| Parameters | Type (Required) | Default Value | Description |
|---|---|---|---|
| Temperature (0~1) | Number(x) | 1 | Controls the range and diversity of responses by influencing response selection probabilities. Values close to 0 make the model choose the highest probability next word, while values close to 1 yield more varied results. Generally, for predictable responses, set closer to 0; for creative and diverse responses, set closer to 1. |
| Top-K (1~) | Number(x) | 10 | Randomly selects from the top k most probable next tokens. |
| Top-P (0~1) | Number(x) | 1 | Selects only from candidates within the top cumulative probability distribution (p%). Higher p values yield more varied word choices. |
| Repetition Penalty (1~) | Number(x) | 1.15 | Higher values help prevent repeated word generation by applying a penalty to recurring tokens, reducing their probability of being selected again. |
| Max token Length (1~2048) | Number(x) | 1024 | Limits the response to a specific length. (For Mi:dm-text-S, it's 1 to 1024.) |
| Number of Beams (1~) | Number(x) | 1 | Specifies the range of word options during decoding. Higher values increase the variety of candidate words, enhancing the model's ability to generate longer, more complex responses by considering multiple options. Conversely, lower values reduce computational cost, resulting in faster responses. |
| Presence penalty | Number(x) | 0 | A value between -2.0 and 2.0. Positive values impose a penalty based on whether new tokens appear in the existing text, thereby increasing the likelihood of the model discussing new topics. |
| Stop words | List(String) | [] | Up to four sequences can be specified, and the API will terminate token generation upon the occurrence of any of these sequences. |
| Inference seed | Number(x) | -1 | The seed controls the reproducibility of the task. Using the same seed and task parameters should produce the same results, though slight differences may occasionally occur. If no seed is specified, one is generated automatically. |
| Log probs | Number(x) | 0 | An integer between 0 and 20 that specifies the number of tokens with the highest probability to return at each token position. Each token is accompanied by its log probability. |
