Your question is Post-Training Pipeline for 7B Coding Model at Cohere. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Design a post-training pipeline to improve a 7B model's coding capabilities (rest of prompt hidden behind points wall).
Asked in the Round 1 system design stage. Visible text: 设计post training流程提升7B model的coding capabilities for an ...
Describe an end-to-end practical pipeline, including data collection and filtering, supervised fine-tuning, preference or reinforcement learning, evaluation, deployment, and monitoring. Explain how you would choose training data, prevent contamination and reward hacking, manage compute, detect regressions, and iterate from production feedback. State the clarifying questions and assumptions you need before committing to an architecture.