tpu-inference 0.11.1__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Potentially problematic release.
This version of tpu-inference might be problematic. Click here for more details.
- tests/__init__.py +0 -0
- tests/core/__init__.py +0 -0
- tests/core/test_adapters.py +83 -0
- tests/core/test_core_tpu.py +523 -0
- tests/core/test_disagg_executor.py +60 -0
- tests/core/test_disagg_utils.py +53 -0
- tests/core/test_init.py +49 -0
- tests/kernels/__init__.py +0 -0
- tests/kernels/quantized_matmul_kernel_test.py +191 -0
- tests/kernels/ragged_kv_cache_update_v2_test.py +234 -0
- tests/kernels/ragged_paged_attention_kernel_v2_test.py +400 -0
- tests/kernels/ragged_paged_attention_kernel_v3_test.py +504 -0
- tests/lora/__init__.py +0 -0
- tests/lora/test_lora.py +123 -0
- tests/test_base.py +201 -0
- tests/test_quantization.py +836 -0
- tests/test_tpu_info.py +120 -0
- tests/test_utils.py +218 -0
- tests/tpu_backend_test.py +59 -0
- tpu_inference/__init__.py +30 -0
- tpu_inference/adapters/__init__.py +0 -0
- tpu_inference/adapters/vllm_adapters.py +42 -0
- tpu_inference/adapters/vllm_config_adapters.py +134 -0
- tpu_inference/backend.py +69 -0
- tpu_inference/core/__init__.py +0 -0
- tpu_inference/core/adapters.py +153 -0
- tpu_inference/core/core_tpu.py +776 -0
- tpu_inference/core/disagg_executor.py +117 -0
- tpu_inference/core/disagg_utils.py +51 -0
- tpu_inference/di/__init__.py +0 -0
- tpu_inference/di/abstracts.py +28 -0
- tpu_inference/di/host.py +76 -0
- tpu_inference/di/interfaces.py +51 -0
- tpu_inference/distributed/__init__.py +0 -0
- tpu_inference/distributed/tpu_connector.py +699 -0
- tpu_inference/distributed/utils.py +59 -0
- tpu_inference/executors/__init__.py +0 -0
- tpu_inference/executors/ray_distributed_executor.py +346 -0
- tpu_inference/experimental/__init__.py +0 -0
- tpu_inference/experimental/llama3_jax_stashed.py +258 -0
- tpu_inference/interfaces/__init__.py +0 -0
- tpu_inference/interfaces/cache.py +31 -0
- tpu_inference/interfaces/config.py +47 -0
- tpu_inference/interfaces/config_parts.py +117 -0
- tpu_inference/interfaces/engine.py +51 -0
- tpu_inference/interfaces/outputs.py +22 -0
- tpu_inference/interfaces/params.py +21 -0
- tpu_inference/interfaces/platform.py +74 -0
- tpu_inference/interfaces/request.py +39 -0
- tpu_inference/interfaces/scheduler.py +31 -0
- tpu_inference/kernels/__init__.py +0 -0
- tpu_inference/kernels/collectives/__init__.py +0 -0
- tpu_inference/kernels/collectives/all_gather_matmul.py +735 -0
- tpu_inference/kernels/collectives/all_gather_matmul_tuned_block_sizes.py +60 -0
- tpu_inference/kernels/collectives/util.py +47 -0
- tpu_inference/kernels/flash_attention/__init__.py +0 -0
- tpu_inference/kernels/flash_attention/kernel.py +772 -0
- tpu_inference/kernels/quantized_matmul/__init__.py +0 -0
- tpu_inference/kernels/quantized_matmul/kernel.py +395 -0
- tpu_inference/kernels/quantized_matmul/tuned_block_sizes.py +609 -0
- tpu_inference/kernels/quantized_matmul/util.py +58 -0
- tpu_inference/kernels/ragged_paged_attention/__init__.py +0 -0
- tpu_inference/kernels/ragged_paged_attention/v2/__init__.py +0 -0
- tpu_inference/kernels/ragged_paged_attention/v2/kernel.py +875 -0
- tpu_inference/kernels/ragged_paged_attention/v2/ragged_kv_cache_update.py +287 -0
- tpu_inference/kernels/ragged_paged_attention/v2/tuned_block_sizes.py +1482 -0
- tpu_inference/kernels/ragged_paged_attention/v3/__init__.py +0 -0
