diff --git a/src/starkware/cairo/common/poseidon_utils.py b/src/starkware/cairo/common/poseidon_utils.py
index 03ad8aac..f9863095 100644
--- a/src/starkware/cairo/common/poseidon_utils.py
+++ b/src/starkware/cairo/common/poseidon_utils.py
@@ -4,12 +4,10 @@
 """
 
 import hashlib
-from typing import Iterable, List, Optional, Type
-
-import numpy as np
+from typing import List, Optional, Type
 
 from starkware.cairo.lang.cairo_constants import DEFAULT_PRIME
-from starkware.python.math_utils import pow_mod, safe_div
+from starkware.python.math_utils import safe_div
 
 
 def generate_round_constant(fn_name: str, field_prime: int, idx: int) -> int:
@@ -40,7 +38,7 @@ class PoseidonParams:
     poseidon_small_params: Optional["PoseidonParams"] = None
 
     def __init__(
-        self, field_prime: int, r: int, c: int, r_f: int, r_p: int, mds: Iterable[Iterable[int]]
+        self, field_prime: int, r: int, c: int, r_f: int, r_p: int
     ):
         self.field_prime = field_prime
         self.r = r
@@ -53,22 +51,16 @@ def __init__(
         self.output_size = c
         assert self.output_size <= r
         # A list of r_f + r_p vectors for the Add-Round Key phase.
-        self.ark = np.array(
-            [
+        self.ark =  [
                 [generate_round_constant("Hades", field_prime, m * i + j) for j in range(m)]
                 for i in range(n_rounds)
-            ],
-            dtype=object,
-        )
-
-        # The MDS matrix for the MixLayer phase.
-        self.mds = np.array(mds, dtype=object)
+            ]
 
     @classmethod
     def get_default_poseidon_params(cls: Type["PoseidonParams"]):
         if cls.poseidon_small_params is None:
             cls.poseidon_small_params = cls(
-                field_prime=DEFAULT_PRIME, r=2, c=1, r_f=8, r_p=83, mds=SmallMds
+                field_prime=DEFAULT_PRIME, r=2, c=1, r_f=8, r_p=83
             )
 
         return cls.poseidon_small_params
@@ -76,22 +68,24 @@ def get_default_poseidon_params(cls: Type["PoseidonParams"]):
 
 def hades_round(values, params: PoseidonParams, is_full_round: bool, round_idx: int):
     # Add-Round Key.
-    values = (values + params.ark[round_idx]) % params.field_prime
-
+    values = [
+        (val + ark) % params.field_prime
+        for val, ark in zip(values, params.ark[round_idx])
+    ]
     # SubWords.
     if is_full_round:
-        values = pow_mod(values, 3, params.field_prime)
+        values =  [pow(val, 3, params.field_prime) for val in values]
     else:
-        values[-1:] = pow_mod(values[-1:], 3, params.field_prime)
+        values[-1] = pow(values[-1], 3, params.field_prime)
 
     # MixLayer.
-    values = params.mds.dot(values) % params.field_prime
+    values = mds_mul(values, params.field_prime)
     return values
 
 
 def hades_permutation(values: List[int], params: PoseidonParams) -> List[int]:
     assert len(values) == params.m
-    values = np.array(values, dtype=object)
+
     round_idx = 0
     # Apply r_f/2 full rounds.
     for _ in range(safe_div(params.r_f, 2)):
@@ -106,8 +100,21 @@ def hades_permutation(values: List[int], params: PoseidonParams) -> List[int]:
         values = hades_round(values, params, True, round_idx)
         round_idx += 1
     assert round_idx == params.n_rounds
-    return list(values)
+    return values
+
 
+def mds_mul(vector, field):
+    """
+    Multiplies a vector by the SmallMds matrix.
+	[3, 1, 1]    [r0]    [3* r0 + r1 + r2 ]
+	[1, -1, 1] * [r1]  = [r0 - r1 + r2    ]
+	[1, 1, -2]   [r2]    [r0 + r1 - 2 * r2]
+    """
+    return [
+        (3 * vector[0] + vector[1] + vector[2]) % field,
+        (vector[0] - vector[1] + vector[2]) % field,
+        (vector[0] + vector[1] - 2 * vector[2]) % field,
+    ]
 
 # The actual config to be in use, with extremely small MDS coefficients.
-SmallMds = [[3, 1, 1], [1, -1, 1], [1, 1, -2]]
+# SmallMds = [[3, 1, 1], [1, -1, 1], [1, 1, -2]]