refactor: pass model version in save_load

VowpalWabbit · Nov 11, 2023 · 47f7bf0 · 47f7bf0
1 parent a541d85
commit 47f7bf0
Show file tree

Hide file tree

Showing 39 changed files with 108 additions and 110 deletions.
diff --git a/vowpalwabbit/core/include/vw/core/learner.h b/vowpalwabbit/core/include/vw/core/learner.h
@@ -70,6 +70,7 @@ using multipredict_func =
 
 using sensitivity_func = std::function<float(example& ex)>;
 using save_load_func = std::function<void(io_buf&, bool read, bool text)>;
+using save_load_ver_func = std::function<void(io_buf&, bool read, bool text, const VW::version_struct&)>;
 using pre_save_load_func = std::function<void(VW::workspace& all)>;
 using save_metric_func = std::function<void(metric_sink& metrics)>;
 
@@ -182,7 +183,7 @@ class learner final : public std::enable_shared_from_this<learner>
   float sensitivity(example& ec, size_t i = 0);
 
   // Called anytime saving or loading needs to happen. Autorecursive.
-  void save_load(io_buf& io, const bool read, const bool text);
+  void save_load(io_buf& io, const bool read, const bool text, const VW::version_struct& model_version);
 
   // Called to edit the command-line from a learner. Autorecursive
   void pre_save_load(VW::workspace& all);
@@ -287,6 +288,7 @@ class learner final : public std::enable_shared_from_this<learner>
   details::cleanup_example_func _cleanup_example_f;
 
   details::save_load_func _save_load_f;
+  details::save_load_ver_func _save_load_ver_f;
   details::void_func _end_pass_f;
   details::void_func _end_examples_f;
   details::pre_save_load_func _pre_save_load_f;
@@ -417,12 +419,19 @@ class common_learner_builder
     learner_ptr->learn_returns_prediction = learn_returns_prediction;
   )
 
+  // TODO: deprecate?
   LEARNER_BUILDER_DEFINE(set_save_load(void (*fn_ptr)(DataT&, io_buf&, bool, bool)),
     assert(fn_ptr != nullptr);
     DataT* data = this->learner_data.get();
     this->learner_ptr->_save_load_f = [fn_ptr, data](io_buf& buf, bool read, bool text)
     { fn_ptr(*data, buf, read, text); };
   )
+  LEARNER_BUILDER_DEFINE(set_save_load(void (*fn_ptr)(DataT&, io_buf&, bool, bool, const VW::version_struct&)),
+    assert(fn_ptr != nullptr);
+    DataT* data = this->learner_data.get();
+    this->learner_ptr->_save_load_ver_f = [fn_ptr, data](io_buf& buf, bool read, bool text, const VW::version_struct& ver)
+    { fn_ptr(*data, buf, read, text, ver); };
+  )
 
   LEARNER_BUILDER_DEFINE(set_finish(void (*fn_ptr)(DataT&)),
     assert(fn_ptr != nullptr);

diff --git a/vowpalwabbit/core/include/vw/core/reductions/active.h b/vowpalwabbit/core/include/vw/core/reductions/active.h
@@ -16,12 +16,10 @@ namespace reductions
 class active
 {
 public:
-  active(float active_c0, std::shared_ptr<shared_data> shared_data, std::shared_ptr<rand_state> random_state,
-      VW::version_struct model_version)
+  active(float active_c0, std::shared_ptr<shared_data> shared_data, std::shared_ptr<rand_state> random_state)
       : active_c0(active_c0)
       , _shared_data(shared_data)
       , _random_state(std::move(random_state))
-      , _model_version{std::move(model_version)}
   {
   }
 
@@ -31,7 +29,6 @@ class active
 
   float _min_seen_label = 0.f;
   float _max_seen_label = 1.f;
-  VW::version_struct _model_version;
 };
 
 std::shared_ptr<VW::LEARNER::learner> active_setup(VW::setup_base_i& stack_builder);

diff --git a/vowpalwabbit/core/include/vw/core/reductions/cb/cb_explore_adf_common.h b/vowpalwabbit/core/include/vw/core/reductions/cb/cb_explore_adf_common.h
@@ -93,7 +93,7 @@ class cb_explore_adf_base
     if (with_metrics) { _metrics = VW::make_unique<cb_explore_metrics>(); }
   }
 
