Page MenuHomePhabricator

Phase 2: Article categorization metrics, fine-tuning metrics, optimization tooling
Closed, ResolvedPublic

Description

  • Goals:
    • Generate metrics from first phase for article categorization(baseline datasets from Aitolkyn, prompting strategy from Mykola)
    • Generate metrics for optimized tooling (quantized models; using onnx, ggml/llama.cpp)
    • Generate metrics for model fine-tuning (what LLM models can we fine-tune with WMF infra), without focus on improving model quality
  • Infrastructure: the ml-lab instances that each have two AMD Instinct MI210 GPUs.
  • Stretch goals:
    • Explore model parallel inference (as we have 2 GPU per instance)
    • Fine-tuned model for article categorization classification task (e.g. no just for metrics, try to learn a good model)

Details

Due Date
Dec 6 2024, 12:00 AM

Event Timeline

leila triaged this task as High priority.Oct 23 2024, 4:08 PM
leila updated the task description. (Show Details)
leila set Due Date to Dec 6 2024, 12:00 AM.
leila subscribed.

We discussed this task in backlog refinement today. set the deadline and priority based on earlier conversations and agreements.

Stretch goals may not be completed by end of Q2 - will continue in Q3.

@MunizaA can you update the status of this task, please? Thanks and let me know where I can help.

This work is in this repo. It includes extendible ways to compute metrics (including evaluation and latency/performance), future needs for new measurements/environments can be use this tooling.