Datasets
Models
Downloads: 1
Diff of
/Code/Drug Discovery/LLM-RAG/Drug Synthesis/README.md
[000000]
..
[404218]
Switch to unified view
| a | b/Code/Drug Discovery/LLM-RAG/Drug Synthesis/README.md | ||
|---|---|---|---|
| 1 | ## [Seminar PDF: Drug Synthesis and 2024 LLM Inferencing](https://drive.google.com/file/d/1kvjRnw1QDM-ZqRQNj2Jj-cCiPqKu5mWM/view?usp=sharing) |
||
| 2 | |||
| 3 | 2024 LLMs: Faster than PhD level chemists, but quality across many prompts needs improvement. |
||
| 4 | |||
| 5 | A) Many low cost/no cost LLMs with emergent properties exist for developing new medicine. |
||
| 6 | |||
| 7 | B) Explainability Matters - some generations reveal a complete summary of their findings. Document, page number, selection of text, and line number IDs need improvement. Subsequent synthetic chemistry steps improve if humans and other software have more confidence with current step explanations. |
||
| 8 | |||
| 9 | C) Context Matters - LLM Retrieval Augmented Generations pulls document info w/ context. LLM further uses its own context to solve problems in addition to document details. |
||
| 10 | |||
| 11 | D) Speed Matters - More Synthetic Chemistry Reactions run in a single day with more speed. High Tokens/Second is a function of model size/architecture and API hardware. More Reactions/Day assist projects with deadlines greater when speed is higher. |
||
| 12 | |||
| 13 | 35 Minute Future of Industry [Discussion](https://www.youtube.com/watch?v=U467_vWLShc), see slide 04 Example. |
||
| 14 | |||
| 15 | 25Jul24, Kevin Kawchak |