- tpu_inference/kernels/ragged_paged_attention/v3/kernel.py +1447 -0
- tpu_inference/kernels/ragged_paged_attention/v3/tuned_block_sizes.py +3834 -0
- tpu_inference/kernels/ragged_paged_attention/v3/util.py +47 -0
- tpu_inference/layers/__init__.py +0 -0
- tpu_inference/layers/common/__init__.py +0 -0
- tpu_inference/layers/common/attention_metadata.py +34 -0
- tpu_inference/layers/jax/__init__.py +0 -0
- tpu_inference/layers/jax/attention/__init__.py +0 -0
- tpu_inference/layers/jax/attention/attention.py +254 -0
- tpu_inference/layers/jax/attention/deepseek_v3_attention.py +354 -0
- tpu_inference/layers/jax/attention/llama4_attention.py +153 -0
- tpu_inference/layers/jax/attention_interface.py +356 -0
- tpu_inference/layers/jax/base.py +151 -0
- tpu_inference/layers/jax/binary_search.py +295 -0
- tpu_inference/layers/jax/constants.py +88 -0
- tpu_inference/layers/jax/layers.py +301 -0
- tpu_inference/layers/jax/misc.py +16 -0
- tpu_inference/layers/jax/moe/__init__.py +0 -0
- tpu_inference/layers/jax/moe/deepseek_v3_moe.py +608 -0
- tpu_inference/layers/jax/moe/moe.py +209 -0
- tpu_inference/layers/jax/rope.py +172 -0
- tpu_inference/layers/jax/rope_interface.py +214 -0
- tpu_inference/layers/jax/sample/__init__.py +0 -0
- tpu_inference/layers/jax/sample/rejection_sampler.py +515 -0
- tpu_inference/layers/jax/sample/sampling.py +95 -0
- tpu_inference/layers/jax/sample/sampling_metadata.py +69 -0
- tpu_inference/layers/jax/sharding.py +406 -0
- tpu_inference/layers/jax/transformer_block.py +76 -0
- tpu_inference/layers/vllm/__init__.py +0 -0
- tpu_inference/layers/vllm/attention.py +184 -0
- tpu_inference/layers/vllm/fused_moe.py +399 -0
- tpu_inference/layers/vllm/linear_common.py +186 -0
- tpu_inference/layers/vllm/quantization/__init__.py +34 -0
- tpu_inference/layers/vllm/quantization/awq.py +207 -0
- tpu_inference/layers/vllm/quantization/common.py +105 -0
- tpu_inference/layers/vllm/quantization/compressed_tensors/__init__.py +0 -0
- tpu_inference/layers/vllm/quantization/compressed_tensors/compressed_tensors.py +121 -0
- tpu_inference/layers/vllm/quantization/compressed_tensors/schemes/__init__.py +0 -0
- tpu_inference/layers/vllm/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py +208 -0
- tpu_inference/layers/vllm/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py +136 -0
- tpu_inference/layers/vllm/quantization/unquantized.py +263 -0
- tpu_inference/layers/vllm/sharding.py +151 -0
- tpu_inference/logger.py +10 -0
- tpu_inference/lora/__init__.py +0 -0
- tpu_inference/lora/torch_lora_ops.py +103 -0
- tpu_inference/lora/torch_punica_tpu.py +308 -0
- tpu_inference/mock/__init__.py +0 -0
- tpu_inference/mock/vllm_config_utils.py +28 -0
- tpu_inference/mock/vllm_envs.py +1233 -0
- tpu_inference/mock/vllm_logger.py +212 -0
- tpu_inference/mock/vllm_logging_utils.py +15 -0
- tpu_inference/models/__init__.py +0 -0
- tpu_inference/models/common/__init__.py +0 -0
- tpu_inference/models/common/model_loader.py +433 -0
- tpu_inference/models/jax/__init__.py +0 -0
- tpu_inference/models/jax/deepseek_v3.py +868 -0
- tpu_inference/models/jax/llama3.py +366 -0
- tpu_inference/models/jax/llama4.py +473 -0
- tpu_inference/models/jax/llama_eagle3.py +333 -0
- tpu_inference/models/jax/phi3.py +376 -0
- tpu_inference/models/jax/qwen2.py +375 -0
- tpu_inference/models/jax/qwen2_5_vl.py +976 -0
- tpu_inference/models/jax/qwen3.py +302 -0
- tpu_inference/models/jax/utils/__init__.py +0 -0
