simpletuner/minimax-music3

Public

7 runs

Run simpletuner/minimax-music3 with an API

Use one of our client libraries to get started quickly. Clicking on a library will take you to the Playground tab where you can tweak different inputs, see the results, and copy the corresponding code to use in your own project.

Input schema

The fields you can use to run this model with an API. If you don't give a value for a field its default value will be used.

Field	Type	Default value	Description
train_data	string		ZIP or tar of matching .wav/.mp3/.flac, .txt style captions, and .lyrics files
regularisation_data	string		Optional audio/.txt/.lyrics archive for frozen-base prior preservation
trigger_word	string		Optional instance prompt replacing per-track captions
max_train_steps	integer	750 Min: 1 Max: 50000	Optimizer updates
checkpoint_interval	integer	250 Min: 1 Max: 5000	Save every N updates
lora_rank	integer	64 Min: 1 Max: 256	LoRA rank
lora_dropout	number	0 Max: 0.5	LoRA dropout
learning_rate	number	0.00008 Min: 0.0000001 Max: 0.001	Learning rate
window_mode	None	continuation	Continuation supervises the end of a context window; prefix and random supervise the entire selected window
continuation_crop_mode	None	full	Full keeps the song opening; random may start later while retaining causal context
caption_dropout	number	0 Max: 1	Probability of omitting style captions while retaining lyrics
max_audio_duration	number	30 Min: 5.12 Max: 300	Maximum visible training window in seconds
enable_regional_compile	boolean	True	Compile repeated transformer blocks
attention_mechanism	None	kohaku-fa-auto	Attention backend
gradient_checkpointing	boolean	True	Recompute language-model activations to reduce memory use
dataloader_prefetch	boolean	True	Prepare upcoming training batches in the background
seed	integer	42	Training seed
hub_model_id	string		Optional Hugging Face repository for the trained adapter
hf_token	string		Write-capable Hugging Face token for uploading the adapter
push_checkpoints_to_hub	boolean	True	Upload intermediate checkpoints to the Hub in the background

{
  "type": "object",
  "title": "Input",
  "required": [
    "train_data"
  ],
  "properties": {
    "seed": {
      "type": "integer",
      "title": "Seed",
      "default": 42,
      "minimum": 0,
      "x-order": 16,
      "description": "Training seed"
    },
    "hf_token": {
      "type": "string",
      "title": "Hf Token",
      "format": "password",
      "x-order": 18,
      "nullable": true,
      "writeOnly": true,
      "description": "Write-capable Hugging Face token for uploading the adapter",
      "x-cog-secret": true
    },
    "lora_rank": {
      "type": "integer",
      "title": "Lora Rank",
      "default": 64,
      "maximum": 256,
      "minimum": 1,
      "x-order": 5,
      "description": "LoRA rank"
    },
    "train_data": {
      "type": "string",
      "title": "Train Data",
      "format": "uri",
      "x-order": 0,
      "description": "ZIP or tar of matching .wav/.mp3/.flac, .txt style captions, and .lyrics files"
    },
    "window_mode": {
      "enum": [
        "continuation",
        "prefix",
        "random"
      ],
      "type": "string",
      "title": "window_mode",
      "description": "Continuation supervises the end of a context window; prefix and random supervise the entire selected window",
      "default": "continuation",
      "x-order": 8
    },
    "hub_model_id": {
      "type": "string",
      "title": "Hub Model Id",
      "x-order": 17,
      "nullable": true,
      "description": "Optional Hugging Face repository for the trained adapter"
    },
    "lora_dropout": {
      "type": "number",
      "title": "Lora Dropout",
      "default": 0,
      "maximum": 0.5,
      "minimum": 0,
      "x-order": 6,
      "description": "LoRA dropout"
    },
    "trigger_word": {
      "type": "string",
      "title": "Trigger Word",
      "default": "",
      "x-order": 2,
      "description": "Optional instance prompt replacing per-track captions"
    },
    "learning_rate": {
      "type": "number",
      "title": "Learning Rate",
      "default": 8e-05,
      "maximum": 0.001,
      "minimum": 1e-07,
      "x-order": 7,
      "description": "Learning rate"
    },
    "caption_dropout": {
      "type": "number",
      "title": "Caption Dropout",
      "default": 0,
      "maximum": 1,
      "minimum": 0,
      "x-order": 10,
      "description": "Probability of omitting style captions while retaining lyrics"
    },
    "max_train_steps": {
      "type": "integer",
      "title": "Max Train Steps",
      "default": 750,
      "maximum": 50000,
      "minimum": 1,
      "x-order": 3,
      "description": "Optimizer updates"
    },
    "max_audio_duration": {
      "type": "number",
      "title": "Max Audio Duration",
      "default": 30,
      "maximum": 300,
      "minimum": 5.12,
      "x-order": 11,
      "description": "Maximum visible training window in seconds"
    },
    "attention_mechanism": {
      "enum": [
        "kohaku-fa-auto",
        "native-efficient",
        "diffusers"
      ],
      "type": "string",
      "title": "attention_mechanism",
      "description": "Attention backend",
      "default": "kohaku-fa-auto",
      "x-order": 13
    },
    "checkpoint_interval": {
      "type": "integer",
      "title": "Checkpoint Interval",
      "default": 250,
      "maximum": 5000,
      "minimum": 1,
      "x-order": 4,
      "description": "Save every N updates"
    },
    "dataloader_prefetch": {
      "type": "boolean",
      "title": "Dataloader Prefetch",
      "default": true,
      "x-order": 15,
      "description": "Prepare upcoming training batches in the background"
    },
    "regularisation_data": {
      "type": "string",
      "title": "Regularisation Data",
      "format": "uri",
      "x-order": 1,
      "nullable": true,
      "description": "Optional audio/.txt/.lyrics archive for frozen-base prior preservation"
    },
    "continuation_crop_mode": {
      "enum": [
        "full",
        "random"
      ],
      "type": "string",
      "title": "continuation_crop_mode",
      "description": "Full keeps the song opening; random may start later while retaining causal context",
      "default": "full",
      "x-order": 9
    },
    "gradient_checkpointing": {
      "type": "boolean",
      "title": "Gradient Checkpointing",
      "default": true,
      "x-order": 14,
      "description": "Recompute language-model activations to reduce memory use"
    },
    "enable_regional_compile": {
      "type": "boolean",
      "title": "Enable Regional Compile",
      "default": true,
      "x-order": 12,
      "description": "Compile repeated transformer blocks"
    },
    "push_checkpoints_to_hub": {
      "type": "boolean",
      "title": "Push Checkpoints To Hub",
      "default": true,
      "x-order": 19,
      "description": "Upload intermediate checkpoints to the Hub in the background"
    }
  }
}

Output schema

The shape of the response you’ll get when you run this model with an API.

Schema

{
  "type": "string",
  "title": "Output",
  "format": "uri"
}