Spaces:
Running
Running
File size: 14,578 Bytes
5f923cd | 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 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 | // Copyright 2025 The ODML Authors.
//
// 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
//
// http://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.
#include "runtime/components/sampling_cpu_util.h"
#include <memory>
#include <random>
#include <vector>
#include <gmock/gmock.h>
#include <gtest/gtest.h>
#include "absl/types/span.h" // from @com_google_absl
namespace litert::lm {
namespace {
using ::testing::ElementsAre;
using ::testing::UnorderedElementsAre;
TEST(SamplingCpuUtilTest, TopKTokenIds_BatchSize1) {
const std::vector<float> logits = {0.1, 0.5, 0.4, 0.2};
auto topk_token_ids = TopKTokenIds(absl::MakeConstSpan(logits), /*k=*/2,
/*batch_size=*/1, /*sequence_size=*/1);
ASSERT_TRUE(topk_token_ids.ok());
EXPECT_EQ(topk_token_ids->size(), 1);
EXPECT_THAT((*topk_token_ids)[0], UnorderedElementsAre(1, 2));
}
TEST(SamplingCpuUtilTest, TopKTokenIds_BatchSize2) {
const std::vector<float> logits = {0.1, 0.5, 0.4, 0.2};
auto topk_token_ids = TopKTokenIds(absl::MakeConstSpan(logits), /*k=*/1,
/*batch_size=*/2, /*sequence_size=*/1);
ASSERT_TRUE(topk_token_ids.ok());
EXPECT_EQ(topk_token_ids->size(), 2);
EXPECT_THAT((*topk_token_ids)[0], ElementsAre(1));
EXPECT_THAT((*topk_token_ids)[1], ElementsAre(0));
}
TEST(SamplingCpuUtilTest, TopKTokenIds_SequenceLength2) {
const std::vector<float> logits = {0.1, 0.5, 0.4, 0.2, 0.6, 0.3};
auto topk_token_ids = TopKTokenIds(absl::MakeConstSpan(logits), /*k=*/1,
/*batch_size=*/1,
/*sequence_size=*/2);
ASSERT_TRUE(topk_token_ids.ok());
EXPECT_EQ(topk_token_ids->size(), 1);
EXPECT_THAT((*topk_token_ids)[0], ElementsAre(1, 1));
}
TEST(SamplingCpuUtilTest, Softmax_BatchSize1) {
const std::vector<float> logits = {0.1f, 0.1f};
const std::vector<int> topk_indices = {0, 1};
std::vector<std::vector<float>> max_logit_values;
auto probabilities =
Softmax(absl::MakeConstSpan(logits), absl::MakeConstSpan(topk_indices),
/*temperature=*/1.0, /*batch_size=*/1, /*sequence_size=*/1,
max_logit_values);
ASSERT_TRUE(probabilities.ok());
EXPECT_EQ(probabilities->size(), 1);
EXPECT_THAT((*probabilities)[0], ElementsAre(0.5, 0.5));
EXPECT_EQ(max_logit_values.size(), 1);
EXPECT_THAT(max_logit_values[0], ElementsAre(0.1f));
}
TEST(SamplingCpuUtilTest, Softmax_AllZeroLogits) {
const std::vector<float> logits = {0.0f, 0.0f};
const std::vector<int> topk_indices = {0, 1};
std::vector<std::vector<float>> max_logit_values;
auto probabilities =
Softmax(absl::MakeConstSpan(logits), absl::MakeConstSpan(topk_indices),
/*temperature=*/1.0, /*batch_size=*/1, /*sequence_size=*/1,
max_logit_values);
ASSERT_TRUE(probabilities.ok());
EXPECT_EQ(probabilities->size(), 1);
EXPECT_THAT((*probabilities)[0], ElementsAre(0.5, 0.5));
EXPECT_EQ(max_logit_values.size(), 1);
EXPECT_THAT(max_logit_values[0], ElementsAre(0.0f));
}
TEST(SamplingCpuUtilTest, Softmax_TemperatureVerySmall) {
const std::vector<float> logits = {0.0f, 1.0f, 2.0f};
const std::vector<int> topk_indices = {0, 1, 2};
std::vector<std::vector<float>> max_logit_values;
auto probabilities =
Softmax(absl::MakeConstSpan(logits), absl::MakeConstSpan(topk_indices),
/*temperature=*/0.00000001f, /*batch_size=*/1,
/*sequence_size=*/1, max_logit_values);
ASSERT_TRUE(probabilities.ok());
// Very small temperature should mimic greedy sampling.
