| import sys
|
| from pathlib import Path
|
| import pytest
|
| import torch
|
|
|
| sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
|
|
|
|
|
| class FakeTokenizer:
|
| """Char-level tokenizer (ids = ord % vocab); enough for encode_record / encode_glint_record."""
|
| vocab = 997
|
| pad_token_id = 0
|
|
|
| class _Out:
|
| def __init__(self, ids):
|
| self.input_ids = ids
|
|
|
| def __call__(self, text, add_special_tokens=False):
|
| return self._Out([1 + ord(c) % (self.vocab - 1) for c in text])
|
|
|
| def convert_tokens_to_ids(self, token):
|
| return self.vocab - 1
|
|
|
|
|
| class FakeBackbone(torch.nn.Module):
|
| """Stands in for Qwen3_5ForConditionalGeneration: .model(...) -> last_hidden_state, get_output_embeddings()."""
|
|
|
| def __init__(self, vocab=FakeTokenizer.vocab, hidden=32):
|
| super().__init__()
|
|
|
| class Inner(torch.nn.Module):
|
| def __init__(self):
|
| super().__init__()
|
| self.embed = torch.nn.Embedding(vocab, hidden)
|
| self.mix = torch.nn.Linear(hidden, hidden)
|
|
|
| def forward(self, input_ids, attention_mask, use_cache=False, return_dict=True, **media):
|
| class O:
|
| pass
|
| o = O(); o.last_hidden_state = torch.tanh(self.mix(self.embed(input_ids)))
|
| return o
|
|
|
| self.model = Inner()
|
| self.lm_head = torch.nn.Linear(hidden, vocab, bias=False)
|
|
|
| def get_output_embeddings(self):
|
| return self.lm_head
|
|
|
|
|
| @pytest.fixture
|
| def tok():
|
| return FakeTokenizer()
|
|
|
|
|
| @pytest.fixture
|
| def records():
|
| choice = {"screen": {"type": "choice", "instructions": "What screen?", "criteria": {"home": "Home", "error": "Error"}},
|
| "has_error": {"type": "noul", "instructions": "Is an error visible?"}}
|
| box = {"btn": {"type": "box", "instructions": "the primary button in `screen`"},
|
| "btn_visible": {"type": "noul", "instructions": "Is the primary button visible?"},
|
| "badge": {"type": "box", "instructions": "the red badge"}}
|
| return {"choice": {"id": "c", "state": {"app": "X"}, "questions": choice},
|
| "mixed": {"id": "m", "state": "s", "questions": {**choice, **box}},
|
| "box_only": {"id": "b", "state": "s", "questions": {"btn": box["btn"]}}}
|
|
|
