tinker.types
All public types in the tinker.types module, organized by function.
Data & Input
Types for preparing training data and model inputs.
| Type | Description |
|---|---|
Datum |
A single training example with input tokens and loss mask |
ModelInput |
Token sequence input for sampling or training |
TensorData |
Tensor container with dtype and shape metadata |
EncodedTextChunk |
A chunk of encoded text tokens |
ImageChunk |
Image data chunk for vision-language models |
ImageAssetPointerChunk |
Pointer to a pre-uploaded image asset |
Training
Types for configuring and observing training operations.
| Type | Description |
|---|---|
ForwardBackwardOutput |
Result of a forward-backward pass (loss, gradients, metrics) |
OptimStepResponse |
Result of an optimizer step |
AdamParams |
Adam optimizer hyperparameters (learning rate, betas, epsilon) |
LoraConfig |
LoRA adapter configuration (rank, which layers to train) |
GetInfoResponse |
Information about a training client's model |
ModelData |
Metadata about a model's architecture and configuration |
Sampling & Inference
Types for text generation and sampling results.
| Type | Description |
|---|---|
SamplingParams |
Sampling configuration (temperature, max tokens, stop sequences) |
SampleResponse |
Response from a sampling request |
SampledSequence |
A single generated sequence with tokens and log-probabilities |
Models & Checkpoints
Types for managing models, checkpoints, and training runs.
| Type | Description |
|---|---|
Checkpoint |
A saved model checkpoint |
ParsedCheckpointTinkerPath |
Parsed components of a tinker:// checkpoint path |
TrainingRun |
Metadata about a training run |
SupportedModel |
Information about a model supported by the server |
GetServerCapabilitiesResponse |
Server capabilities including supported models |
CheckpointsListResponse |
Paginated list of checkpoints |
CheckpointArchiveUrlResponse |
Download URL for a checkpoint archive |
TrainingRunsResponse |
Paginated list of training runs |
Cursor |
Pagination cursor for list responses |
SaveWeightsResponse |
Response from saving model weights |
SaveWeightsForSamplerResponse |
Response from saving weights for a sampler |
WeightsInfoResponse |
Minimal information for loading public checkpoints |