EXPECT_EQ(probabilities->size(), 1);
EXPECT_THAT((*probabilities)[0], ElementsAre(0.0f, 0.0f, 1.0f));
EXPECT_EQ(max_logit_values.size(), 1);
EXPECT_THAT(max_logit_values[0], ElementsAre(2.0f));
}
TEST(SamplingCpuUtilTest, Softmax_TemperatureExactlyZero) {
const std::vector<float> logits = {0.0f, 1.0f, 2.0f};
const std::vector<int> topk_indices = {0, 1, 2};
std::vector<std::vector<float>> max_logit_values;
auto probabilities =
Softmax(absl::MakeConstSpan(logits), absl::MakeConstSpan(topk_indices),
/*temperature=*/0.0f, /*batch_size=*/1, /*sequence_size=*/1,
max_logit_values);
ASSERT_TRUE(probabilities.ok());
// Exactly zero temperature should mimic greedy sampling.
EXPECT_EQ(probabilities->size(), 1);
EXPECT_THAT((*probabilities)[0], ElementsAre(0.0f, 0.0f, 1.0f));
EXPECT_EQ(max_logit_values.size(), 1);
EXPECT_THAT(max_logit_values[0], ElementsAre(2.0f));
}
TEST(SamplingCpuUtilTest, Softmax_TemperatureInf) {
const std::vector<float> logits = {0.0f, 1.0f, 2.0f, 3.0f};
const std::vector<int> topk_indices = {0, 1, 2, 3};
std::vector<std::vector<float>> max_logit_values;
auto probabilities =
Softmax(absl::MakeConstSpan(logits), absl::MakeConstSpan(topk_indices),
/*temperature=*/100000000000.0f, /*batch_size=*/1,
/*sequence_size=*/1, max_logit_values);
ASSERT_TRUE(probabilities.ok());
// Very large temperature should mimic uniform sampling.
EXPECT_EQ(probabilities->size(), 1);
EXPECT_THAT((*probabilities)[0], ElementsAre(0.25f, 0.25f, 0.25f, 0.25f));
EXPECT_EQ(max_logit_values.size(), 1);
EXPECT_THAT(max_logit_values[0], ElementsAre(3.0f));
}
TEST(SamplingCpuUtilTest, Softmax_BatchSize3) {
// Batch size of 3, vocab size of 2.
const std::vector<float> logits = {0.1f, 0.1f, 0.0f, 5.0f, 1.0f, 0.0f};
absl::Span<const float> logits_span = absl::MakeConstSpan(logits);
const std::vector<int> topk_indices = {0, 1, 0, 1, 0, 1};
absl::Span<const int> topk_indices_span = absl::MakeConstSpan(topk_indices);
std::vector<std::vector<float>> max_logit_values;
auto probabilities = Softmax(logits_span, topk_indices_span,
/*temperature=*/1.0f, /*batch_size=*/3,
/*sequence_size=*/1, max_logit_values);
ASSERT_TRUE(probabilities.ok());
EXPECT_EQ(probabilities->size(), 3);
EXPECT_EQ((*probabilities)[0].size(), 2);
EXPECT_THAT((*probabilities)[0], ElementsAre(0.5f, 0.5f));
EXPECT_EQ((*probabilities)[1].size(), 2);
EXPECT_THAT((*probabilities)[1], ElementsAre(0.00669285096f, 0.993307173f));
EXPECT_EQ((*probabilities)[2].size(), 2);
EXPECT_THAT((*probabilities)[2], ElementsAre(0.731058598f, 0.268941432f));
EXPECT_EQ(max_logit_values.size(), 3);
EXPECT_EQ(max_logit_values[0].size(), 1);
EXPECT_THAT(max_logit_values[0], ElementsAre(0.1f));
EXPECT_EQ(max_logit_values[1].size(), 1);
EXPECT_THAT(max_logit_values[1], ElementsAre(5.0f));
EXPECT_EQ(max_logit_values[2].size(), 1);
EXPECT_THAT(max_logit_values[2], ElementsAre(1.0f));
}
TEST(SamplingCpuUtilTest, TopKTopPSampling_InvalidInputs) {
const std::vector<float> probabilities = {0.0, 0.0, 0.3};
auto rng = std::make_shared<std::default_random_engine>(0);
// Negative k.
