DeepForgeby svx

0/5730 solved

← All interview tracks

DL engineers · kernel authorsKernels · numerics · distributed training

NVIDIA · Deep Learning Engineer

Kernel-level efficiency: quantization scales, attention tiling, occupancy, and the memory math that decides whether a model fits.

3 phases · 26 problems · 26 unique

Start timed mock5 problems · 4 min each

Your progress

0/26 · 0%

Continue track · Model Size (MB)

1Warm-up

0/8

Model Size (MB)
dl-057Deep Learning
Easy
Int8 Quantization Scale
dl-058Deep Learning
Easy
Quantize Int8
dl-059Deep Learning
Easy
Dequantize Int8
dl-060Deep Learning
Easy
Throughput Estimate
dl-101Deep Learning
Easy
Latency Estimate
dl-102Deep Learning
Easy
Conv Kernel Tiling
dl-276Deep Learning
Easy
Tensor-Core Fragment Shape
dl-277Deep Learning
Easy

2Core

0/10

KV Cache Size
dl-075Deep Learning
Medium
Per-Channel Quant Scale
dl-077Deep Learning
Medium
Flash Attention Block Count
dl-124Deep Learning
Medium
Online Softmax Rescale Step
dl-211Deep Learning
Medium
Dynamic Batch Admission
dl-213Deep Learning
Medium
Winograd Transform Lite
dl-292Deep Learning
Medium
Occupancy Calculator
dl-295Deep Learning
Medium
Shared-Memory Bank Conflicts
dl-296Deep Learning
Medium
Quantized Inference Latency
dl-301Deep Learning
Medium
Batch-Latency Tradeoff
dl-305Deep Learning
Medium

3Hard

0/8

Inference FLOPs Estimate
dl-093Deep Learning
Hard
Flash Decode Block Split
dl-224Deep Learning
Hard
Roofline Bound Check
dl-228Deep Learning
Hard
MFU Utilization
dl-229Deep Learning
Hard
Fused Attention SRAM Tiles
dl-272Deep Learning
Hard
CPU Offload Time
dl-318Deep Learning
Hard
Software Pipeline Depth
dl-319Deep Learning
Hard
Register Pressure Estimate
dl-320Deep Learning
Hard