[WIP]Add text tagging by prompt mapper op #408
Conversation
| skip_special_tokens=True) | ||
|
|
||
| text_tags = [] | ||
| text_tags.append(output) |
There was a problem hiding this comment.
It is better to strip output
| """ # noqa | ||
|
|
||
| super().__init__(*args, **kwargs) | ||
| self.num_proc = 1 |
There was a problem hiding this comment.
If enable_vllm is False, num_proc=1 will force this OP to run in single process/GPU. Is this the desired behavior?
|
|
||
| text_tags = [] | ||
| text_tags.append(output) | ||
| sample[Fields.text_tags] = text_tags |
There was a problem hiding this comment.
Please refer to #423 for adding user-specified tag field name.
| if not string_list: | ||
| assert False, "输入的列表不能是空的" | ||
|
|
||
| for string in string_list: |
There was a problem hiding this comment.
Why not directly check output in string_list?
| max_model_len=1024, | ||
| max_num_seqs=16, | ||
| sampling_params={'temperature': 0.1, 'top_p': 0.95, 'max_tokens': 256}) | ||
|
|
There was a problem hiding this comment.
Can this OP run in multiple processes, especially without vllm? Please add more tests.
| max_num_seqs=16, | ||
| sampling_params={'temperature': 0.1, 'top_p': 0.95, 'max_tokens': 256}) | ||
|
|
||
|
|
There was a problem hiding this comment.
It would be better to add tests for tensor_parallel_size.
|
This PR is marked as stale because there has been no activity for 21 days. Remove stale label or add new comments or this PR will be closed in 3 day. |
|
Close this stale PR. |
As the title says.
Use llm to perform arbitrary tagging and classification.