whisper : tokenizer fix + re-enable tokenizer test for LLaMa (#3096)

* Fix für #2721

* Reenable tokenizer test for LLaMa

* Add `console.cpp` dependency

* Fix dependency to `common`

* Fixing wrong fix.

* Make console usage platform specific

Work on compiler warnings.

* Adapting makefile

* Remove trailing whitespace

* Adapting the other parts of the makefile

* Fix typo.

Makefile

@@ -2,7 +2,7 @@
 BUILD_TARGETS = main quantize quantize-stats perplexity embedding vdot train-text-from-scratch convert-llama2c-to-ggml simple save-load-state server embd-input-test gguf llama-bench baby-llama beam-search speculative tests/test-c.o
 
 # Binaries only useful for tests
-TEST_TARGETS = tests/test-llama-grammar tests/test-grammar-parser tests/test-double-float tests/test-grad0 tests/test-opt tests/test-quantize-fns tests/test-quantize-perf tests/test-sampling tests/test-tokenizer-0-llama tests/test-tokenizer-0-falcon tests/test-tokenizer-1
+TEST_TARGETS = tests/test-llama-grammar tests/test-grammar-parser tests/test-double-float tests/test-grad0 tests/test-opt tests/test-quantize-fns tests/test-quantize-perf tests/test-sampling tests/test-tokenizer-0-llama tests/test-tokenizer-0-falcon tests/test-tokenizer-1-llama
 
 # Code coverage output files
 COV_TARGETS = *.gcno tests/*.gcno *.gcda tests/*.gcda *.gcov tests/*.gcov lcov-report gcovr-report
@@ -49,7 +49,7 @@ test: $(TEST_TARGETS)
 			./$$test_target $(CURDIR)/models/ggml-vocab-llama.gguf; \
 		elif [ "$$test_target" = "tests/test-tokenizer-0-falcon" ]; then \
 			continue; \
-		elif [ "$$test_target" = "tests/test-tokenizer-1" ]; then \
+		elif [ "$$test_target" = "tests/test-tokenizer-1-llama" ]; then \
 			continue; \
 		else \
 			echo "Running test $$test_target..."; \
@@ -605,7 +605,7 @@ tests/test-tokenizer-0-falcon: tests/test-tokenizer-0-falcon.cpp build-info.h gg
 tests/test-tokenizer-0-llama: tests/test-tokenizer-0-llama.cpp build-info.h ggml.o llama.o common.o $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
 
-tests/test-tokenizer-1: tests/test-tokenizer-1.cpp build-info.h ggml.o llama.o common.o $(OBJS)
+tests/test-tokenizer-1-llama: tests/test-tokenizer-1-llama.cpp build-info.h ggml.o llama.o common.o $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
 
 tests/test-c.o: tests/test-c.c llama.h

llama.cpp

@@ -3121,10 +3121,9 @@ struct llm_tokenizer_spm {
         while (offs < text.size()) {
             llm_symbol sym;
             size_t len = utf8_len(text[offs]);
-            GGML_ASSERT(offs + len <= text.size());
             sym.text = text.c_str() + offs;
-            sym.n = len;
+            sym.n = std::min(len, text.size() - offs);
-            offs += len;
+            offs += sym.n;
             sym.prev = index - 1;
             sym.next = offs == text.size() ? -1 : index + 1;
             index++;
@@ -6218,7 +6217,7 @@ int llama_tokenize_with_model(
     auto res = llama_tokenize_internal(model->vocab, text, add_bos);
 
     if (n_max_tokens < (int) res.size()) {
-        LLAMA_LOG_ERROR("%s: too many tokens\n", __func__);
+        // LLAMA_LOG_ERROR("%s: too many tokens\n", __func__);
         return -((int) res.size());
     }
 

tests/CMakeLists.txt

@@ -29,9 +29,8 @@ llama_build_executable(test-tokenizer-0-llama.cpp)
 llama_test_executable (test-tokenizer-0-llama test-tokenizer-0-llama.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama.gguf)
 llama_build_executable(test-tokenizer-0-falcon.cpp)
 #llama_test_executable (test-tokenizer-0-falcon test-tokenizer-0-falcon.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-falcon.gguf)
-llama_build_executable(test-tokenizer-1.cpp)
+llama_build_executable(test-tokenizer-1-llama.cpp)
-# test-tokenizer-1 requires a BPE vocab. re-enable when we have one.
+llama_test_executable (test-tokenizer-1-llama test-tokenizer-1-llama.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-llama.gguf)
-#llama_test_executable (test-tokenizer-1.llama test-tokenizer-1.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-falcon.gguf)
 #llama_test_executable(test-tokenizer-1.aquila test-tokenizer-1.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab-aquila.gguf)
 llama_build_and_test_executable(test-grammar-parser.cpp)
 llama_build_and_test_executable(test-llama-grammar.cpp)

tests/test-tokenizer-0-llama.cpp

@@ -1,5 +1,6 @@
 #include "llama.h"
 #include "common.h"
+#include "console.h"
 
