(
batch_size: int,
min_seq_len: int,
max_seq_len: int,
max_num_logprobs: int = None,
)
| 52 | |
| 53 | |
| 54 | def _create_default_sampling_metadata( |
| 55 | batch_size: int, |
| 56 | min_seq_len: int, |
| 57 | max_seq_len: int, |
| 58 | max_num_logprobs: int = None, |
| 59 | ) -> SamplingMetadata: |
| 60 | |
| 61 | fake_sampling_metadata = SamplingMetadata( |
| 62 | temperature=paddle.full(shape=[batch_size, 1], fill_value=0.9, dtype="float32"), |
| 63 | top_p=paddle.full(shape=[batch_size, 1], fill_value=0.7, dtype="float32"), |
| 64 | prompt_ids=paddle.full(shape=[batch_size, max_seq_len], fill_value=0, dtype="int64"), |
| 65 | prompt_lens=paddle.full(shape=[batch_size, 1], fill_value=5, dtype="int64"), |
| 66 | step_idx=paddle.full(shape=[batch_size, 1], fill_value=0, dtype="int64"), |
| 67 | pre_token_ids=_create_tokens_tensor(batch_size, max_seq_len), |
| 68 | frequency_penalties=_create_penalty_tensor(batch_size, 0.0), |
| 69 | presence_penalties=_create_penalty_tensor(batch_size, 0.0), |
| 70 | repetition_penalties=_create_penalty_tensor(batch_size, 1.0), |
| 71 | min_dec_lens=paddle.full(shape=[batch_size, 1], fill_value=min_seq_len, dtype="int64"), |
| 72 | bad_words_token_ids=paddle.full(shape=[batch_size], fill_value=-1, dtype="int64"), |
| 73 | bad_words_token_len=paddle.full(shape=[batch_size, 1], fill_value=0, dtype="int64"), |
| 74 | eos_token_ids=paddle.full(shape=[batch_size], fill_value=-2, dtype="int64"), |
| 75 | min_p=paddle.randn([batch_size]), |
| 76 | seed=paddle.to_tensor([[2025]]), |
| 77 | logits_processors=None, |
| 78 | ) |
| 79 | if max_num_logprobs is not None: |
| 80 | fake_sampling_metadata.max_num_logprobs = max_num_logprobs |
| 81 | return fake_sampling_metadata |
| 82 | |
| 83 | |
| 84 | def build_config_json() -> str: |
no test coverage detected