Get model limits - Together AI docs

{ "model_name": "<string>", "lora_training": { "max_batch_size": 123, "max_batch_size_dpo": 123, "min_batch_size": 123, "max_rank": 123, "target_modules": [ "<string>" ] }, "max_num_epochs": 123, "max_num_evals": 123, "max_learning_rate": 123, "min_learning_rate": 123, "supports_full_training": true, "supports_vision": true, "supports_tools": true, "supports_reasoning": true, "merge_output_lora": true, "default_gradient_accumulation_steps": 123, "max_num_checkpoints": 123, "min_max_seq_length": 123, "max_seq_length_sft": 123, "max_seq_length_dpo": 123, "full_training": { "max_batch_size": 123, "max_batch_size_dpo": 123, "min_batch_size": 123 } }

Authorizations

Authorization

string

header

default:default

required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Query Parameters

model_name

string

required

The model name to get limits for.

Response

Model limits.

Model limits for fine-tuning.

model_name

string

required

The name of the model.

lora_training

object

required

Limits for LoRA training.

Show child attributes

max_num_epochs

integer

required

Maximum number of training epochs.

max_num_evals

integer

required

Maximum number of evaluations.

max_learning_rate

number

required

Maximum learning rate.

min_learning_rate

number

required

Minimum learning rate.

supports_full_training

boolean

required

Whether the model supports full (non-LoRA) fine-tuning. When false, only LoRA fine-tuning is available and the full_training limits are reported as zero.

supports_vision

boolean

required

Whether the model supports vision/multimodal inputs.

supports_tools

boolean

required

Whether the model supports tool/function calling.

supports_reasoning

boolean

required

Whether the model supports reasoning.

merge_output_lora

boolean

required

Whether a merged checkpoint (the base model with the trained LoRA adapter fused in) is produced for LoRA fine-tunes of this model, in addition to the standalone adapter.

default_gradient_accumulation_steps

integer

required

Default gradient accumulation steps used when a fine-tune request omits the value or sets it to 0.

max_num_checkpoints

integer

required

Maximum number of checkpoints that can be saved during a fine-tuning job.

min_max_seq_length

integer

required

Minimum value allowed for the max_seq_length hyperparameter.

max_seq_length_sft

integer

required

Maximum sequence length supported for SFT training.

max_seq_length_dpo

integer

required

Maximum sequence length supported for DPO training.

full_training

object

Limits for full training.

Show child attributes