Update configs/inference.yaml

configs/inference.yaml

@@ -1,65 +1,65 @@
-model: 
-  name: 'google/t5-v1_1-small'
-  spectrogram:
-    sample_rate: 16000
-    hop_length: 128
-    n_fft: 1024
-    n_mels: 388
-  do_style_embed: false
-  input_features: false
-
-model_path: './checkpoint'
-audio_path: ''            # Path to input audio
-total_duration_ms: 0    # Total duration of audio in milliseconds, 0 for full audio
-output_path: ''         # Path to output directory
-bpm: 120         # Beats per minute of input audio
-offset: 0            # Start of beat, in miliseconds, from the beginning of input audio
-resnap_objects: false    # Resnap objects beat timing ticks, requires accurate BPM and offset
-slider_multiplier: 1.7  # Multiplier for slider velocity
-title: ''               # Song title
-artist: ''              # Song artist
-beatmap_path: ''        # Path to .osu file which will be remapped
-other_beatmap_path: ''  # Path to .osu file of other beatmap in the mapset to use as reference
-beatmap_id: -1          # Beatmap ID to use as style
-difficulty: -1  # Difficulty star rating to map
-creator: ''        # Beatmap creator
-version: ''        # Beatmap version
-full_set: true         # Generate full mapset
-set_difficulties: 5   # Number of difficulties to generate.
-
-# Diffusion settings
-generate_positions: true      # Use diffusion to generate object positions
-diff_ckpt: './osudiffusion/DiT-B-0700000.pt'                  # Path to checkpoint for diffusion model
-diff_refine_ckpt: ''           # Path to checkpoint for refining diffusion model
-
-diffusion:
-  style_id: 1451282                # Style ID to use for diffusion
-  num_sampling_steps: 100    # Number of sampling steps
-  cfg_scale: 1           # Scale of classifier-free guidance
-  num_classes: 52670          # Number of classes stored in the model
-  beatmap_idx: 'osudiffusion/beatmap_idx.pickle'  # Path to beatmap index
-  use_amp: true                      # Use automatic mixed precision
-  refine_iters: 10                  # Number of refinement iterations
-  seq_len: 128                      # Sequence length
-  model: 'DiT-B'                    # Model architecture
-
-
-data:                  # Data settings
-  src_seq_len: 640
-  tgt_seq_len: 480
-  sample_rate: ${model.spectrogram.sample_rate}
-  hop_length: ${model.spectrogram.hop_length}
-  sequence_stride: 1    # Fraction of audio sequence length to shift inference window
-  center_pad_decoder: false            # Center pad decoder input
-  add_pre_tokens: true
-  special_token_len: 2
-  diff_token_index: 0
-  style_token_index: -1
-  max_pre_token_len: 4
-  add_gd_context: false  # Prefix the decoder with tokens of another beatmap in the mapset
-
-hydra:
-  job:
-    chdir: False
-  run:
+model: 
+  name: 'google/t5-v1_1-small'
+  spectrogram:
+    sample_rate: 16000
+    hop_length: 128
+    n_fft: 1024
+    n_mels: 256
+  do_style_embed: false
+  input_features: false
+
+model_path: './checkpoint'
+audio_path: ''            # Path to input audio
+total_duration_ms: 0    # Total duration of audio in milliseconds, 0 for full audio
+output_path: ''         # Path to output directory
+bpm: 120         # Beats per minute of input audio
+offset: 0            # Start of beat, in miliseconds, from the beginning of input audio
+resnap_objects: false    # Resnap objects beat timing ticks, requires accurate BPM and offset
+slider_multiplier: 1.7  # Multiplier for slider velocity
+title: ''               # Song title
+artist: ''              # Song artist
+beatmap_path: ''        # Path to .osu file which will be remapped
+other_beatmap_path: ''  # Path to .osu file of other beatmap in the mapset to use as reference
+beatmap_id: -1          # Beatmap ID to use as style
+difficulty: -1  # Difficulty star rating to map
+creator: ''        # Beatmap creator
+version: ''        # Beatmap version
+full_set: false         # Generate full mapset
+set_difficulties: 5   # Number of difficulties to generate.
+
+# Diffusion settings
+generate_positions: true      # Use diffusion to generate object positions
+diff_ckpt: './osudiffusion/DiT-B-0700000.pt'                  # Path to checkpoint for diffusion model
+diff_refine_ckpt: ''           # Path to checkpoint for refining diffusion model
+
+diffusion:
+  style_id: 1451282                # Style ID to use for diffusion
+  num_sampling_steps: 100    # Number of sampling steps
+  cfg_scale: 1           # Scale of classifier-free guidance
+  num_classes: 52670          # Number of classes stored in the model
+  beatmap_idx: 'osudiffusion/beatmap_idx.pickle'  # Path to beatmap index
+  use_amp: true                      # Use automatic mixed precision
+  refine_iters: 10                  # Number of refinement iterations
+  seq_len: 128                      # Sequence length
+  model: 'DiT-B'                    # Model architecture
+
+
+data:                  # Data settings
+  src_seq_len: 800
+  tgt_seq_len: 600
+  sample_rate: ${model.spectrogram.sample_rate}
+  hop_length: ${model.spectrogram.hop_length}
+  sequence_stride: 1    # Fraction of audio sequence length to shift inference window
+  center_pad_decoder: false            # Center pad decoder input
+  add_pre_tokens: true
+  special_token_len: 2
+  diff_token_index: 0
+  style_token_index: -1
+  max_pre_token_len: 4
+  add_gd_context: false  # Prefix the decoder with tokens of another beatmap in the mapset
+
+hydra:
+  job:
+    chdir: False
+  run:
     dir: ./logs/${now:%Y-%m-%d}/${now:%H-%M-%S}