- tpu_inference/models/jax/utils/file_utils.py +96 -0
- tpu_inference/models/jax/utils/multi_modal_utils.py +164 -0
- tpu_inference/models/jax/utils/quantization/__init__.py +0 -0
- tpu_inference/models/jax/utils/quantization/quantization_utils.py +588 -0
- tpu_inference/models/jax/utils/weight_utils.py +510 -0
- tpu_inference/models/vllm/__init__.py +0 -0
- tpu_inference/models/vllm/vllm_model_wrapper.py +272 -0
- tpu_inference/models/vllm/vllm_model_wrapper_context.py +45 -0
- tpu_inference/platforms/__init__.py +2 -0
- tpu_inference/platforms/tpu_jax.py +257 -0
- tpu_inference/runner/__init__.py +0 -0
- tpu_inference/runner/block_table_jax.py +122 -0
- tpu_inference/runner/compilation_manager.py +672 -0
- tpu_inference/runner/input_batch_jax.py +435 -0
- tpu_inference/runner/kv_cache.py +119 -0
- tpu_inference/runner/kv_cache_manager.py +460 -0
- tpu_inference/runner/lora_utils.py +92 -0
- tpu_inference/runner/multimodal_manager.py +208 -0
- tpu_inference/runner/persistent_batch_manager.py +244 -0
- tpu_inference/runner/speculative_decoding_manager.py +250 -0
- tpu_inference/runner/structured_decoding_manager.py +89 -0
- tpu_inference/runner/tpu_jax_runner.py +771 -0
- tpu_inference/runner/utils.py +426 -0
- tpu_inference/spec_decode/__init__.py +0 -0
- tpu_inference/spec_decode/jax/__init__.py +0 -0
- tpu_inference/spec_decode/jax/eagle3.py +334 -0
- tpu_inference/tpu_info.py +77 -0
- tpu_inference/utils.py +294 -0
- tpu_inference/worker/__init__.py +0 -0
- tpu_inference/worker/_temporary_vllm_compat.py +129 -0
- tpu_inference/worker/base.py +100 -0
- tpu_inference/worker/tpu_worker_jax.py +321 -0
- tpu_inference-0.11.1.dist-info/METADATA +101 -0
- tpu_inference-0.11.1.dist-info/RECORD +168 -0
- tpu_inference-0.11.1.dist-info/WHEEL +5 -0
- tpu_inference-0.11.1.dist-info/licenses/LICENSE +201 -0
- tpu_inference-0.11.1.dist-info/top_level.txt +2 -0
tests/test_base.py
ADDED
|
@@ -0,0 +1,201 @@
|
|
|
1
|
+
import logging
|
|
2
|
+
import unittest
|
|
3
|
+
import warnings
|
|
4
|
+
from dataclasses import dataclass, field, fields
|
|
5
|
+
from typing import Any, List, Mapping
|
|
6
|
+
|
|
7
|
+
from tpu_inference.layers.jax.base import Config
|
|
8
|
+
|
|
9
|
+
# Use the 'warnings' module to globally ignore warnings within this block
|
|
10
|
+
vllm_logger = logging.getLogger("vllm")
|
|
11
|
+
original_level = vllm_logger.level
|
|
12
|
+
|
|
13
|
+
with warnings.catch_warnings():
|
|
14
|
+
warnings.simplefilter("ignore")
|
|
15
|
+
|
|
16
|
+
# Set the vLLM logger to ERROR to suppress its messages
|
|
17
|
+
vllm_logger.setLevel(logging.ERROR)
|
|
18
|
+
|
|
19
|
+
# Import the class; all warnings will be suppressed
|
|
20
|
+
from vllm.config import ModelConfig
|
|
21
|
+
|
|
22
|
+
vllm_logger.setLevel(logging.WARNING)
|
|
23
|
+
|
|
24
|
+
|
|
25
|
+
def setup_vllm_config(subconfig_types: List[str],
|
|
26
|
+
overrides: List[Mapping[str, Any]]):
|
|
27
|
+
vllm_config = SimpleVllmConfig()
|
|
28
|
+
for (subconfig_type, override) in zip(subconfig_types, overrides):
|
|
29
|
+
if subconfig_type == "model":
|
|
30
|
+
for key in override:
|
|
31
|
+
setattr(vllm_config.model_config, key, override[key])
|
|
32
|
+
else:
|
|
33
|
+
for key in override:
|
|
34
|
+
setattr(vllm_config, key, override[key])
|
|
35
|
+
return vllm_config
|
|
36
|
+
|
|
37
|
+
|
|
38
|
+
@dataclass
|
|
39
|
+
class SimpleVllmConfig():
|
|
40
|
+
additional_config: Mapping[str, Any] = field(default_factory=dict)
