| { |
| "mic": "0.3", |
| "name": "MusicGen Melody text- and melody-conditioned music generation", |
| "usage": "AutoProcessor + MusicgenMelodyForConditionalGeneration.generate, text and melody prompts, classifier-free guidance, batched", |
| "architecture": "MusicgenMelodyForConditionalGeneration", |
| "checkpoints": { |
| "facebook/musicgen-melody": { |
| "fixture": "hf-internal-testing/mic-musicgen-melody-slim" |
| } |
| }, |
| "fixtures": { |
| "hf-internal-testing/mic-musicgen-melody-slim": { |
| "built_from": "facebook/musicgen-melody", |
| "covers": [ |
| "facebook/musicgen-melody" |
| ], |
| "build": { |
| "seed": 17, |
| "config_overrides": { |
| "fields": { |
| "audio_encoder.num_filters": 4, |
| "audio_encoder.codebook_dim": 64 |
| }, |
| "reason": "The generic rule shrinks EnCodec's hidden_size but not num_filters, which doubles at each of its four stages (64 filters reach 1024 channels at the LSTM, 56M of 59M parameters), nor codebook_dim, which must equal hidden_size for the decoder to accept the quantizer's output." |
| }, |
| "overfit": "Random weights pass the input-sensitivity gate (teacher-forced logits move), but greedy generation ignores the conditioning: both prompts, with or without a melody, give the same audio codes, so a change that drops the text or melody conditioning would go unseen." |
| } |
| } |
| }, |
| "integrations": { |
| "transformers": { |
| "entry_point": "integrations/transformers/usage.md", |
| "format": "runnable-markdown", |
| "requires": [ |
| "transformers>=5", |
| "torch", |
| "sentencepiece" |
| ], |
| "targets": [ |
| "fixtures", |
| "checkpoints" |
| ], |
| "variants": { |
| "default": {}, |
| "eager": { |
| "attn_implementation": "eager" |
| }, |
| "bfloat16": { |
| "dtype": "bfloat16" |
| } |
| }, |
| "training": true, |
| "resources": { |
| "fixture": { |
| "memory_gb": 1, |
| "download_mb": 20 |
| }, |
| "facebook/musicgen-melody": { |
| "memory_gb": 12, |
| "download_mb": 6230 |
| } |
| } |
| } |
| }, |
| "outputs": { |
| "runtime": { |
| "compare": "exact" |
| }, |
| "input_feature_shape": { |
| "compare": "exact" |
| }, |
| "codes": { |
| "compare": "exact" |
| }, |
| "audio_shape": { |
| "compare": "exact" |
| }, |
| "audio_head": { |
| "compare": "allclose", |
| "atol": 0.0001, |
| "rtol": 0.0001 |
| }, |
| "text_only_codes": { |
| "compare": "exact" |
| }, |
| "training.loss": { |
| "compare": "allclose", |
| "atol": 0.0001, |
| "rtol": 0.0001 |
| }, |
| "training.gradients": { |
| "compare": "all_true" |
| }, |
| "training.updated": { |
| "compare": "true" |
| } |
| } |
| } |
|
|
