cleanup

generate.py

@@ -190,7 +190,6 @@ class GenerateMidiText:
         verbose=True,
         expected_length=None,
     ):
-
         """generate until the TRACK_END token is reached
         full_piece = input_prompt + generated"""
         if expected_length is None:
@@ -270,14 +269,17 @@ class GenerateMidiText:
 
     def generate_piece(self, instrument_list, density_list, temperature_list):
         """generate a sequence with mutiple tracks
-        - inst_list sets the list of instruments of the order of generation
-        - density is paired with inst_list
-        Each track/intrument is generated on a prompt which contains the previously generated track/instrument
-        This means that the first instrument is generated with less bias than the next one, and so on.
 
-        'generated_piece' keeps track of the entire piece
-        'generated_piece' is returned by self.generate_until_track_end
-        # it is returned by self.generate_until_track_end"""
+        Args:
+            - inst_list sets the list of instruments and the the order of generation
+            - density and
+            - temperature are paired with inst_list
+
+        Each track/intrument is generated based on a prompt which contains the previously generated track/instrument
+
+        Returns:
+        'generated_piece' which keeps track of the entire piece
+        """
 
         generated_piece = "PIECE_START "
         for instrument, density, temperature in zip(
@@ -321,7 +323,9 @@ class GenerateMidiText:
                     for bar in track["bars"][-self.model_n_bar :]:
                         pre_promt += bar
                     pre_promt += "TRACK_END "
-                elif False:  # len_diff <= 0: # THIS GENERATES EMPTINESS
+                elif (
+                    False
+                ):  # len_diff <= 0: # THIS DOES NOT WORK - It just fills things with empty bars
                     # adding an empty bars at the end of the other tracks if they have not been processed yet
                     pre_promt += othertracks["bars"][0]
                     for bar in track["bars"][-(self.model_n_bar - 1) :]:
@@ -378,7 +382,7 @@ class GenerateMidiText:
     def check_the_piece_for_errors(self, piece: str = None):
 
         if piece is None:
-            piece = generate_midi.get_whole_piece_from_bar_dict()
+            piece = self.get_whole_piece_from_bar_dict()
         errors = []
         errors.append(
             [
@@ -396,91 +400,4 @@ class GenerateMidiText:
 
 
 if __name__ == "__main__":
-
-    # worker
-    DEVICE = "cpu"
-
-    # define generation parameters
-    N_FILES_TO_GENERATE = 2
-    Temperatures_to_try = [0.7]
-
-    USE_FAMILIZED_MODEL = True
-    force_sequence_length = True
-
-    if USE_FAMILIZED_MODEL:
-        # model_repo = "misnaej/the-jam-machine-elec-famil"
-        # model_repo = "misnaej/the-jam-machine-elec-famil-ft32"
-
-        # model_repo = "JammyMachina/elec-gmusic-familized-model-13-12__17-35-53"
-        # n_bar_generated = 8
-
-        model_repo = "JammyMachina/improved_4bars-mdl"
-        n_bar_generated = 4
-        instrument_promt_list = ["4", "DRUMS", "3"]
-        # DRUMS = drums, 0 = piano, 1 = chromatic percussion, 2 = organ, 3 = guitar, 4 = bass, 5 = strings, 6 = ensemble, 7 = brass, 8 = reed, 9 = pipe, 10 = synth lead, 11 = synth pad, 12 = synth effects, 13 = ethnic, 14 = percussive, 15 = sound effects
-        density_list = [3, 2, 2]
-        # temperature_list = [0.7, 0.7, 0.75]
-    else:
-        model_repo = "misnaej/the-jam-machine"
-        instrument_promt_list = ["30"]  # , "DRUMS", "0"]
-        density_list = [3]  # , 2, 3]
-        # temperature_list = [0.7, 0.5, 0.75]
-        pass
-
-    # define generation directory
-    generated_sequence_files_path = define_generation_dir(model_repo)
-
-    # load model and tokenizer
-    model, tokenizer = LoadModel(
-        model_repo, from_huggingface=True
-    ).load_model_and_tokenizer()
-
-    # does the prompt make sense
-    check_if_prompt_inst_in_tokenizer_vocab(tokenizer, instrument_promt_list)
-
-    for temperature in Temperatures_to_try:
-        print(f"================= TEMPERATURE {temperature} =======================")
-        for _ in range(N_FILES_TO_GENERATE):
-            print(f"========================================")
-            # 1 - instantiate
-            generate_midi = GenerateMidiText(model, tokenizer)
-            # 0 - set the n_bar for this model
-            generate_midi.set_nb_bars_generated(n_bars=n_bar_generated)
-            # 1 - defines the instruments, densities and temperatures
-            # 2- generate the first 8 bars for each instrument
-            generate_midi.set_improvisation_level(30)
-            generate_midi.generate_piece(
-                instrument_promt_list,
-                density_list,
-                [temperature for _ in density_list],
-            )
-            # 3 - force the model to improvise
-            # generate_midi.set_improvisation_level(20)
-            # 4 - generate the next 4 bars for each instrument
-            # generate_midi.generate_n_more_bars(n_bar_generated)
-            # 5 - lower the improvisation level
-            generate_midi.generated_piece = (
-                generate_midi.get_whole_piece_from_bar_dict()
-            )
-
-            # print the generated sequence in terminal
-            print("=========================================")
-            print(generate_midi.generated_piece)
-            print("=========================================")
-
-            # write to JSON file
-            filename = WriteTextMidiToFile(
-                generate_midi,
-                generated_sequence_files_path,
-            ).text_midi_to_file()
-
-            # decode the sequence to MIDI """
-            decode_tokenizer = get_miditok()
-            TextDecoder(decode_tokenizer, USE_FAMILIZED_MODEL).get_midi(
-                generate_midi.generated_piece, filename=filename.split(".")[0] + ".mid"
-            )
-            inst_midi, mixed_audio = get_music(filename.split(".")[0] + ".mid")
-            max_time = get_max_time(inst_midi)
-            plot_piano_roll(inst_midi)
-
-            print("Et voilà! Your MIDI file is ready! GO JAM!")
+    pass

generation_utils.py

@@ -15,10 +15,6 @@ matplotlib.rcParams["axes.edgecolor"] = "grey"
 
 
 def define_generation_dir(model_repo_path):
-    #### to remove later ####
-    if model_repo_path == "models/model_2048_fake_wholedataset":
-        model_repo_path = "misnaej/the-jam-machine"
-    #### to remove later ####
     generated_sequence_files_path = f"midi/generated/{model_repo_path}"
     if not os.path.exists(generated_sequence_files_path):
         os.makedirs(generated_sequence_files_path)
@@ -61,6 +57,11 @@ def check_if_prompt_inst_in_tokenizer_vocab(tokenizer, inst_prompt_list):
             )
 
 
+# TODO
+def check_if_prompt_density_in_tokenizer_vocab(tokenizer, density_prompt_list):
+    pass
+
+
 def forcing_bar_count(input_prompt, generated, bar_count, expected_length):
     """Forcing the generated sequence to have the expected length
     expected_length and bar_count refers to the length of newly_generated_only (without input prompt)"""