-  static void save_load(cb_explore_adf_base<ExploreType>& data, io_buf& io, bool read, bool text);
+  static void save_load(cb_explore_adf_base<ExploreType>& data, io_buf& io, bool read, bool text, const VW::version_struct& ver);
   static void persist_metrics(cb_explore_adf_base<ExploreType>& data, metric_sink& metrics);
   static void predict(cb_explore_adf_base<ExploreType>& data, VW::LEARNER::learner& base, multi_ex& examples);
   static void learn(cb_explore_adf_base<ExploreType>& data, VW::LEARNER::learner& base, multi_ex& examples);
@@ -302,9 +302,9 @@ void cb_explore_adf_base<ExploreType>::_print_update(
 
 template <typename ExploreType>
 inline void cb_explore_adf_base<ExploreType>::save_load(
-    cb_explore_adf_base<ExploreType>& data, io_buf& io, bool read, bool text)
+    cb_explore_adf_base<ExploreType>& data, io_buf& io, bool read, bool text, const VW::version_struct& ver)
 {
-  data.explore.save_load(io, read, text);
+  data.explore.save_load(io, read, text, ver);
 }
 
 template <typename ExploreType>

diff --git a/vowpalwabbit/core/src/learner.cc b/vowpalwabbit/core/src/learner.cc
@@ -442,7 +442,7 @@ float learner::sensitivity(example& ec, size_t i)
   return ret;
 }
 
-void learner::save_load(io_buf& io, const bool read, const bool text)
+void learner::save_load(io_buf& io, const bool read, const bool text, const VW::version_struct& model_version)
 {
   if (_save_load_f)
   {
@@ -457,7 +457,20 @@ void learner::save_load(io_buf& io, const bool read, const bool text)
       throw VW::save_load_model_exception(vwex.filename(), vwex.line_number(), better_msg.str());
     }
   }
-  if (_base_learner) { _base_learner->save_load(io, read, text); }
+  else if (_save_load_ver_f)
+  {
+    try
+    {
+      _save_load_ver_f(io, read, text, model_version);
+    }
+    catch (VW::vw_exception& vwex)
+    {
+      std::stringstream better_msg;
+      better_msg << "model " << std::string(read ? "load" : "save") << " failed. Error Details: " << vwex.what();
+      throw VW::save_load_model_exception(vwex.filename(), vwex.line_number(), better_msg.str());
+    }
+  }
+  if (_base_learner) { _base_learner->save_load(io, read, text, model_version); }
 }
 
 void learner::pre_save_load(VW::workspace& all)

diff --git a/vowpalwabbit/core/src/parse_args.cc b/vowpalwabbit/core/src/parse_args.cc
@@ -1362,7 +1362,7 @@ void load_input_model(VW::workspace& all, VW::io_buf& io_temp)
       all.feature_mask == all.initial_weights_config.initial_regressors[0])
   {
     // load rest of regressor
-    all.l->save_load(io_temp, true, false);
+    all.l->save_load(io_temp, true, false, all.runtime_state.model_file_ver);
     io_temp.close_file();
 
     VW::details::parse_mask_regressor_args(all, all.feature_mask, all.initial_weights_config.initial_regressors);
@@ -1372,7 +1372,7 @@ void load_input_model(VW::workspace& all, VW::io_buf& io_temp)
     VW::details::parse_mask_regressor_args(all, all.feature_mask, all.initial_weights_config.initial_regressors);
 
     // load rest of regressor
-    all.l->save_load(io_temp, true, false);
+    all.l->save_load(io_temp, true, false, all.runtime_state.model_file_ver);
     io_temp.close_file();
   }
 }

diff --git a/vowpalwabbit/core/src/parse_regressor.cc b/vowpalwabbit/core/src/parse_regressor.cc
@@ -489,7 +489,7 @@ void VW::details::dump_regressor(VW::workspace& all, VW::io_buf& buf, bool as_te
   std::string unused;
   if (all.l != nullptr) { all.l->pre_save_load(all); }
   VW::details::save_load_header(all, buf, false, as_text, unused, *all.options);
-  if (all.l != nullptr) { all.l->save_load(buf, false, as_text); }
+  if (all.l != nullptr) { all.l->save_load(buf, false, as_text, all.runtime_state.model_file_ver); }
 
   buf.flush();  // close_file() should do this for me ...
   buf.close_file();
@@ -580,7 +580,7 @@ void VW::details::parse_mask_regressor_args(
     io_temp_mask.add_file(VW::io::open_file_reader(feature_mask));
 
     save_load_header(all, io_temp_mask, true, false, file_options, *all.options);
-    all.l->save_load(io_temp_mask, true, false);
+    all.l->save_load(io_temp_mask, true, false, all.runtime_state.model_file_ver);
     io_temp_mask.close_file();
 
