Implement LoRA for ML Models | Dataford Interview Questions

Implement LoRA (low-rank adaptation) as an ML coding exercise. Asked in the Tech screen stage for an NLP/LM research engineer team, followed by basic ML questions and resume project discussion.

Implement lora_forward(x, W, A, B, alpha), where x is [batch, in_features], W is [in_features, out_features], A is [in_features, rank], and B is [rank, out_features]. Return xW + (alpha / rank)xAB as a nested list. Do not use NumPy or other matrix libraries.

Constraints: all dimensions are positive, rank <= in_features, and values are finite numbers.

Constraints

1 <= batch, in_features, out_features <= 64
1 <= rank <= in_features
x has shape [batch, in_features]
W has shape [in_features, out_features]
A has shape [in_features, rank]
B has shape [rank, out_features]
All values are finite numbers

Function Signature

def lora_forward(x, W, A, B, alpha):

Your question is Implement LoRA for ML Models. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.

Problem

Implement LoRA (low-rank adaptation) as an ML coding exercise. Asked in the Tech screen stage for an NLP/LM research engineer team, followed by basic ML questions and resume project discussion.

Constraints: all dimensions are positive, rank <= in_features, and values are finite numbers.

Constraints

1 <= batch, in_features, out_features <= 64
1 <= rank <= in_features
x has shape [batch, in_features]
W has shape [in_features, out_features]
A has shape [in_features, rank]
B has shape [rank, out_features]
All values are finite numbers

Function Signature

def lora_forward(x, W, A, B, alpha):

Your question is Implement LoRA for ML Models. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.