std::vector<std::vector<float>> sampled_scores;
auto sampled_ids =
TopKTopPSampling(absl::MakeConstSpan(probabilities), /*k=*/-1,
/*p=*/0.5,
/*temperature=*/1.0, rng, /*batch_size=*/1,
/*sequence_size=*/1, sampled_scores);
EXPECT_FALSE(sampled_ids.ok());
// Negative p.
sampled_ids = TopKTopPSampling(absl::MakeConstSpan(probabilities),
/*k=*/1,
/*p=*/-0.5, /*temperature=*/1.0f, rng,
/*sequence_size=*/1,
/*batch_size=*/1, sampled_scores);
EXPECT_FALSE(sampled_ids.ok());
}
TEST(SamplingCpuUtilTest, TopKTopPSampling_BatchSize1) {
const std::vector<float> probabilities = {0.0, 0.0, 0.3};
auto rng = std::make_shared<std::default_random_engine>(0);
std::vector<std::vector<float>> sampled_scores;
auto sampled_ids =
TopKTopPSampling(absl::MakeConstSpan(probabilities), /*k=*/1,
/*p=*/0.5,
/*temperature=*/1.0f, rng, /*batch_size=*/1,
/*sequence_size=*/1, sampled_scores);
ASSERT_TRUE(sampled_ids.ok());
EXPECT_EQ(sampled_ids->size(), 1);
EXPECT_THAT((*sampled_ids)[0], ElementsAre(2));
EXPECT_EQ(sampled_scores.size(), 1);
EXPECT_THAT(sampled_scores[0], ElementsAre(1.0));
}
TEST(SamplingCpuUtilTest, TopKTopPSampling_BatchSize1_TopK) {
// Test that the sampler does return a sampled token from the top k
// instead of always returning the first or the last token.
const std::vector<float> logits = {-1.0e7f, 1.0f, -1e3f};
auto rng = std::make_shared<std::default_random_engine>(0);
std::vector<std::vector<float>> sampled_scores;
auto sampled_ids =
TopKTopPSampling(absl::MakeConstSpan(logits), /*k=*/3,
/*p=*/1.0,
/*temperature=*/1.0f, rng,
/*batch_size=*/1, /*sequence_size=*/1, sampled_scores);
ASSERT_TRUE(sampled_ids.ok());
EXPECT_EQ(sampled_ids->size(), 1);
EXPECT_THAT((*sampled_ids)[0], ElementsAre(1));
EXPECT_EQ(sampled_scores.size(), 1);
EXPECT_THAT(sampled_scores[0], ElementsAre(1.0));
}
TEST(SamplingCpuUtilTest, TopKTopPSampling_BatchSize3) {
// Batch of 3, vocab size of 3. The sampled ids are 2, 1, 0.
const std::vector<float> logits = {0.0, 0.0, 1.0, 0.0, 1.0,
0.0, 1.0, 0.0, 0.0};
auto rng = std::make_shared<std::default_random_engine>(0);
std::vector<std::vector<float>> sampled_scores;
auto sampled_ids =
TopKTopPSampling(absl::MakeConstSpan(logits), /*k=*/2, /*p=*/0.5,
/*temperature=*/0.00001f, rng, /*batch_size=*/3,
/*sequence_size=*/1, sampled_scores);
ASSERT_TRUE(sampled_ids.ok());
EXPECT_EQ(sampled_ids->size(), 3);
EXPECT_EQ((*sampled_ids)[0][0], 2);
EXPECT_EQ((*sampled_ids)[1][0], 1);
EXPECT_EQ((*sampled_ids)[2][0], 0);
EXPECT_EQ(sampled_scores.size(), 3);
EXPECT_EQ(sampled_scores[0].size(), 1);
EXPECT_EQ(sampled_scores[1].size(), 1);
EXPECT_EQ(sampled_scores[2].size(), 1);
}
TEST(SamplingCpuUtilTest, TopKTopPSampling_LargeVocabIndices) {
// Tests that sampling works correctly when top-k topk_token_ids are larger
// than k. This exposes a bug where vocab topk_token_ids were incorrectly used
// as offsets.
std::vector<float> logits = {1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 10.0};
auto rng = std::make_shared<std::default_random_engine>(0);
std::vector<std::vector<float>> sampled_scores;
// Top p = 0.0001f, should always return the token with the highest logit,
// which is the 14th token in this case.
auto sampled_ids =
TopKTopPSampling(absl::MakeConstSpan(logits), /*k=*/15, /*p=*/0.0001f,
/*temperature=*/1.0f, rng, /*batch_size=*/1,
/*sequence_size=*/1, sampled_scores);
ASSERT_TRUE(sampled_ids.ok());
// With very small temperature, it should pick the logit with the highest
// value, which is at index 11.
