📍 VOD Development Roadmap V0

[MotzWanted]

We summarize our planned features in this issue. This issue will keep being updated.

Dataloaders: Support more datasets for common IR & Gen AI

Primary:

• [x] MSMarco
• [x] Natural Question
• [ ] ELI5
• [ ] MMLU
• [ ] Wikipedia
• [ ] WikiHow

Secondary:

• [x] Interface for external datasets

Modelling: REALM for Generative Tasks

Primary:

• [x] Implement a REALM model (Encoder + OSS Generative LLM)

Secondary:

• [x] Handle LoRA parameterization
• [ ] Handle DeepSpeed

Search API: Add Filtering Capabilities

Primary:

• [x] Qdrant backend

Secondary:

• [ ] vector support in elasticsearch

Gradients: VOD for Generative Tasks

Primary:

• [ ] Implement VOD gradients for generative tasks
• [ ] Test VOD gradients on MNIST

UX: first-steps

Primary:

• [x] Make the code plug-an-play / extendable
• [x] Simplify existing workflows