# tinker.types.PromptAltTokens

*Generated from tinker 0.34.0 at commit [`64eef5e`](https://github.com/thinking-machines-lab/tinker/tree/64eef5e00c4dc4f55ef6521e14defde58c891c52). Source links point at that snapshot.*

## *class* [**tinker.types.PromptAltTokens**](https://github.com/thinking-machines-lab/tinker/blob/64eef5e00c4dc4f55ef6521e14defde58c891c52/src/tinker/types/prompt_alt_tokens.py#L11)()

Independent draws from the model's next-token distribution at every prompt position after the first, taken in the prefill that served the request (`SampleRequest.prompt_alt_tokens_k`).

Both tensors are dense `[len(prompt) - 1, k]`. Row `i` holds `k` draws, with replacement and at the request's temperature, from the distribution over prompt token `i + 1` (position 0 has no preceding context), so `tokens[i]` are alternatives to `prompt[i + 1]`.

**Fields:**

- [**tokens**](https://github.com/thinking-machines-lab/tinker/blob/64eef5e00c4dc4f55ef6521e14defde58c891c52/src/tinker/types/prompt_alt_tokens.py#L22) (*[TensorData](https://tinker-docs.thinkingmachines.ai/tinker/api-reference/types/tensordata/index.md)*) – int64 token ids, shape `[len(prompt) - 1, k]`.
- [**logprobs**](https://github.com/thinking-machines-lab/tinker/blob/64eef5e00c4dc4f55ef6521e14defde58c891c52/src/tinker/types/prompt_alt_tokens.py#L25) (*[TensorData](https://tinker-docs.thinkingmachines.ai/tinker/api-reference/types/tensordata/index.md)*) – float32 logprobs, shape `[len(prompt) - 1, k]`: `logprobs[i][j]` is the model's logprob of `tokens[i][j]` at prompt position `i + 1`.

## Referenced by

- [tinker.types.SampleResponse](https://tinker-docs.thinkingmachines.ai/tinker/api-reference/types/sampleresponse/index.md)