Factor out model loading and provide a way to stop generation. (#2002)

Summary:
Pull Request resolved: #2002

.

Reviewed By: manuelcandales

Differential Revision: D53848385

fbshipit-source-id: 6bab78da0e8f484966e153007bb4bd86f98b4e76

Author: shoumikhin

examples/models/llama2/runner/runner.cpp

@@ -24,29 +24,42 @@
 #include <executorch/runtime/core/exec_aten/util/scalar_type_util.h>
 #include <executorch/runtime/platform/log.h>
 
-namespace torch {
-namespace executor {
+namespace torch::executor {
+namespace {
+static constexpr auto kTopp = 0.9f;
+} // namespace
 
 Runner::Runner(
-    const char* model_path,
-    const char* tokenizer_path,
-    float temperature) {
-  // Constants definition
-  float topp = 0.9f;
-  unsigned long long rng_seed =
-      (unsigned int)time(nullptr); // seed rng with time by default
-  // Create module
-  module_ = std::make_unique<Module>(
-      model_path, Module::MlockConfig::UseMlockIgnoreErrors);
+    const std::string& model_path,
+    const std::string& tokenizer_path,
+    const float temperature)
+    : module_(std::make_unique<Module>(
+          model_path,
+          Module::MlockConfig::UseMlockIgnoreErrors)),
+      tokenizer_path_(tokenizer_path),
+      temperature_(temperature) {
+  ET_LOG(
+      Info,
+      "Creating LLaMa runner: model_path=%s, tokenizer_path=%s",
+      model_path.c_str(),
+      tokenizer_path.c_str());
+}
+
+bool Runner::is_loaded() const {
+  return module_->is_loaded() && tokenizer_ && sampler_;
+}
+
+Error Runner::load() {
+  if (is_loaded()) {
+    return Error::Ok;
+  }
+  ET_CHECK_OK_OR_RETURN_ERROR(module_->load_method("forward"));
 
   // Read out metadata: vocab_size (expected by the model), BOS, EOS, n_BOS,
   // n_EOS max_seq_len from the model
   ET_LOG(Info, "Reading metadata from model");
   const auto method_names = module_->method_names();
-  ET_CHECK_MSG(
-      method_names.ok(),
-      "Failed to read method names from model: %s",
-      model_path);
+  ET_CHECK_MSG(method_names.ok(), "Failed to read method names from model");
   model_methods_ = method_names.get();
   vocab_size_ = getMetadataHelper<int64_t>("get_vocab_size", 32000);
   bos_id_ = getMetadataHelper<int64_t>("get_bos_id", 1);
@@ -59,7 +72,7 @@ Runner::Runner(
 
   // Load tokenizer
   tokenizer_ = std::make_unique<Tokenizer>(vocab_size_, bos_id_, eos_id_);
-  tokenizer_->load(tokenizer_path);
+  tokenizer_->load(tokenizer_path_);
   if (tokenizer_->bos_tok() != bos_id_) {
     ET_LOG(
         Error,
@@ -75,8 +88,13 @@ Runner::Runner(
         eos_id_);
   }
   // Create sampler
-  sampler_ =
-      std::make_unique<Sampler>(vocab_size_, temperature, topp, rng_seed);
+  sampler_ = std::make_unique<Sampler>(
+      vocab_size_,
+      temperature_,
+      kTopp,
+      static_cast<unsigned long long>(std::time(nullptr)));
+
+  return Error::Ok;
 }
 
 template <typename T>
@@ -141,6 +159,9 @@ Error Runner::generate(
   // Prepare the inputs.
   // Use ones-initialized inputs.
   ET_CHECK_MSG(!prompt.empty(), "Prompt cannot be null");
+  ET_CHECK_OK_OR_RETURN_ERROR(load());
+
+  shouldStop_ = false;
 
   // encode the (string) prompt into tokens sequence
   int num_prompt_tokens = 0;
@@ -292,6 +313,10 @@ Error Runner::generate(
       callback(piece);
     }
 
+    if (shouldStop_) {
+      break;
+    }
+
     // data-dependent terminating condition: we have n_eos_ number of EOS
     if (pos >= num_prompt_tokens && next == eos_id_) {
       eos_counter++;
@@ -338,12 +363,15 @@ Error Runner::generate(
   return Error::Ok;
 }
 
+void Runner::stop() {
+  shouldStop_ = true;
+}
+
 // explicit instantiation of template methods
 template int64_t Runner::getMetadataHelper<int64_t>(
     std::string method_name,
     int64_t default_val);
 template bool Runner::getMetadataHelper<bool>(
     std::string method_name,
     bool default_val);
-} // namespace executor
-} // namespace torch
+} // namespace torch::executor

examples/models/llama2/runner/runner.h

@@ -21,19 +21,21 @@
 #include <executorch/examples/models/llama2/tokenizer/tokenizer.h>
 #include <executorch/extension/module/module.h>
 
-namespace torch {
-namespace executor {
+namespace torch::executor {
 
 class Runner {
  public:
   explicit Runner(
-      const char* model_path,
-      const char* tokenizer_path,
-      float temperature = 0.8f);
+      const std::string& model_path,
+      const std::string& tokenizer_path,
+      const float temperature = 0.8f);
 
+  bool is_loaded() const;
+  Error load();
   Error generate(
       const std::string& prompt,
       std::function<void(const std::string&)> callback = {});
+  void stop();
 
  private:
   // metadata
@@ -53,13 +55,12 @@ class Runner {
   bool use_kv_cache_;
   bool append_eos_;
   std::unordered_set<std::string> model_methods_;
-  // module
   std::unique_ptr<Module> module_;
-  // tokenizer
+  std::string tokenizer_path_;
+  float temperature_;
   std::unique_ptr<Tokenizer> tokenizer_;
-  // sampler
   std::unique_ptr<Sampler> sampler_;
+  bool shouldStop_{false};
 };
 
-} // namespace executor
-} // namespace torch
+} // namespace torch::executor

examples/models/llama2/tokenizer/tokenizer.cpp

@@ -47,15 +47,15 @@ Tokenizer::Tokenizer(int32_t vocab_size, int32_t bos_tok, int32_t eos_tok)
  * @param tokenizer_path The path to the tokenizer file.
  * @return Error
  */
-Error Tokenizer::load(const char* tokenizer_path) {
+Error Tokenizer::load(const std::string& tokenizer_path) {
   if (initialized_) {
     ET_LOG(Info, "Tokenizer already initialized");
     return Error::Ok;
   }
   // read in the file
-  FILE* file = fopen(tokenizer_path, "rb");
+  FILE* file = fopen(tokenizer_path.c_str(), "rb");
   if (!file) {
-    ET_LOG(Error, "couldn't load %s", tokenizer_path);
+    ET_LOG(Error, "couldn't load %s", tokenizer_path.c_str());
     return Error::InvalidArgument;
   }
   int32_t metadata[2];

examples/models/llama2/tokenizer/tokenizer.h

@@ -15,6 +15,7 @@
 #include <cstdlib>
 #include <cstring>
 #include <memory>
+#include <string>
 
 #include <executorch/runtime/core/error.h>
 #include <executorch/runtime/core/exec_aten/exec_aten.h>
@@ -34,7 +35,7 @@ class Tokenizer {
   explicit Tokenizer(int32_t vocab_size, int32_t bos_tok, int32_t eos_tok);
   ~Tokenizer();
 
-  Error load(const char* tokenizer_path);
+  Error load(const std::string& tokenizer_path);
 
   Error encode(
       const char* text,