Replicate API

.cog/tmp/build4097781755/requirements.txt

@@ -0,0 +1,21 @@
+av
+einops
+flashy>=0.0.1
+hydra-core>=1.1
+hydra_colorlog
+julius
+num2words
+numpy
+sentencepiece
+spacy>=3.6.1
+torch==2.1.0
+torchaudio>=2.0.0
+huggingface_hub
+tqdm
+transformers>=4.31.0
+demucs
+librosa
+torchmetrics
+encodec
+protobuf
+xformers --index-url https://download.pytorch.org/whl/cu118

.cog/tmp/build4114755976/requirements.txt

@@ -0,0 +1,11 @@
+av
+einops
+flashy
+gradio==3.50.2
+julius
+num2words
+omegaconf
+sentencepiece
+torchmetrics
+torch==2.1.0
+torchaudio==2.1.0

.cog/tmp/build678619104/requirements.txt

@@ -0,0 +1,21 @@
+av
+einops
+flashy>=0.0.1
+hydra-core>=1.1
+hydra_colorlog
+julius
+num2words
+numpy
+sentencepiece
+spacy>=3.6.1
+torch==2.1.0
+torchaudio>=2.0.0
+huggingface_hub
+tqdm
+transformers>=4.31.0
+demucs
+librosa
+gradio
+torchmetrics
+encodec
+protobuf

.cog/tmp/build953554414/requirements.txt

@@ -0,0 +1,21 @@
+av
+einops
+flashy>=0.0.1
+hydra-core>=1.1
+hydra_colorlog
+julius
+num2words
+numpy
+sentencepiece
+spacy>=3.6.1
+torch==2.1.0
+torchaudio>=2.0.0
+huggingface_hub
+tqdm
+transformers>=4.31.0
+demucs
+librosa
+torchmetrics
+encodec
+protobuf
+xformers --index-url https://download.pytorch.org/whl/cu118

.dockerignore

@@ -0,0 +1,17 @@
+# The .dockerignore file excludes files from the container build process.
+#
+# https://docs.docker.com/engine/reference/builder/#dockerignore-file
+
+# Exclude Git files
+.git
+.github
+.gitignore
+
+# Exclude Python cache files
+__pycache__
+.mypy_cache
+.pytest_cache
+.ruff_cache
+
+# Exclude Python virtual environment
+/venv

cog.yaml

@@ -0,0 +1,35 @@
+# Configuration for Cog ⚙️
+# Reference: https://github.com/replicate/cog/blob/main/docs/yaml.md
+
+build:
+  gpu: true
+  cuda: "11.8"
+  system_packages:
+    - "ffmpeg"
+    - "aria2"
+  python_version: "3.10"
+  python_packages:
+    - "av"
+    - "einops"
+    - "flashy>=0.0.1"
+    - "hydra-core>=1.1"
+    - "hydra_colorlog"
+    - "julius"
+    - "num2words"
+    - "numpy"
+    - "sentencepiece"
+    - "spacy>=3.6.1"
+    - "torch==2.1.0"
+    - "torchaudio>=2.0.0"
+    - "huggingface_hub"
+    - "tqdm"
+    - "transformers>=4.31.0"
+    - "demucs"
+    - "librosa"
+    - "torchmetrics"
+    - "encodec"
+    - "protobuf"
+    - "xformers --index-url https://download.pytorch.org/whl/cu118"
+
+# predict.py defines how predictions are run on your model
+predict: "predict.py:Predictor"

predict.py

@@ -0,0 +1,93 @@
+# Prediction interface for Cog ⚙️
+# https://github.com/replicate/cog/blob/main/docs/python.md
+from cog import BasePredictor, Input, Path
+import torch
+import torchaudio
+from typing import List
+from audiocraft.models import MAGNeT
+from audiocraft.data.audio import audio_write
+
+class Predictor(BasePredictor):
+    def setup(self) -> None:
+        """Load the model into memory to make running multiple predictions efficient"""
+        self.model = MAGNeT.get_pretrained("facebook/magnet-small-10secs")
+
+    @torch.inference_mode()
+    def predict(
+        self,
+        prompt: str = Input(
+            description="Input Text",
+            default="80s electronic track with melodic synthesizers, catchy beat and groovy bass"
+        ),
+        model: str = Input(
+            description="Model to use",
+            default="facebook/magnet-small-10secs",
+            choices=[
+                'facebook/magnet-small-10secs',
+                'facebook/magnet-medium-10secs',
+                'facebook/magnet-small-30secs',
+                'facebook/magnet-medium-30secs',
+                'facebook/audio-magnet-small',
+                'facebook/audio-magnet-medium']
+        ),
+        variations: int = Input(
+            description="Number of variations to generate",
+            default=3, ge=1, le=4,
+        ),
+        span_score: str = Input(
+            default="prod-stride1",
+            choices=["max-nonoverlap", "prod-stride1"],
+        ),
+        temperature: float = Input(
+            default=3.0,
+            description="Temperature for sampling",
+        ),
+        top_p: float = Input(
+            default=0.9, ge=0.0, le=1.0,
+            description="Top p for sampling",
+        ),
+        max_cfg: float = Input(
+            default=10.0,
+            description="Max CFG coefficient",
+        ),
+        min_cfg: float = Input(
+            default=1.0,
+            description="Min CFG coefficient",
+        ),
+        decoding_steps_stage_1: int = Input(
+            default=20,
+            description="Number of decoding steps for stage 1",
+        ),
+        decoding_steps_stage_2: int = Input(
+            default=10,
+            description="Number of decoding steps for stage 2",
+        ),
+        decoding_steps_stage_3: int = Input(
+            default=10,
+            description="Number of decoding steps for stage 3",
+        ),
+        decoding_steps_stage_4: int = Input(
+            default=10,
+            description="Number of decoding steps for stage 4",
+        ),
+    ) -> List[Path]:
+        """Run a single prediction on the model"""
+        descriptions = [prompt for _ in range(variations)]
+
+        self.model = MAGNeT.get_pretrained(model)
+        self.model.set_generation_params(
+            temperature=temperature,
+            top_p=top_p,
+            max_cfg_coef=max_cfg, min_cfg_coef=min_cfg, 
+            decoding_steps=[decoding_steps_stage_1, decoding_steps_stage_2, decoding_steps_stage_3, decoding_steps_stage_4],
+            span_arrangement='stride1' if (span_score == 'prod-stride1') else 'nonoverlap',)
+        wav = self.model.generate(descriptions)
+
+        for idx, one_wav in enumerate(wav):
+            audio_write(f'/tmp/{idx}', one_wav.cpu(), self.model.sample_rate, strategy="loudness", loudness_compressor=True)
+
+        output_paths = []
+        for idx in range(variations):
+            output_paths.append(Path(f'/tmp/{idx}.wav'))
+
+        return output_paths