-
Notifications
You must be signed in to change notification settings - Fork 281
Expand file tree
/
Copy pathmodel_test.py
More file actions
86 lines (70 loc) · 2.51 KB
/
model_test.py
File metadata and controls
86 lines (70 loc) · 2.51 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
# Copyright 2026 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Tests for Gemma 4 model."""
from __future__ import annotations
from absl.testing import absltest
from flax import nnx
import jax
import jax.numpy as jnp
from tunix.models.gemma4 import model as model_lib
class ModelTest(absltest.TestCase):
def test_forward_pass_dense(self):
config = model_lib.ModelConfig.gemma4_e2b()
config.num_layers = 1
config.embed_dim = 256
config.hidden_dim = 512
config.num_heads = 4
config.head_dim = 64
config.num_kv_heads = 1
config.frac_shared_layers = 0.0
rngs = nnx.Rngs(0)
model = model_lib.Gemma4(config, rngs=rngs)
tokens = jax.random.randint(
jax.random.PRNGKey(0), (2, 32), 0, config.num_embed
)
positions = jnp.tile(
jnp.arange(tokens.shape[1])[None, :], (tokens.shape[0], 1)
)
attn_mask = jnp.tril(
jnp.ones((tokens.shape[1], tokens.shape[1]), dtype=jnp.bool_)
)[None, ...]
logits, _ = model(tokens, positions=positions, attention_mask=attn_mask)
self.assertEqual(logits.shape, (2, 32, config.num_embed))
print(f"{logits.shape=}")
def test_forward_pass_moe(self):
config = model_lib.ModelConfig.gemma4_26b_a4b()
config.num_layers = 1
config.embed_dim = 256
config.hidden_dim = 512
config.num_heads = 4
config.head_dim = 64
config.num_kv_heads = 1
config.num_experts = 4
config.num_experts_per_tok = 2
config.expert_dim = 128
rngs = nnx.Rngs(0)
model = model_lib.Gemma4(config, rngs=rngs)
tokens = jax.random.randint(
jax.random.PRNGKey(0), (2, 32), 0, config.num_embed
)
positions = jnp.tile(
jnp.arange(tokens.shape[1])[None, :], (tokens.shape[0], 1)
)
attn_mask = jnp.tril(
jnp.ones((tokens.shape[1], tokens.shape[1]), dtype=jnp.bool_)
)[None, ...]
logits, _ = model(tokens, positions=positions, attention_mask=attn_mask)
self.assertEqual(logits.shape, (2, 32, config.num_embed))
if __name__ == "__main__":
absltest.main()