|
|
41
|
+
# Set default max_model_len to turn off warnings.
|
|
42
|
+
model_config: ModelConfig = field(
|
|
43
|
+
default_factory=lambda: ModelConfig(max_model_len=1024))
|
|
44
|
+
|
|
45
|
+
|
|
46
|
+
@dataclass
|
|
47
|
+
class SimpleConfig(Config):
|
|
48
|
+
vllm_config: SimpleVllmConfig
|
|
49
|
+
arg1: str
|
|
50
|
+
arg2: str
|
|
51
|
+
arg3: int
|
|
52
|
+
|
|
53
|
+
def is_equal(self, other: Config):
|
|
54
|
+
for f in fields(self):
|
|
55
|
+
if f.name != "vllm_config":
|
|
56
|
+
if getattr(self, f.name) != getattr(other, f.name):
|
|
57
|
+
return False
|
|
58
|
+
return True
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
class ConfigOverrideTests(unittest.TestCase):
|
|
62
|
+
|
|
63
|
+
def test_additional_config_overrides(self):
|
|
64
|
+
subconfig_types = ['']
|
|
65
|
+
overrides = [{"additional_config": {"arg1": "val1", "arg2": "val2"}}]
|
|
66
|
+
override_vllm_config = setup_vllm_config(subconfig_types, overrides)
|
|
67
|
+
default_vllm_config = SimpleVllmConfig()
|
|
68
|
+
config = SimpleConfig(vllm_config=override_vllm_config,
|
|
69
|
+
arg1="foo",
|
|
70
|
+
arg2="bar",
|
|
71
|
+
arg3=123)
|
|
72
|
+
expected_config = SimpleConfig(vllm_config=default_vllm_config,
|
|
73
|
+
arg1="val1",
|
|
74
|
+
arg2="val2",
|
|
75
|
+
arg3=123)
|
|
76
|
+
self.assertTrue(config.is_equal(expected_config))
|
|
77
|
+
|
|
78
|
+
def test_hf_overrides(self):
|
|
79
|
+
subconfig_types = ['model']
|
|
80
|
+
overrides = [{"hf_overrides": {"arg2": "val2", "arg3": 456}}]
|
|
81
|
+
default_vllm_config = SimpleVllmConfig()
|
|
82
|
+
override_vllm_config = setup_vllm_config(subconfig_types, overrides)
|
|
83
|
+
config = SimpleConfig(vllm_config=override_vllm_config,
|
|
84
|
+
arg1="foo",
|
|
85
|
+
arg2="bar",
|
|
86
|
+
arg3=123)
|
|
87
|
+
expected_config = SimpleConfig(vllm_config=default_vllm_config,
|
|
88
|
+
arg1="foo",
|
|
89
|
+
arg2="val2",
|
|
90
|
+
arg3=456)
|
|
91
|
+
self.assertTrue(config.is_equal(expected_config))
|
|
92
|
+
|
|
93
|
+
def test_additional_and_hf_overrides(self):
|
|
94
|
+
subconfig_types = ['', 'model']
|
|
95
|
+
overrides = [{
|
|
96
|
+
"additional_config": {
|
|
97
|
+
"arg1": "val1",
|
|
98
|
+
"arg2": "val2"
|
|
99
|
+
}
|
|
100
|
+
}, {
|
|
101
|
+
"hf_overrides": {
|
|
102
|
+
"arg2": "val3",
|
|
103
|
+
"arg3": 456
|
|
104
|
+
}
|
|
105
|
+
}]
|
|
106
|
+
default_vllm_config = SimpleVllmConfig()
|
|
107
|
+
override_vllm_config = setup_vllm_config(subconfig_types, overrides)
|
|
108
|
+
config = SimpleConfig(vllm_config=override_vllm_config,
|
|
109
|
+
arg1="foo",
|
|
110
|
+
arg2="bar",
|
|
111
|
+
arg3=123)
|
|
112
|
+
expected_config = SimpleConfig(vllm_config=default_vllm_config,
|
|
113
|
+
arg1="val1",
|
|
114
|
+
arg2="val3",
|
|
115
|
+
arg3=456)
|
|
116
|
+
self.assertTrue(config.is_equal(expected_config))
|
|
117
|
+
