Function calling ability not stable after mutli-rounds conversations
#32
by
xl2533
- opened
when using deepseek v3.2 in multi-round conversations, each round envolve multiple funtion calls, after 6 ~8 rounds of conversations, the model occasionly stop function calling, and answer the question directly.
When processing the context, we drop the intermediate function call and function result, only keep the conversation input and output for each round.
Does anyone encounter the simliary issue?