Description

Project Title : opensource LLM to review the SLEPerf performance dashboard result

Problem / Pain Point:

we need review all the dashboard result links one by one and manually summary it.

Proposed Solution:

Use LLM + MCP possible to do it automatically.

MCP server can handle the communication with SLEPerf backend REST API.

LLM play the human role to review the result and give summary

Expected Benefits

Quick generate summary without human interactive 

Cross check/Double check the result compare with manually review result

Scope and Data Requirements

All Benchmark compare result.

Because this project only using offline opensource LLM, data is secure .

(we are not sure LLM is smart to analyze the result, we will find it out . ph056 has A30 NVIDIA card 24G vram) 

Technical Approach

Nvidia tool-stack,accelerate LLM

Ollama , that can launch the LLM and  provide openai like API 

Opensource LLM, try deepseek (replaceable)

MCP , we will create our SLEPerf MCP server to provide all data query functions

Risks / Limitations

Opensource LLM may not  able to analyze data because of hardware limitations.

Effort Estimate

1 Month

Owner / Contributors Jerry,Thomas

Success Criteria have a dashboard page to connecting the LLM and MCP, It can generate and summary 2 different products performance on specific benchmarks result . (Number of benchmarks depends on the context size of LLM, will try more)

Additional Notes: NONE

Goals

• Launch the offline opensource LLM and provide LLM chat api

• Create MCP server provide all data query function from SLEPerf backend

• Try it first on AI Client (VS code with Cline) , once it can work , will create wrapper can use LLM + MCP

Resources