 #include <cstdio>
 #include <string>
@@ -89,6 +90,12 @@ int main(int argc, char **argv) {
         return 2;
     }
 
+#ifdef _WIN32
+    // We need this for unicode console support
+    console::init(false, false);
+    atexit([]() { console::cleanup(); });
+#endif
+
     bool success = true;
 
     for (const auto & test_kv : k_tests()) {

tests/test-tokenizer-1-llama.cpp

@@ -0,0 +1,127 @@
+#include "llama.h"
+#include "common.h"
+#include "console.h"
+
+#include <cassert>
+#include <cstdio>
+#include <cstring>
+#include <string>
+#include <codecvt>
+#include <map>
+#include <vector>
+#include <locale>
+
+typedef int codepoint;
+
+std::string codepoint_to_utf8(codepoint cp) {
+    std::string result;
+    if (0x00 <= cp && cp <= 0x7f) {
+        result.push_back(cp);
+    } else if (0x80 <= cp && cp <= 0x7ff) {
+        result.push_back(0xc0 | ((cp >> 6) & 0x1f));
+        result.push_back(0x80 | (cp & 0x3f));
+    } else if (0x800 <= cp && cp <= 0xffff) {
+        result.push_back(0xe0 | ((cp >> 12) & 0x0f));
+        result.push_back(0x80 | ((cp >> 6) & 0x3f));
+        result.push_back(0x80 | (cp & 0x3f));
+    } else if (0x10000 <= cp && cp <= 0x10ffff) {
+        result.push_back(0xf0 | ((cp >> 18) & 0x07));
+        result.push_back(0x80 | ((cp >> 12) & 0x3f));
+        result.push_back(0x80 | ((cp >> 6) & 0x3f));
+        result.push_back(0x80 | (cp & 0x3f));
+    } else {
+        throw std::invalid_argument("invalid codepoint");
+    }
+    return result;
+}
+
+int main(int argc, char **argv) {
+    if (argc < 2) {
+        fprintf(stderr, "Usage: %s <vocab-file>\n", argv[0]);
+        return 1;
+    }
+
+    const std::string fname = argv[1];
+
+    fprintf(stderr, "%s : reading vocab from: '%s'\n", __func__, fname.c_str());
+
+    llama_model * model;
+    llama_context * ctx;
+
+    llama_backend_init(false);
+
+    // load the vocab
+    {
+        auto lparams = llama_context_default_params();
+
+        lparams.vocab_only = true;
+
+        model = llama_load_model_from_file(fname.c_str(), lparams);
+
+        if (model == NULL) {
+            fprintf(stderr, "%s: error: failed to load vocab '%s'\n", __func__, fname.c_str());
+            return 1;
+        }
+
+        ctx = llama_new_context_with_model(model, lparams);
+
+        if (ctx == NULL) {
+            fprintf(stderr, "%s: error: failed to load vocab '%s'\n", __func__, fname.c_str());
+            llama_free_model(model);
+            return 1;
+        }
+    }
+
+    GGML_ASSERT(llama_vocab_type(ctx) == LLAMA_VOCAB_TYPE_SPM);
+
+#ifdef _WIN32
+    // We need this for unicode console support
+    console::init(false, false);
+    atexit([]() { console::cleanup(); });
+#endif
+
+    const int n_vocab = llama_n_vocab(ctx);
+
+    for (int i = 0; i < n_vocab; ++i) {
+        std::string str = llama_detokenize_spm(ctx, std::vector<int>(1, i));
+        std::vector<llama_token> tokens = llama_tokenize(ctx, str, false);
+        std::string check = llama_detokenize_spm(ctx, tokens);
+        if (check != str) {
+            fprintf(stderr, "%s : error: token %d detokenizes to >%s<(%llu) but tokenization of this detokenizes to >%s<(%llu)\n",
+                __func__, i, str.c_str(), str.length(), check.c_str(), check.length());
+            if(i != 3)
+                return 2;
+        }
+    }
+
+    for (codepoint cp = 0x0000; cp < 0xffff; ++cp) {
+        if (cp < 0xd800 || cp > 0xdfff) {
+            std::string str = codepoint_to_utf8(cp);
+            std::vector<llama_token> tokens = llama_tokenize(ctx, str, false);
+            std::string check = llama_detokenize_spm(ctx, tokens);
+            if (str != check) {
+                fprintf(stderr, "%s : error: codepoint %d detokenizes to >%s<(%llu) instead of >%s<(%llu)\n",
+                    __func__, cp, check.c_str(), check.length(), str.c_str(), str.length());
+                if(cp != 0 && cp != 9601)
+                    return 3;
+            }
+        }
+    }
+    for (codepoint cp = 0x10000; cp < 0x0010ffff; ++cp) {
+        std::string str = codepoint_to_utf8(cp);
+        std::vector<llama_token> tokens = llama_tokenize(ctx, str, false);
+        std::string check = llama_detokenize_spm(ctx, tokens);
+        if (str != check) {
+            fprintf(stderr, "%s : error: codepoint %d detokenizes to >%s<(%llu) instead of >%s<(%llu)\n",
+                __func__, cp, check.c_str(), check.length(), str.c_str(), str.length());
+            return 4;
+        }
+    }
+
+    llama_free_model(model);
+    llama_free(ctx);
+
+    llama_backend_free();
+
+    return 0;
+}

