Update Gemma.py
Browse files
Gemma.py
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
# Copyright 2024
|
2 |
#
|
3 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
4 |
# you may not use this file except in compliance with the License.
|
@@ -66,12 +66,17 @@ def apply_rotary_emb(x, freqs_cis):
|
|
66 |
return x_out
|
67 |
|
68 |
|
69 |
-
class Embedder:
|
70 |
"""Embedder module."""
|
71 |
def __init__(self, config: GemmaConfig):
|
72 |
self.vocab_size = config.vocab_size
|
73 |
self.embed_dim = config.hidden_size
|
74 |
-
self.input_embedding_table =
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
def encode(self, x):
|
77 |
x = tf.gather(self.input_embedding_table, x)
|
@@ -92,7 +97,12 @@ class RMSNorm:
|
|
92 |
):
|
93 |
self.eps = eps
|
94 |
self.add_unit_offset = add_unit_offset
|
95 |
-
self.weight =
|
|
|
|
|
|
|
|
|
|
|
96 |
|
97 |
def _norm(self, x):
|
98 |
return x * tf.math.rsqrt(tf.reduce_mean(tf.math.pow(x, 2), axis=-1, keepdims=True) + self.eps)
|
|
|
1 |
+
# Copyright 2024 NoteDance
|
2 |
#
|
3 |
# Licensed under the Apache License, Version 2.0 (the "License");
|
4 |
# you may not use this file except in compliance with the License.
|
|
|
66 |
return x_out
|
67 |
|
68 |
|
69 |
+
class Embedder(tf.keras.layers.Layer):
|
70 |
"""Embedder module."""
|
71 |
def __init__(self, config: GemmaConfig):
|
72 |
self.vocab_size = config.vocab_size
|
73 |
self.embed_dim = config.hidden_size
|
74 |
+
self.input_embedding_table = self.add_weight(
|
75 |
+
name='input_embedding_table',
|
76 |
+
shape=(self.vocab_size, self.embed_dim),
|
77 |
+
initializer=tf.keras.initializers.RandomNormal(stddev=0.02),
|
78 |
+
trainable=True
|
79 |
+
)
|
80 |
|
81 |
def encode(self, x):
|
82 |
x = tf.gather(self.input_embedding_table, x)
|
|
|
97 |
):
|
98 |
self.eps = eps
|
99 |
self.add_unit_offset = add_unit_offset
|
100 |
+
self.weight = self.add_weight(
|
101 |
+
name='weight',
|
102 |
+
shape=(self.dim,),
|
103 |
+
initializer=tf.keras.initializers.Zeros(),
|
104 |
+
trainable=True
|
105 |
+
)
|
106 |
|
107 |
def _norm(self, x):
|
108 |
return x * tf.math.rsqrt(tf.reduce_mean(tf.math.pow(x, 2), axis=-1, keepdims=True) + self.eps)
|