vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Added XTC to test-sampling

Browse source
This commit is contained in:
MaggotHATE 2024-10-08 01:19:39 +05:00 • committed by GitHub
parent 4c44e3da5a
commit dbe9ef7783
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 34 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

40
tests/test-sampling.cpp
View file

@ -111,6 +111,28 @@ static void test_min_p(const std::vector<float> & probs, const std::vector<float
} }
} }
static void test_xtc(const std::vector<float> & probs, const std::vector<float> & expected_probs, float p, float t, float t_max) {
const size_t n_vocab = probs.size();
std::vector<llama_token_data> cur;
cur.reserve(n_vocab);
for (llama_token token_id = 0; token_id < (llama_token)n_vocab; token_id++) {
const float logit = logf(probs[token_id]);
cur.emplace_back(llama_token_data{token_id, logit, 0.0f});
}
llama_token_data_array cur_p = { cur.data(), cur.size(), -1, false };
APPLY(llama_sampler_init_softmax(), &cur_p);
DUMP(&cur_p);
APPLY(llama_sampler_init_xtc(p, t, t_max, 0, 0), &cur_p);
DUMP(&cur_p);
GGML_ASSERT(cur_p.size == expected_probs.size());
for (size_t i = 0; i < cur_p.size; i++) {
GGML_ASSERT(fabs(cur_p.data[i].p - expected_probs[i]) < 1e-5);
}
}
static void test_typical(const std::vector<float> & probs, const std::vector<float> & expected_probs, float p) { static void test_typical(const std::vector<float> & probs, const std::vector<float> & expected_probs, float p) {
const size_t n_vocab = probs.size(); const size_t n_vocab = probs.size();
@ -279,12 +301,13 @@ static void test_perf() {
data.emplace_back(llama_token_data{i, logit, 0.0f}); data.emplace_back(llama_token_data{i, logit, 0.0f});
} }
BENCH(llama_sampler_init_top_k (40), data, 32); BENCH(llama_sampler_init_top_k (40), data, 32);
BENCH(llama_sampler_init_top_p (0.8f, 1), data, 32); BENCH(llama_sampler_init_top_p (0.8f, 1), data, 32);
BENCH(llama_sampler_init_min_p (0.2f, 1), data, 32); BENCH(llama_sampler_init_min_p (0.2f, 1), data, 32);
BENCH(llama_sampler_init_tail_free(0.5f, 1), data, 32); BENCH(llama_sampler_init_tail_free(0.5f, 1), data, 32);
BENCH(llama_sampler_init_typical (0.5f, 1), data, 32); BENCH(llama_sampler_init_typical (0.5f, 1), data, 32);
BENCH(llama_sampler_init_softmax (), data, 32); BENCH(llama_sampler_init_xtc (1.0f, 0.1f, 0.8f, 1, 1), data, 32);
BENCH(llama_sampler_init_softmax (), data, 32);
} }
int main(void) { int main(void) {
@ -309,6 +332,11 @@ int main(void) {
test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f}, 0.76f); test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f}, 0.76f);
test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f}, 1.00f); test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f}, 1.00f);
test_xtc({0.4f, 0.3f, 0.2f, 0.1f}, {0.1f}, 0.99f, 0.10f, 1.00f);
test_xtc({0.4f, 0.3f, 0.2f, 0.1f}, {0.4f, 0.1f}, 0.99f, 0.10f, 0.35f);
test_xtc({0.4f, 0.3f, 0.2f, 0.1f}, {0.4f, 0.3f, 0.1f}, 0.99f, 0.10f, 0.25f);
test_xtc({0.4f, 0.3f, 0.2f, 0.1f}, {0.4f, 0.2f, 0.1f}, 0.99f, 0.20f, 0.35f);
test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f}, 0.25f); test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f}, 0.25f);
test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f, 0.25f}, 0.75f); test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f, 0.25f}, 0.75f);
test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f, 0.25f}, 0.99f); test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f, 0.25f}, 0.99f);