Skip to content
Skills
Problems
Questions
Roles
Skills
Interview
Contest
Leaderboard
More
Problems
Questions
Roles
Skills
Interview
Contest
Leaderboard
More
Home
/
MLOps Engineer
/
Questions
/
GPUs and distributed training
MLOps Engineer
GPUs and distributed training questions
15 questions on GPUs and distributed training for mlops engineers, easiest first.
Easy
1
Memory for 7B parameters in fp16
easy
Medium
7
Scheduling a GPU pod
medium
Low GPU use under load
medium
KV cache size for one long request
medium
Training on spot GPUs
medium
Speeding up training on 8 GPUs
medium
Why bf16 is easier than fp16
medium
CPU pods filling GPU nodes
medium
Hard
7
Why LLM servers use continuous batching
hard
Same seed, different results
hard
Where to use tensor parallelism
hard
KV cache with grouped-query attention
hard
Memory to fully fine-tune 7B with Adam
hard
GPU shows 100% but is underused
hard
Loss spike after resuming training
hard
