Add dora-lmdeploy multimodel

[7SOMAY]

Add dora-lmdeploy Node for Efficient LLM Inference

Summary

This PR - #817 introduces the dora-lmdeploy node, which leverages LMDeploy's Turbomind engine for efficient inference of large language models. It supports both text and image inputs, making it suitable for multimodal tasks.

Key Features

• Efficient Inference: Optimized for CUDA performance using Turbomind.
• Multimodal Support: Handles text and image inputs.
• Configurable Models: Supports various models like InternLM and Qwen.
• Conversation History: Maintains context for responses.

Configuration Example

env:
  MODEL_NAME: "internlm/internlm2-7b"
  MAX_LENGTH: 2048
  TEMPERATURE: 0.7
  TURBOMIND_CACHE_DIR: "./workspace"

Testing Status

The node is under working condition; however, testing is pending due to an incompatible version of LMDeploy with my Mac. I am currently working on finding a solution to resolve this issue.

Checklist

• [x] Added tests
• [x] Updated documentation
• [ ] Example configurations included

[7SOMAY]

@haixuanTao, The lmdeploy doesn't support the macos platform. What should be the way I should handle that? Can I make it available for windows and linux?

[7SOMAY]

Please refer this Issue: https://github.com/InternLM/MindSearch/issues/46

Potential solution: https://github.com/InternLM/MindSearch/issues/46#issuecomment-2343880680

[haixuanTao]

I think it would be better to implement this using a compatible OS otherwise it's going to be very hard to debug