     // Deal with the over-written header from initial regressor

diff --git a/vowpalwabbit/core/src/reductions/active.cc b/vowpalwabbit/core/src/reductions/active.cc
@@ -127,16 +127,16 @@ void active_print_result(
   if (t != len) { logger.err_error("write error: {}", VW::io::strerror_to_string(errno)); }
 }
 
-void save_load(active& a, VW::io_buf& io, bool read, bool text)
+void save_load(active& a, VW::io_buf& io, bool read, bool text, const VW::version_struct& ver)
 {
   using namespace VW::version_definitions;
   if (io.num_files() == 0) { return; }
   // This code is valid if version is within
   // [VERSION_FILE_WITH_ACTIVE_SEEN_LABELS, VERSION_FILE_WITH_ACTIVE_SEEN_LABELS_REVERTED)
   // or >= VERSION_FILE_WITH_ACTIVE_SEEN_LABELS_FIXED
-  if ((a._model_version >= VERSION_FILE_WITH_ACTIVE_SEEN_LABELS &&
-          a._model_version < VERSION_FILE_WITH_ACTIVE_SEEN_LABELS_REVERTED) ||
-      a._model_version >= VERSION_FILE_WITH_ACTIVE_SEEN_LABELS_FIXED)
+  if ((ver >= VERSION_FILE_WITH_ACTIVE_SEEN_LABELS &&
+          ver < VERSION_FILE_WITH_ACTIVE_SEEN_LABELS_REVERTED) ||
+      ver >= VERSION_FILE_WITH_ACTIVE_SEEN_LABELS_FIXED)
   {
     if (read)
     {
@@ -201,7 +201,7 @@ std::shared_ptr<VW::LEARNER::learner> VW::reductions::active_setup(VW::setup_bas
   if (!options.add_parse_and_check_necessary(new_options)) { return nullptr; }
 
   if (options.was_supplied("lda")) { THROW("lda cannot be combined with active learning") }
-  auto data = VW::make_unique<active>(active_c0, all.sd, all.get_random_state(), all.runtime_state.model_file_ver);
+  auto data = VW::make_unique<active>(active_c0, all.sd, all.get_random_state());
   auto base = require_singleline(stack_builder.setup_base_learner());
 
   using learn_pred_func_t = void (*)(active&, VW::LEARNER::learner&, VW::example&);

diff --git a/vowpalwabbit/core/src/reductions/automl.cc b/vowpalwabbit/core/src/reductions/automl.cc
@@ -112,7 +112,7 @@ void pre_save_load_automl(VW::workspace& all, automl<CMType>& data)
 }
 
 template <typename CMType>
-void save_load_automl(automl<CMType>& aml, VW::io_buf& io, bool read, bool text)
+void save_load_automl(automl<CMType>& aml, VW::io_buf& io, bool read, bool text, const VW::version_struct&)
 {
   if (io.num_files() == 0) { return; }
   if (read) { VW::model_utils::read_model_field(io, aml); }

diff --git a/vowpalwabbit/core/src/reductions/baseline_challenger_cb.cc b/vowpalwabbit/core/src/reductions/baseline_challenger_cb.cc
@@ -163,7 +163,7 @@ void learn_or_predict(baseline_challenger_data& data, learner& base, VW::multi_e
   data.learn_or_predict<is_learn>(base, examples);
 }
 
-void save_load(baseline_challenger_data& data, VW::io_buf& io, bool read, bool text)
+void save_load(baseline_challenger_data& data, VW::io_buf& io, bool read, bool text, const VW::version_struct&)
 {
   if (io.num_files() == 0) { return; }
   if (read) { VW::model_utils::read_model_field(io, data); }

diff --git a/vowpalwabbit/core/src/reductions/bfgs.cc b/vowpalwabbit/core/src/reductions/bfgs.cc
@@ -1048,7 +1048,7 @@ void save_load_regularizer(VW::workspace& all, bfgs& b, VW::io_buf& model_file,
   if (read) { regularizer_to_weight(all, b); }
 }
 
-void save_load(bfgs& b, VW::io_buf& model_file, bool read, bool text)
+void save_load(bfgs& b, VW::io_buf& model_file, bool read, bool text, const VW::version_struct&)
 {
   VW::workspace* all = b.all;
 

diff --git a/vowpalwabbit/core/src/reductions/boosting.cc b/vowpalwabbit/core/src/reductions/boosting.cc
@@ -239,7 +239,7 @@ void predict_or_learn_adaptive(boosting& o, VW::LEARNER::learner& base, VW::exam
   ec.pred.scalar = final_prediction;
 }
 
