From fa6d8cefd6a2666705eae832676e56b4b40ffab3 Mon Sep 17 00:00:00 2001 From: Prashant Gupta Date: Fri, 11 Oct 2024 15:51:47 -0700 Subject: [PATCH] =?UTF-8?q?=E2=99=BB=EF=B8=8F=20use=20generic=20check=20fo?= =?UTF-8?q?r=20encode=5Fplus?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Prashant Gupta --- src/vllm_tgis_adapter/grpc/grpc_server.py | 14 ++++++-------- 1 file changed, 6 insertions(+), 8 deletions(-) diff --git a/src/vllm_tgis_adapter/grpc/grpc_server.py b/src/vllm_tgis_adapter/grpc/grpc_server.py index 9fe9bd8..43c1e02 100644 --- a/src/vllm_tgis_adapter/grpc/grpc_server.py +++ b/src/vllm_tgis_adapter/grpc/grpc_server.py @@ -23,7 +23,6 @@ log_tracing_disabled_warning, ) from vllm.transformers_utils.tokenizer import AnyTokenizer # noqa: TCH002 -from vllm.transformers_utils.tokenizers.mistral import MistralTokenizer from vllm.utils import iterate_with_cancellation from vllm_tgis_adapter.logging import init_logger @@ -856,12 +855,16 @@ async def Tokenize( tokenizer = await self._get_tokenizer(adapter_kwargs) responses: list[TokenizeResponse] = [] - is_mistral_tokenizer = isinstance(tokenizer, MistralTokenizer) # TODO: maybe parallelize, also move convert_ids_to_tokens into the # other threads for req in request.requests: - if is_mistral_tokenizer: + if not hasattr(tokenizer, "encode_plus"): + if request.return_offsets: + raise ValueError( + f"{type(tokenizer)} doesn't support " + "return_offsets at the moment. " + ) token_ids = tokenizer.encode( prompt=req.text, ) @@ -885,11 +888,6 @@ async def Tokenize( offsets = None if request.return_offsets: - if is_mistral_tokenizer: - raise ValueError( - "Mistral tokenizer doesn't support " - "return_offsets at the moment. " - ) offsets = [ {"start": start, "end": end} for start, end in batch_encoding.offset_mapping