| """Configuration for a Qwen backbone with a dynamic-option decision head.""" |
| import math |
| from transformers.models.qwen3_5.configuration_qwen3_5 import Qwen3_5Config |
|
|
|
|
| class AutoJevConfig(Qwen3_5Config): |
| model_type = "autojev_decision" |
|
|
| def __init__(self, codes=None, token_ids=None, max_options=255, temperature=1.0, |
| base_model="Qwen/Qwen3.8-27B", initialization_step=0, |
| max_input_length=8192, **kwargs): |
| super().__init__(**kwargs) |
| if max_options != 255: |
| raise ValueError("This artifact uses exactly 255 option positions") |
| if not math.isfinite(temperature) or temperature <= 0: |
| raise ValueError("temperature must be positive and finite") |
| self.max_options = max_options |
| self.codes = list(codes or []) |
| self.token_ids = list(token_ids or []) |
| if self.codes or self.token_ids: |
| if len(self.codes) != max_options or len(set(self.codes)) != max_options: |
| raise ValueError("Expected 255 unique answer codes") |
| if len(self.token_ids) != max_options or len(set(self.token_ids)) != max_options: |
| raise ValueError("Expected 255 unique answer token IDs") |
| if any(type(i) is not int or not 0 <= i < self.text_config.vocab_size for i in self.token_ids): |
| raise ValueError("Answer token IDs outside base vocabulary") |
| self.temperature = float(temperature) |
| self.base_model = base_model |
| self.initialization_step = initialization_step |
| self.max_input_length = int(max_input_length) |
| if self.max_input_length < 1: |
| raise ValueError("max_input_length must be positive") |
|
|
