A junk drawer for the prompt-learning–KG Venn diagram

What it does PromptKG is a research-group clearinghouse: a paper list, a few library stubs (lambdaKG for PLM-based KG embeddings, deltaKG for editing them), and tutorial notebooks. Think of it as a living literature review with occasional runnable code.

The interesting bit The repo tries to map two booming fields—prompt tuning and knowledge graphs—that keep bumping into each other but rarely share a bibliography. The “Knowledge as Prompt” vs. “Prompt for Knowledge” framing is a genuine attempt to impose order on that collision.

Key highlights

• lambdaKG: a library and benchmark for using pre-trained language models as KG embedders (link prediction, etc.)
• deltaKG: a smaller, more experimental library for dynamically editing those embeddings
• Paper list: extensive, with sections on retrieval-augmented generation, multimodal prompting, and even robot task planning
• Tutorial notebooks: pitched at beginners, though the README doesn’t say how many or how deep
• Surveys: a solid starter pack of 10+ recent survey papers, including the well-cited “Pre-train, Prompt, and Predict”

Caveats

• The README is mostly a paper dump with sparse code documentation; library usage is unclear without digging into subdirectories
• “Advanced Tasks” ranges from recommendation to robot manipulation—ambitious breadth, but depth per topic is unverified
• Last-commit badge is green, but no version releases or install instructions are visible

Verdict Grad students and researchers hunting for a foothold in prompt+KG literature should bookmark this. Production engineers looking for a pip-installable toolkit should look elsewhere—at least until lambdaKG and deltaKG grow proper docs.