update langchain agent with qwen2.5 for better accuracy

[openvino-dev-samples]

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[eaidova]

@openvino-dev-samples please update test patching:

  File "/tmp/scripts/patch_notebooks.py", line 204, in <module>
    patch_notebooks(args.notebooks_dir, args.test_device, args.skip_ov_install)
  File "/tmp/scripts/patch_notebooks.py", line 181, in patch_notebooks
    raise ValueError(f"Processing {notebookfile} failed: {source_value} does not exist in cell")
ValueError: Processing /tmp/notebooks/llm-agent-react/llm-agent-react-langchain.ipynb failed: mistralai/Mistral-7B-Instruct-v0.3 does not exist in cell

[openvino-dev-samples]

I think a small model may lead parsing error on output of agent.

[eaidova]

I think a small model may lead parsing error on output of agent.

@openvino-dev-samples ok, then I think it will be better to remove patching and disable notebook for precommit as we can not run 7B models due to github limitations (it is still will validated in internal infra with more powerful hw)

[openvino-dev-samples]

I think a small model may lead parsing error on output of agent.

@openvino-dev-samples ok, then I think it will be better to remove patching and disable notebook for precommit as we can not run 7B models due to github limitations (it is still will validated in internal infra with more powerful hw)

I'dont know if "TinyLlama/TinyLlama-1.1B-Chat-v1.0" can pass the CI, let me try

[eaidova]

I think a small model may lead parsing error on output of agent.

@openvino-dev-samples ok, then I think it will be better to remove patching and disable notebook for precommit as we can not run 7B models due to github limitations (it is still will validated in internal infra with more powerful hw)

I'dont know if "TinyLlama/TinyLlama-1.1B-Chat-v1.0" can pass the CI, let me try

what about Qwen2.5-1.5B-Instruct? do you think it less accurate than tinyllama?

[openvino-dev-samples]

I think a small model may lead parsing error on output of agent.

@openvino-dev-samples ok, then I think it will be better to remove patching and disable notebook for precommit as we can not run 7B models due to github limitations (it is still will validated in internal infra with more powerful hw)

I'dont know if "TinyLlama/TinyLlama-1.1B-Chat-v1.0" can pass the CI, let me try

what about Qwen2.5-1.5B-Instruct? do you think it less accurate than tinyllama?

It depends on how good LLM can follow prompt's instruction, Let me try ting-llama first