Models

List Together AI models and upload your own from Hugging Face or S3.

Upload

Upload a model from Hugging Face or S3 for inference on a dedicated endpoint.

Basic

tg models upload \
    --model-name [TEXT] \
    --model-source [URI]

# Upload a model from Hugging Face.
tg models upload \
    --model-name together-m1-3b-personal-clone \
    --model-source https://huggingface.co/togethercomputer/M1-3B \
    --hf-token "$HUGGINGFACEHUB_API_TOKEN"

# Upload a model from S3.
PRESIGNED_URL=$(sh ./get-presigned-url)

tg models upload \
  --model-name my-s3-upload-model \
  --model-source "$PRESIGNED_URL"

Parameters

Flag	Type	Description
--model-name	string	The name to give your uploaded model. required
--model-source	string	The source URI of the model. required
--model-type	model or adapter	Whether the model is a full model or an adapter.
--hf-token	string	Hugging Face token, used when uploading from Hugging Face.
--description	string	A description of your model.
--base-model	string	The base model for an adapter when running against a serverless pool. Only used with --model-type adapter.
--lora-model	string	The LoRA pool for an adapter when running against a dedicated pool. Only used with --model-type adapter.

List all models

Basic

# List models
tg models list

# List models that can be deployed on dedicated endpoints
tg models list --type dedicated

# Output in JSON mode and pipe to jq.
tg models list --json | jq 'length'

Parameters

Flag	Type	Description
--type	dedicated	Filter to models that can be deployed on dedicated endpoints. dedicated is the only available value.
--after	string	The cursor to start from for pagination.