Shinichi Hemmi

shemmipfn

·

AI & ML interests

None yet

Recent Activity

new activity 14 days ago

pfnet/plamo-3-610m-fin-instruct:Wire enable_thinking through to the assistant generation prompt The generation-prompt call site passed allow_reasoning=True unconditionally, so the enable_thinking chat-template kwarg never changed the assistant header, and the prompt always advertised think support. The call site now reads the enable_thinking template variable: assistant_role_message_header( allow_reasoning=(enable_thinking if enable_thinking is defined else true), allow_tool_request=tools) Default behavior is unchanged: when enable_thinking is not supplied (or true), think is advertised as before. Assistant history rendering keeps allow_reasoning=True; only the generation prompt is affected. Verified against vLLM: resolve_chat_template_kwargs forwards enable_thinking to the template, and the PLaMo3 reasoning parser consumes the same kwarg, so the prompt header and the parser agree on whether reasoning is enabled.

new activity 14 days ago

pfnet/plamo-3-610m-fin-instruct:Wire enable_thinking through to the assistant generation prompt

liked a model 14 days ago

pfnet/plamo-3-610m-fin-instruct

View all activity

Organizations

models 0

None public yet

datasets 0

None public yet