Update

[ghstack-poisoned]

Author: helunwencser

examples/models/llama/llama_transformer.py

@@ -154,7 +154,7 @@ def __init__(
     ):
         super().__init__()
         self.max_seq_length = max_seq_length
-        self.is_tranposed = transpose_cache
+        self.is_transposed = transpose_cache
         if transpose_cache:
             cache_shape = (max_batch_size, n_heads, max_seq_length, head_dim)
         else:

examples/models/llama/source_transformation/quantized_kv_cache.py

@@ -193,7 +193,7 @@ def update(self, input_pos, k_val, v_val):
     @classmethod
     def from_float(cls, kv_cache, cache_type: QuantizedCacheType):
         cache_shape = kv_cache.k_cache.shape
-        if kv_cache.is_tranposed:
+        if kv_cache.is_transposed:
             max_batch_size, n_heads, max_seq_length, head_dim = cache_shape
         else:
             max_batch_size, max_seq_length, n_heads, head_dim = cache_shape
@@ -203,7 +203,7 @@ def from_float(cls, kv_cache, cache_type: QuantizedCacheType):
             n_heads,
             head_dim,
             cache_type,
-            kv_cache.is_tranposed,
+            kv_cache.is_transposed,
             kv_cache.enable_dynamic_shape,
         )
 

examples/models/llama/source_transformation/test_quantized_kv_cache.py

@@ -48,8 +48,8 @@ def setUp(self):
         self.transpose_kv_cache = False
         self.dtype = torch.float32
 
-    def _test_simple_update_fetch(self, is_tranposed=False, is_dynamic_shape=False):
-        self.transpose_kv_cache = is_tranposed
+    def _test_simple_update_fetch(self, is_transposed=False, is_dynamic_shape=False):
+        self.transpose_kv_cache = is_transposed
         self.enable_dynamic_shape = is_dynamic_shape
         input_pos = torch.tensor([0, 1, 2])
         self.seq_len = input_pos.size(0)
@@ -122,7 +122,7 @@ def test_simple_update_fetch_not_transposed_dynamic_shape(self):
         self._test_simple_update_fetch(is_dynamic_shape=True)
 
     def test_simple_update_fetch_transposed(self):
-        self._test_simple_update_fetch(is_tranposed=True)
+        self._test_simple_update_fetch(is_transposed=True)
 
     def test_simple_update_fetch_transposed_dynamic_shape(self):
-        self._test_simple_update_fetch(is_tranposed=True, is_dynamic_shape=True)
+        self._test_simple_update_fetch(is_transposed=True, is_dynamic_shape=True)