| 29 | static const int N_CASES = sizeof(CASES) / sizeof(CASES[0]); |
| 30 | |
| 31 | int main(int argc, char ** argv) { |
| 32 | if (argc < 2) { |
| 33 | fprintf(stderr, "Usage: %s <model.ggml>\n", argv[0]); |
| 34 | return 1; |
| 35 | } |
| 36 | |
| 37 | if (!sam3_test_load_tokenizer(argv[1])) { |
| 38 | fprintf(stderr, "FAIL: could not load tokenizer from '%s'\n", argv[1]); |
| 39 | return 1; |
| 40 | } |
| 41 | |
| 42 | int n_pass = 0; |
| 43 | for (int i = 0; i < N_CASES; ++i) { |
| 44 | auto tokens = sam3_test_tokenize(CASES[i].text); |
| 45 | const auto & exp = CASES[i].expected; |
| 46 | bool pass = true; |
| 47 | |
| 48 | // Must be exactly 32 tokens (context length) |
| 49 | if ((int)tokens.size() != 32) { |
| 50 | fprintf(stderr, "FAIL [%d]: expected 32 tokens, got %d\n", i, (int)tokens.size()); |
| 51 | pass = false; |
| 52 | } |
| 53 | |
| 54 | // Check non-padding prefix matches expected exactly |
| 55 | for (size_t j = 0; j < exp.size() && j < tokens.size(); ++j) { |
| 56 | if (tokens[j] != exp[j]) { |
| 57 | fprintf(stderr, "FAIL [%d]: token[%zu] = %d, expected %d\n", |
| 58 | i, j, tokens[j], exp[j]); |
| 59 | pass = false; |
| 60 | } |
| 61 | } |
| 62 | |
| 63 | // Check remaining positions are zero-padded |
| 64 | for (size_t j = exp.size(); j < tokens.size(); ++j) { |
| 65 | if (tokens[j] != 0) { |
| 66 | fprintf(stderr, "FAIL [%d]: expected 0 at position %zu, got %d\n", |
| 67 | i, j, tokens[j]); |
| 68 | pass = false; |
| 69 | break; |
| 70 | } |
| 71 | } |
| 72 | |
| 73 | // Print result |
| 74 | fprintf(stderr, "%s [%d] \"%s\" → [", pass ? "PASS" : "FAIL", i, CASES[i].text); |
| 75 | for (size_t j = 0; j < tokens.size(); ++j) { |
| 76 | if (j > 0) fprintf(stderr, ", "); |
| 77 | fprintf(stderr, "%d", tokens[j]); |
| 78 | if (tokens[j] == 0 && j > 1) { fprintf(stderr, ", ..."); break; } |
| 79 | } |
| 80 | fprintf(stderr, "]\n"); |
| 81 | |
| 82 | if (pass) n_pass++; |
| 83 | } |
| 84 | |
| 85 | fprintf(stderr, "\n%d/%d tests passed\n", n_pass, N_CASES); |
| 86 | return (n_pass == N_CASES) ? 0 : 1; |
| 87 | } |
nothing calls this directly
no test coverage detected