diff --git a/challenges/hard/53_casual_attention/challenge.html b/challenges/hard/53_casual_attention/challenge.html index 1558f9d6..4a7d9a90 100644 --- a/challenges/hard/53_casual_attention/challenge.html +++ b/challenges/hard/53_casual_attention/challenge.html @@ -249,7 +249,7 @@

Constraints

  • Matrix Q, K, and V are all of size M×d
  • 1 ≤ M ≤ 10000
  • 1 ≤ d ≤ 128
  • -
  • All elements in Q, K, and V are sampled from[-100.0, 100.0]
  • +
  • All elements in Q, K, and V are sampled from[-1.0, 1.0]
  • Data type for all matrices is float32
  • Performance is measured with M = 5,000
  • diff --git a/challenges/hard/53_casual_attention/challenge.py b/challenges/hard/53_casual_attention/challenge.py index af087d0f..2def7986 100644 --- a/challenges/hard/53_casual_attention/challenge.py +++ b/challenges/hard/53_casual_attention/challenge.py @@ -141,9 +141,9 @@ def generate_functional_test(self) -> List[Dict[str, Any]]: def generate_performance_test(self) -> Dict[str, Any]: M, d = 5000, 128 return { - "Q": RandTensor((M, d), -100.0, 100.0), - "K": RandTensor((M, d), -100.0, 100.0), - "V": RandTensor((M, d), -100.0, 100.0), + "Q": RandTensor((M, d), -1.0, 1.0), + "K": RandTensor((M, d), -1.0, 1.0), + "V": RandTensor((M, d), -1.0, 1.0), "output": OutTensor((M, d)), "M": M, "d": d,