Support passing additional params to apply_chat_template by albertvillanova · Pull Request #1902 · huggingface/smolagents (original) (raw)

Support passing additional params to tokenizer.apply_chat_template in:

• VLLMModel
• TransformersModel

by using the apply_chat_template_kwargs param at model instantiation.

Previuosly done for MLXModel in:

• Support passing additional params to MLXModel load and tokenizer.apply_chat_template #1406

Context

As requested by @sergiopaniego, a use case requires passing enable_thinking=False to apply_chat_template.

model = TransformersModel( model_id="Qwen/Qwen3-1.7B", apply_chat_template_kwargs={"enable_thinking": False}, )

CC: @sergiopaniego