Modified Attention Mechanism Implementation | Dataford Interview Questions

Implement the attention mechanism with specific modifications.

Implement attention(q, k, v, causal, mask, temperature) using scaled dot-product attention. Apply causal masking and an optional boolean mask, then divide logits by temperature before softmax. Return [output, weights], where both values are nested Python lists. Round every returned floating-point value to 6 decimal places. Inputs contain one attention head, and every query has at least one permitted key.

Constraints

1 <= len(q) <= 256
len(q) == len(mask) when mask is provided
1 <= len(k) == len(v) <= 256
1 <= len(q[i]) == len(k[j])
len(v[j]) >= 1
temperature > 0
Every query has at least one permitted key
All numeric entries are finite real numbers

Function Signature

def attention(q, k, v, causal, mask, temperature):

Your question is Modified Attention Mechanism Implementation. 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 the attention mechanism with specific modifications.

Constraints

1 <= len(q) <= 256
len(q) == len(mask) when mask is provided
1 <= len(k) == len(v) <= 256
1 <= len(q[i]) == len(k[j])
len(v[j]) >= 1
temperature > 0
Every query has at least one permitted key
All numeric entries are finite real numbers

Function Signature

def attention(q, k, v, causal, mask, temperature):

Your question is Modified Attention Mechanism Implementation. 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.