-void save_load_sampling(boosting& o, VW::io_buf& model_file, bool read, bool text)
+void save_load_sampling(boosting& o, VW::io_buf& model_file, bool read, bool text, const VW::version_struct&)
 {
   if (model_file.num_files() == 0) { return; }
   std::stringstream os;
@@ -298,7 +298,7 @@ void save_load_sampling(boosting& o, VW::io_buf& model_file, bool read, bool tex
   o.logger.err_info("{}", fmt::to_string(buffer));
 }
 
-void save_load(boosting& o, VW::io_buf& model_file, bool read, bool text)
+void save_load(boosting& o, VW::io_buf& model_file, bool read, bool text, const VW::version_struct&)
 {
   if (model_file.num_files() == 0) { return; }
   std::stringstream os;
@@ -340,7 +340,7 @@ void save_load(boosting& o, VW::io_buf& model_file, bool read, bool text)
   }
 }
 
-void save_load_boosting_noop(boosting&, VW::io_buf&, bool, bool) {}
+void save_load_boosting_noop(boosting&, VW::io_buf&, bool, bool, const VW::version_struct&) {}
 }  // namespace
 
 std::shared_ptr<VW::LEARNER::learner> VW::reductions::boosting_setup(VW::setup_base_i& stack_builder)
@@ -383,7 +383,7 @@ std::shared_ptr<VW::LEARNER::learner> VW::reductions::boosting_setup(VW::setup_b
   std::string name_addition;
   void (*learn_ptr)(boosting&, VW::LEARNER::learner&, VW::example&);
   void (*pred_ptr)(boosting&, VW::LEARNER::learner&, VW::example&);
-  void (*save_load_fn)(boosting&, io_buf&, bool, bool);
+  void (*save_load_fn)(boosting&, io_buf&, bool, bool, const VW::version_struct&);
 
   if (data->alg == "BBM")
   {

diff --git a/vowpalwabbit/core/src/reductions/cb/cb_adf.cc b/vowpalwabbit/core/src/reductions/cb/cb_adf.cc
@@ -345,10 +345,9 @@ void print_update_cb_adf(VW::workspace& all, VW::shared_data& /* sd */, const VW
   else { VW::details::print_update_cb(all, !labeled_example, ec, &ec_seq, true, nullptr); }
 }
 
-void save_load(VW::reductions::cb_adf& c, VW::io_buf& model_file, bool read, bool text)
+void save_load(VW::reductions::cb_adf& c, VW::io_buf& model_file, bool read, bool text, const VW::version_struct& ver)
 {
-  if (c.get_model_file_ver() != nullptr &&
-      *c.get_model_file_ver() < VW::version_definitions::VERSION_FILE_WITH_CB_ADF_SAVE)
+  if (ver < VW::version_definitions::VERSION_FILE_WITH_CB_ADF_SAVE)
   {
     return;
   }

diff --git a/vowpalwabbit/core/src/reductions/cb/cb_explore.cc b/vowpalwabbit/core/src/reductions/cb/cb_explore.cc
@@ -54,7 +54,6 @@ class cb_explore
   float psi = 0.f;
   bool nounif = false;
   bool epsilon_decay = false;
-  VW::version_struct model_file_version;
   VW::io::logger logger;
 
   size_t counter = 0;
@@ -270,11 +269,11 @@ float calc_loss(const cb_explore& data, const VW::example& ec, const VW::cb_labe
   return loss;
 }
 
-void save_load(cb_explore& cb, VW::io_buf& io, bool read, bool text)
+void save_load(cb_explore& cb, VW::io_buf& io, bool read, bool text, const VW::version_struct& ver)
 {
   if (io.num_files() == 0) { return; }
 
-  if (!read || cb.model_file_version >= VW::version_definitions::VERSION_FILE_WITH_CCB_MULTI_SLOTS_SEEN_FLAG)
+  if (!read || ver >= VW::version_definitions::VERSION_FILE_WITH_CCB_MULTI_SLOTS_SEEN_FLAG)
   {
     std::stringstream msg;
     if (!read) { msg << "cb cover storing VW::example counter:  = " << cb.counter << "\n"; }
@@ -372,7 +371,6 @@ std::shared_ptr<VW::LEARNER::learner> VW::reductions::cb_explore_setup(VW::setup
   if (data->epsilon < 0.0 || data->epsilon > 1.0) { THROW("The value of epsilon must be in [0,1]"); }
 
