsafe.Pointer(&result), unsafe.Pointer(&adapter))
+	return uint64(result)
+}
+
+// AdapterGetAloraInvocationTokens returns a slice of invocation tokens for the adapter.
+func AdapterGetAloraInvocationTokens(adapter AdapterLora) []Token {
+	n := AdapterGetAloraNInvocationTokens(adapter)
+	if n == 0 {
+		return nil
+	}
+
+	var ptr *Token
+	adapterGetAloraInvocationTokensFunc.Call(unsafe.Pointer(&ptr), unsafe.Pointer(&adapter))
+	if ptr == nil {
+		return nil
+	}
+
+	return unsafe.Slice(ptr, n)
+}

pkg/llama/lora_test.go

@@ -244,3 +244,49 @@ func TestClearAdapterLora(t *testing.T) {
 	ClearAdapterLora(ctx)
 	t.Logf("ClearAdapterLora succeeded")
 }
+
+func TestAdapterGetAloraNInvocationTokens(t *testing.T) {
+	modelFile := testLoraModelFileName(t)
+	loraFile := testLoraAdaptorFileName(t)
+
+	testSetup(t)
+	defer testCleanup(t)
+
+	model, err := ModelLoadFromFile(modelFile, ModelDefaultParams())
+	if err != nil {
+		t.Fatalf("ModelLoadFromFile failed: %v", err)
+	}
+	defer ModelFree(model)
+
+	adapter, err := AdapterLoraInit(model, loraFile)
+	if err != nil {
+		t.Fatalf("AdapterLoraInit failed: %v", err)
+	}
+	defer AdapterLoraFree(adapter)
+
+	n := AdapterGetAloraNInvocationTokens(adapter)
+	t.Logf("AdapterGetAloraNInvocationTokens returned: %d", n)
+}
+
+func TestAdapterGetAloraInvocationTokens(t *testing.T) {
+	modelFile := testLoraModelFileName(t)
+	loraFile := testLoraAdaptorFileName(t)
+
+	testSetup(t)
+	defer testCleanup(t)
+
+	model, err := ModelLoadFromFile(modelFile, ModelDefaultParams())
+	if err != nil {
+		t.Fatalf("ModelLoadFromFile failed: %v", err)
+	}
+	defer ModelFree(model)
+
+	adapter, err := AdapterLoraInit(model, loraFile)
+	if err != nil {
+		t.Fatalf("AdapterLoraInit failed: %v", err)
+	}
+	defer AdapterLoraFree(adapter)
+
+	tokens := AdapterGetAloraInvocationTokens(adapter)
+	t.Logf("AdapterGetAloraInvocationTokens returned %d tokens", len(tokens))
+}