Project Title
Improve the WasmEdge-based Rust coding assistant for inference-time scaling
Description
In a previous LFX mentorship project, we have created an LLM-based coding assistant grounded in Rust programming language skills. We aim to further improve the Rust coding assistant by incorporating inference-time compute that utilizes the Rust compiler for feedback.
One of the greatest advantages of Rust is its powerful and strict compiler, and the detailed error message generated by the compiler. The Rust compiler could give valuable feedbacks to code generating LLMs to improve the code quality.
Expected Outcome
The mentee is expected to do the following.
1 Run a Qwen Coder 2.5 LLM locally or access it via an API.
2 Create an LLM system prompt that describes the structure and key elements of a cargo project. It will guide the LLM to generate multiple files (artifacts) for a complete project.
3 Create a Python program to send user requests to the LLM and parse the generated result into locally cached files.
4 Use a local Rust compiler to build the generated project. Sends the error messages back to the LLM to re-generate.
5 Iterate until there is no more errors.
6 Build a web API for the Python program that takes OpenAI compatible requests and return OpenAI compatible results.
Recommend skills
- Rust
- LlamaEdge
- LLMs like ChatGPT
- Coding assistant like GitHub Copilot
Pre-tests
Create a video demo (recorded screen cast) of yourself
1 Use the Rustcoder LLM in your cursor or Zed IDE as follows.
https://docs.gaianet.ai/agent-integrations/cursor
https://docs.gaianet.ai/agent-integrations/zed
2 Open an existing Rust cargo project in the IDE.
3 Ask the IDE to implement a feature and demonstrate the result.
Here is an example: https://x.com/juntao/status/1840445738985111653
Mentor(s)
Michael Yuan @juntao michael@secondstate.io
Apply Link
TBD
Appendix
No response
Project Title
Improve the WasmEdge-based Rust coding assistant for inference-time scaling
Description
In a previous LFX mentorship project, we have created an LLM-based coding assistant grounded in Rust programming language skills. We aim to further improve the Rust coding assistant by incorporating inference-time compute that utilizes the Rust compiler for feedback.
One of the greatest advantages of Rust is its powerful and strict compiler, and the detailed error message generated by the compiler. The Rust compiler could give valuable feedbacks to code generating LLMs to improve the code quality.
Expected Outcome
The mentee is expected to do the following.
1 Run a Qwen Coder 2.5 LLM locally or access it via an API.
2 Create an LLM system prompt that describes the structure and key elements of a
cargoproject. It will guide the LLM to generate multiple files (artifacts) for a complete project.3 Create a Python program to send user requests to the LLM and parse the generated result into locally cached files.
4 Use a local Rust compiler to build the generated project. Sends the error messages back to the LLM to re-generate.
5 Iterate until there is no more errors.
6 Build a web API for the Python program that takes OpenAI compatible requests and return OpenAI compatible results.
Recommend skills
Pre-tests
Create a video demo (recorded screen cast) of yourself
1 Use the Rustcoder LLM in your cursor or Zed IDE as follows.
https://docs.gaianet.ai/agent-integrations/cursor
https://docs.gaianet.ai/agent-integrations/zed
2 Open an existing Rust cargo project in the IDE.
3 Ask the IDE to implement a feature and demonstrate the result.
Here is an example: https://x.com/juntao/status/1840445738985111653
Mentor(s)
Michael Yuan @juntao michael@secondstate.io
Apply Link
TBD
Appendix
No response