diff --git a/README.md b/README.md index 845db067..7978468c 100644 --- a/README.md +++ b/README.md @@ -9,9 +9,8 @@

- GitHub stars PyPI version - PyPI downloads + PyPI Downloads License

diff --git a/optillm/__init__.py b/optillm/__init__.py index b18101a6..0aad5ab0 100644 --- a/optillm/__init__.py +++ b/optillm/__init__.py @@ -1,5 +1,5 @@ # Version information -__version__ = "0.3.21" +__version__ = "0.3.22" import os as _os diff --git a/optillm/autothink/processor.py b/optillm/autothink/processor.py index 2fcb4998..bade0fda 100644 --- a/optillm/autothink/processor.py +++ b/optillm/autothink/processor.py @@ -243,7 +243,7 @@ def process(self, messages: List[Dict[str, str]]) -> str: thinking_messages, continue_final_message=True, return_tensors="pt" - ).to(self.model.device) + ).input_ids.to(self.model.device) # Reset and update token history in steering hooks if self.steering_hooks: diff --git a/optillm/deepconf/processor.py b/optillm/deepconf/processor.py index 17030269..14fc31fb 100644 --- a/optillm/deepconf/processor.py +++ b/optillm/deepconf/processor.py @@ -108,7 +108,7 @@ def generate_single_trace(self, messages: List[Dict[str, str]], messages, return_tensors="pt", add_generation_prompt=True - ).to(self.model.device) + ).input_ids.to(self.model.device) # Initialize generation state kv_cache = DynamicCache() diff --git a/optillm/thinkdeeper.py b/optillm/thinkdeeper.py index 321e9520..56a2a8ca 100644 --- a/optillm/thinkdeeper.py +++ b/optillm/thinkdeeper.py @@ -77,8 +77,7 @@ def reasoning_effort(self, messages) -> str: messages, continue_final_message=True, return_tensors="pt" - ) - tokens = tokens.to(self.model.device) + ).input_ids.to(self.model.device) kv = DynamicCache() n_thinking_tokens = 0 diff --git a/pyproject.toml b/pyproject.toml index 5a69a541..7e2159ca 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta" [project] name = "optillm" -version = "0.3.21" +version = "0.3.22" description = "An optimizing inference proxy for LLMs." readme = "README.md" license = "Apache-2.0"