EXPECT_EQ(sampled_ids->size(), 1);
EXPECT_THAT((*sampled_ids)[0], ElementsAre(14));
EXPECT_EQ(sampled_scores.size(), 1);
EXPECT_THAT(sampled_scores[0], ElementsAre(0.99827528f));
}
TEST(SamplingCpuUtilTest, Softmax_BatchSize2SequenceLength2) {
// Batch size 2, sequence length 2, vocab size 3.
const std::vector<float> logits = {
0.1f, 0.1f, 0.2f, // Batch 0, Sequence 0
0.0f, 5.0f, 1.0f, // Batch 0, Sequence 1
1.0f, 0.0f, 2.0f, // Batch 1, Sequence 0
0.0f, 0.0f, 0.0f // Batch 1, Sequence 1
};
absl::Span<const float> logits_span = absl::MakeConstSpan(logits);
const std::vector<int> topk_indices = {
1, 2, // Batch 0, Sequence 0
1, 2, // Batch 0, Sequence 1
0, 2, // Batch 1, Sequence 0
0, 1 // Batch 1, Sequence 1
};
absl::Span<const int> topk_indices_span = absl::MakeConstSpan(topk_indices);
std::vector<std::vector<float>> max_logit_values;
auto probabilities = Softmax(logits_span, topk_indices_span,
/*temperature=*/1.0f, /*batch_size=*/2,
/*sequence_size=*/2, max_logit_values);
ASSERT_TRUE(probabilities.ok());
EXPECT_EQ(probabilities->size(), 2);
EXPECT_EQ((*probabilities)[0].size(), 4);
EXPECT_THAT((*probabilities)[0],
ElementsAre(testing::FloatNear(0.47502f, 1e-4f),
testing::FloatNear(0.52498f, 1e-4f),
testing::FloatNear(0.98201f, 1e-4f),
testing::FloatNear(0.017986f, 1e-4f)));
EXPECT_EQ((*probabilities)[1].size(), 4);
EXPECT_THAT((*probabilities)[1],
ElementsAre(testing::FloatNear(0.26894f, 1e-4f),
testing::FloatNear(0.73106f, 1e-4f),
testing::FloatNear(0.5f, 1e-4f),
testing::FloatNear(0.5f, 1e-4f)));
EXPECT_EQ(max_logit_values.size(), 2);
EXPECT_EQ(max_logit_values[0].size(), 2);
EXPECT_THAT(max_logit_values[0], ElementsAre(0.2f, 5.0f));
EXPECT_EQ(max_logit_values[1].size(), 2);
EXPECT_THAT(max_logit_values[1], ElementsAre(2.0f, 0.0f));
}
TEST(SamplingCpuUtilTest, TopKTopPSampling_BatchSize2SequenceLength2) {
// Batch of 2, sequence length 2, vocab size of 3.
const std::vector<float> logits = {
0.0f, 0.0f, 1.0f, // b0, s0: top is 2
1.0f, 0.0f, 0.0f, // b0, s1: top is 0
0.0f, 1.0f, 0.0f, // b1, s0: top is 1
0.0f, 0.0f, 1.0f // b1, s1: top is 2
};
auto rng = std::make_shared<std::default_random_engine>(0);
std::vector<std::vector<float>> sampled_scores;
auto sampled_ids =
TopKTopPSampling(absl::MakeConstSpan(logits), /*k=*/2, /*p=*/0.5f,
/*temperature=*/0.00001f, rng, /*batch_size=*/2,
/*sequence_size=*/2, sampled_scores);
ASSERT_TRUE(sampled_ids.ok());
EXPECT_EQ(sampled_ids->size(), 2);
EXPECT_THAT((*sampled_ids)[0], ElementsAre(2, 0));
EXPECT_THAT((*sampled_ids)[1], ElementsAre(1, 2));
EXPECT_EQ(sampled_scores.size(), 2);
EXPECT_EQ(sampled_scores[0].size(), 2);
EXPECT_EQ(sampled_scores[1].size(), 2);
}
} // namespace
} // namespace litert::lm
|