Text Generation
Transformers
Safetensors
PyTorch
English
nvidia
conversational

Updated to handle multi-turn correctly

#2
by davidmnv - opened
NVIDIA org

The new tokenizer renders the messages correctly compared to the old:

<SPECIAL_10>System
{'reasoning': True}
<SPECIAL_11>User
Hi there
<SPECIAL_11>Assistant
Hello! How can I assist you today?
<SPECIAL_11>User
Can you remember this number for me? 8102250
<SPECIAL_11>Assistant
<TOOLCALL>remember_number(number="8102250")</TOOLCALL>
<SPECIAL_11>User
How are you doing?
<SPECIAL_11>Assistant
<think>
davidmnv changed pull request status to closed

Sign up or log in to comment