v0.0.15

• support optional image prefix for vl fn-call
• support video understanding

v0.0.14

• support completion interface for oai
• add qwen-audio support and support env for default settings
• rm upload image prefix

v0.0.13

• update modelscope_studio version and fix Windows encoding issues

v0.0.12

• upgrade gui to gradio 5 based

v0.0.11

• fix signal conflicts in code interpreter
• support nous tool-call prompt

v0.0.10

• Add the Tool-Integrated Reasoning (TIR) agent for Qwen2.5-Math.
• Add support for caching the LLM's responses. LLM caching can be enabled by providing cache_dir in the LLM's config, i.e.,llm = get_chat_model(cfg={..., "cache_dir": "/tmp/directory_to_save_llm_cache"}).
• Add self-play agents to simulate a dialogue between a user and an AI agent, mainly for data synthesis and testing purposes.

v0.0.9

• Bug fix: Incomplete PyPI packaging caused by the absence of __init__.py.

v0.0.8

• Multimodal function calling for Qwen2-VL. Please see the function calling example and the agent example for example usage.

v0.0.7

• Make dependencies for the code interpreter and RAG optional.
  • pip install qwen-agent now only installs minimal dependencies for function calling.
  • pip install qwen-agent[rag,code_interpreter,gui] installs optional dependencies for RAG, Code Interpreter, and GUI.

v0.0.6

• Added experimental support for parallel tool calling and tool choice.