|
|
118
|
+
def test_additional_and_generate_overrides(self):
|
|
119
|
+
subconfig_types = ['', 'model']
|
|
120
|
+
overrides = [{
|
|
121
|
+
"additional_config": {
|
|
122
|
+
"arg1": "val1",
|
|
123
|
+
"arg2": "val2"
|
|
124
|
+
}
|
|
125
|
+
}, {
|
|
126
|
+
"override_generation_config": {
|
|
127
|
+
"arg2": "val3",
|
|
128
|
+
"arg3": 456
|
|
129
|
+
}
|
|
130
|
+
}]
|
|
131
|
+
default_vllm_config = SimpleVllmConfig()
|
|
132
|
+
override_vllm_config = setup_vllm_config(subconfig_types, overrides)
|
|
133
|
+
config = SimpleConfig(vllm_config=override_vllm_config,
|
|
134
|
+
arg1="foo",
|
|
135
|
+
arg2="bar",
|
|
136
|
+
arg3=123)
|
|
137
|
+
expected_config = SimpleConfig(vllm_config=default_vllm_config,
|
|
138
|
+
arg1="val1",
|
|
139
|
+
arg2="val3",
|
|
140
|
+
arg3=456)
|
|
141
|
+
self.assertTrue(config.is_equal(expected_config))
|
|
142
|
+
|
|
143
|
+
def test_hf_and_generate_overrides(self):
|
|
144
|
+
subconfig_types = ['model', 'model']
|
|
145
|
+
overrides = [{
|
|
146
|
+
"hf_overrides": {
|
|
147
|
+
"arg2": "val2",
|
|
148
|
+
"arg3": 456
|
|
149
|
+
}
|
|
150
|
+
}, {
|
|
151
|
+
"override_generation_config": {
|
|
152
|
+
"arg2": "val4",
|
|
153
|
+
"arg3": 789
|
|
154
|
+
}
|
|
155
|
+
}]
|
|
156
|
+
default_vllm_config = SimpleVllmConfig()
|
|
157
|
+
override_vllm_config = setup_vllm_config(subconfig_types, overrides)
|
|
158
|
+
config = SimpleConfig(vllm_config=override_vllm_config,
|
|
159
|
+
arg1="foo",
|
|
160
|
+
arg2="bar",
|
|
161
|
+
arg3=123)
|
|
162
|
+
expected_config = SimpleConfig(vllm_config=default_vllm_config,
|
|
163
|
+
arg1="foo",
|
|
164
|
+
arg2="val4",
|
|
165
|
+
arg3=789)
|
|
166
|
+
self.assertTrue(config.is_equal(expected_config))
|
|
167
|
+
|
|
168
|
+
def test_additional_and_hf_and_generate_overrides(self):
|
|
169
|
+
subconfig_types = ['', 'model', 'model']
|
|
170
|
+
overrides = [{
|
|
171
|
+
"additional_config": {
|
|
172
|
+
"arg1": "val1",
|
|
173
|
+
"arg2": "val2"
|
|
174
|
+
}
|
|
175
|
+
}, {
|
|
176
|
+
"hf_overrides": {
|
|
177
|
+
"arg2": "val2",
|
|
178
|
+
"arg3": 456
|
|
179
|
+
}
|
|
180
|
+
}, {
|
|
181
|
+
"override_generation_config": {
|
|
182
|
+
"arg1": "val3",
|
|
183
|
+
"arg2": "val4",
|
|
184
|
+
"arg3": 789
|
|
185
|
+
}
|
|
186
|
+
}]
|
|
187
|
+
default_vllm_config = SimpleVllmConfig()
|
|
188
|
+
override_vllm_config = setup_vllm_config(subconfig_types, overrides)
|
|
189
|
+
config = SimpleConfig(vllm_config=override_vllm_config,
|
|
190
|
+
arg1="foo",
|
|
191
|
+
arg2="bar",
|
|
192
|
+
arg3=123)
|
|
193
|
+
expected_config = SimpleConfig(vllm_config=default_vllm_config,
|
|
194
|
+
arg1="val3",
|
|
195
|
+
arg2="val4",
|
|
196
|
+
arg3=789)
|
|
197
|
+
self.assertTrue(config.is_equal(expected_config))
|
|
198
|
+
|
|
199
|
+
|
|
200
|
+
if __name__ == '__main__':
|
|
201
|
+
unittest.main()
|