tests/test-tokenizer-1.cpp

@@ -1,108 +0,0 @@
-#include "llama.h"
-#include "common.h"
-
-#include <cassert>
-#include <cstdio>
-#include <cstring>
-#include <string>
-#include <codecvt>
-#include <map>
-#include <vector>
-#include <locale>
-
-static std::string escape_whitespace(const std::string& text) {
-    std::string result = "\xe2\x96\x81";
-    for (size_t offs = 0; offs < text.length(); ++offs) {
-        if (text[offs] == ' ') {
-            result += "\xe2\x96\x81";
-        } else {
-            result += text[offs];
-        }
-    }
-    return result;
-}
-
-int main(int argc, char **argv) {
-    if (argc < 2) {
-        fprintf(stderr, "Usage: %s <vocab-file>\n", argv[0]);
-        return 1;
-    }
-
-    const std::string fname = argv[1];
-
-    fprintf(stderr, "%s : reading vocab from: '%s'\n", __func__, fname.c_str());
-
-    llama_model * model;
-    llama_context * ctx;
-
-    llama_backend_init(false);
-
-    // load the vocab
-    {
-        auto lparams = llama_context_default_params();
-
-        lparams.vocab_only = true;
-
-        model = llama_load_model_from_file(fname.c_str(), lparams);
-
-        if (model == NULL) {
-            fprintf(stderr, "%s: error: failed to load vocab '%s'\n", __func__, fname.c_str());
-            return 1;
-        }
-
-        ctx = llama_new_context_with_model(model, lparams);
-
-        if (ctx == NULL) {
-            fprintf(stderr, "%s: error: failed to load vocab '%s'\n", __func__, fname.c_str());
-            llama_free_model(model);
-            return 1;
-        }
-    }
-
-    GGML_ASSERT(llama_vocab_type(ctx) == LLAMA_VOCAB_TYPE_BPE);
-
-    const int n_vocab = llama_n_vocab(ctx);
-
-    for (int i = 0; i < n_vocab; ++i) {
-        std::string forward = llama_token_to_piece(ctx, i);
-        std::vector<llama_token> tokens = llama_tokenize(ctx, forward, false);
-        if (tokens.size() == 1) {
-            if (i != tokens[0]) {
-                std::string backward = llama_token_to_piece(ctx, tokens[0]);
-                fprintf(stderr, "%s : error: token %d is string %s but bpe returns token %d %s\n",
-                    __func__, i, llama_token_to_piece(ctx, i).c_str(), tokens[0], backward.c_str());
-                return 2;
-            }
-        }
-    }
-
-#ifdef _WIN32
-    std::wstring_convert<typename std::codecvt_utf8<char16_t>, char16_t> u16converter;
-    for (char16_t ch = 0x0000; ch < 0xffff; ++ch) {
-        std::u16string u16str(1, ch);
-        std::string str = u16converter.to_bytes(u16str);
-        std::vector<llama_token> tokens = llama_tokenize(ctx, escape_whitespace(str).c_str(), false);
-        if (tokens.size() == 1) {
-            fprintf(stderr, "%s : info: %s tokenized to %d \n",
-                __func__, str.c_str(), tokens[0]);
-        }
-    }
-
-    std::wstring_convert<typename std::codecvt_utf8<char32_t>, char32_t> u32converter;
-    for (char32_t ch = 0x0000; ch < 0x0010ffff; ++ch) {
-        std::u32string u32str(1, ch);
-        std::string str = u32converter.to_bytes(u32str);
-        std::vector<llama_token> tokens = llama_tokenize(ctx, escape_whitespace(str).c_str(), false);
-        if (tokens.size() == 1) {
-            fprintf(stderr, "%s : info: %s tokenized to %d \n", __func__, str.c_str(), tokens[0]);
-        }
-    }
-#endif
-
-    llama_free_model(model);
-    llama_free(ctx);
-
-    llama_backend_free();
-
-    return 0;
-}