   data->cbcs.cb_type = VW::cb_type_t::DR;
-  data->model_file_version = all.runtime_state.model_file_ver;
 
   size_t params_per_weight = 1;
   if (options.was_supplied("cover")) { params_per_weight = data->cover_size + 1; }

diff --git a/vowpalwabbit/core/src/reductions/cb/cb_explore_adf_cover.cc b/vowpalwabbit/core/src/reductions/cb/cb_explore_adf_cover.cc
@@ -36,12 +36,12 @@ class cb_explore_adf_cover
 public:
   cb_explore_adf_cover(size_t cover_size, float psi, bool nounif, float epsilon, bool epsilon_decay, bool first_only,
       VW::LEARNER::learner* cs_ldf_learner, VW::LEARNER::learner* scorer, VW::cb_type_t cb_type,
-      VW::version_struct model_file_version, VW::io::logger logger);
+      VW::io::logger logger);
 
   // Should be called through cb_explore_adf_base for pre/post-processing
   void predict(VW::LEARNER::learner& base, VW::multi_ex& examples) { predict_or_learn_impl<false>(base, examples); }
   void learn(VW::LEARNER::learner& base, VW::multi_ex& examples) { predict_or_learn_impl<true>(base, examples); }
-  void save_load(VW::io_buf& io, bool read, bool text);
+  void save_load(VW::io_buf& io, bool read, bool text, const VW::version_struct&);
 
 private:
   size_t _cover_size;
@@ -56,7 +56,6 @@ class cb_explore_adf_cover
   VW::details::cb_to_cs_adf_dr _gen_cs_dr;
   VW::cb_type_t _cb_type = VW::cb_type_t::DM;
 
-  VW::version_struct _model_file_version;
   VW::io::logger _logger;
 
   VW::v_array<VW::action_score> _action_probs;
@@ -71,7 +70,7 @@ class cb_explore_adf_cover
 
 cb_explore_adf_cover::cb_explore_adf_cover(size_t cover_size, float psi, bool nounif, float epsilon, bool epsilon_decay,
     bool first_only, VW::LEARNER::learner* cs_ldf_learner, VW::LEARNER::learner* scorer, VW::cb_type_t cb_type,
-    VW::version_struct model_file_version, VW::io::logger logger)
+    VW::io::logger logger)
     : _cover_size(cover_size)
     , _psi(psi)
     , _nounif(nounif)
@@ -81,7 +80,6 @@ cb_explore_adf_cover::cb_explore_adf_cover(size_t cover_size, float psi, bool no
     , _counter(0)
     , _cs_ldf_learner(cs_ldf_learner)
     , _cb_type(cb_type)
-    , _model_file_version(model_file_version)
     , _logger(std::move(logger))
 {
   _gen_cs_dr.scorer = scorer;
@@ -222,10 +220,10 @@ void cb_explore_adf_cover::predict_or_learn_impl(VW::LEARNER::learner& base, VW:
   if (is_learn) { ++_counter; }
 }
 
-void cb_explore_adf_cover::save_load(VW::io_buf& io, bool read, bool text)
+void cb_explore_adf_cover::save_load(VW::io_buf& io, bool read, bool text, const VW::version_struct& ver)
 {
   if (io.num_files() == 0) { return; }
-  if (!read || _model_file_version >= VW::version_definitions::VERSION_FILE_WITH_CCB_MULTI_SLOTS_SEEN_FLAG)
+  if (!read || ver >= VW::version_definitions::VERSION_FILE_WITH_CCB_MULTI_SLOTS_SEEN_FLAG)
   {
     std::stringstream msg;
     if (!read) { msg << "cb cover adf storing example counter:  = " << _counter << "\n"; }
@@ -326,7 +324,7 @@ std::shared_ptr<VW::LEARNER::learner> VW::reductions::cb_explore_adf_cover_setup
   using explore_type = cb_explore_adf_base<cb_explore_adf_cover>;
   auto data = VW::make_unique<explore_type>(all.output_runtime.global_metrics.are_metrics_enabled(),
       VW::cast_to_smaller_type<size_t>(cover_size), psi, nounif, epsilon, epsilon_decay, first_only, cost_sensitive,
-      scorer, cb_type, all.runtime_state.model_file_ver, all.logger);
+      scorer, cb_type, all.logger);
   auto l = make_reduction_learner(std::move(data), base, explore_type::learn, explore_type::predict,
       stack_builder.get_setupfn_name(cb_explore_adf_cover_setup))
                .set_input_label_type(VW::label_type_t::CB)