tests : add test-tokenizer-0.sh

Author: ggerganov

tests/test-tokenizer-0-bpe.py

@@ -55,8 +55,10 @@
 //    return _k_tests;
 //}
 
-static std::map<std::string, std::vector<llama_token>> read_tests(const std::string & fname_inp, const std::string & fname_out) {
-    std::map<std::string, std::vector<llama_token>> tests;
+using llama_tests = std::map<std::string, std::vector<llama_token>>;
+
+static llama_tests read_tests(const std::string & fname_inp, const std::string & fname_out) {
+    llama_tests tests;
 
     std::ifstream ifs_inp(fname_inp);
     if (!ifs_inp) {
@@ -175,12 +177,20 @@ int main(int argc, char **argv) {
 
     bool success = true;
 
-    const auto k_tests = read_tests(fname_inp, fname_out);
+    const auto k_tests = [&]() -> llama_tests {
+        if (!fname_text.empty()) {
+            return {};
+        }
 
-    if (k_tests.empty()) {
-        fprintf(stderr, "%s : error: no tests found\n", __func__);
-        return 1;
-    }
+        const auto res = read_tests(fname_inp, fname_out);
+
+        if (res.empty()) {
+            fprintf(stderr, "%s : error: no tests found\n", __func__);
+            exit(1);
+        }
+
+        return res;
+    }();
 
     const bool add_special = false;
 
@@ -238,7 +248,17 @@ int main(int argc, char **argv) {
 
         fprintf(stderr, "%s : text size: %zu\n", __func__, text.size());
 
-        const std::vector<llama_token> res = llama_tokenize(ctx, text, add_special);
+        std::vector<llama_token> res;
+
+        {
+            const auto t_start = ggml_time_us();
+
+            res = llama_tokenize(ctx, text, add_special);
+
+            const auto t_end = ggml_time_us();
+
+            fprintf(stderr, "%s : tokenized in %.3f ms (cpp)\n", __func__, (t_end - t_start) / 1000.0);
+        }
 
         fprintf(stderr, "%s : tokens: %zu\n", __func__, res.size());
 
@@ -252,7 +272,8 @@ int main(int argc, char **argv) {
             }
 
             for (const auto & tok : res) {
-                ofs << tok << " '" << string_strip(llama_detokenize_bpe(ctx, std::vector<int>{tok})) << "'" << std::endl;
+                //ofs << tok << " '" << string_strip(llama_detokenize_bpe(ctx, std::vector<int>{tok})) << "'" << std::endl;
+                ofs << tok << "\n";
             }
         }
 

tests/test-tokenizer-0-spm.py

@@ -0,0 +1,46 @@
+import time
+import argparse
+
+from transformers import AutoTokenizer
+
+parser = argparse.ArgumentParser()
+parser.add_argument("dir_tokenizer", help="directory containing 'tokenizer.model' file")
+parser.add_argument("--fname-tok",   help="path to a text file to tokenize", required=True)
+args = parser.parse_args()
+
+dir_tokenizer = args.dir_tokenizer
+fname_tok = args.fname_tok
+
+tokenizer = AutoTokenizer.from_pretrained(dir_tokenizer)
+
+print('tokenizing file: ', fname_tok)
+fname_out = fname_tok + '.tok'
+with open(fname_tok, 'r', encoding='utf-8') as f:
+    lines = f.readlines()
+    s = ''.join(lines)
+    t_start = time.time()
+    res = tokenizer.encode(s, add_special_tokens=False)
+    t_end = time.time()
+    print('\nmain : tokenized in', "{:.3f}".format(1000.0 * (t_end - t_start)), 'ms (py)')
+    with open(fname_out, 'w', encoding='utf-8') as f:
+        for x in res:
+            # LLaMA v3 for some reason strips the space for these tokens (and others)
+            # if x == 662:
+            #     f.write(str(x) + ' \' ' + tokenizer.decode(x) + '\'\n')
+            # elif x == 1174:
+            #     f.write(str(x) + ' \' ' + tokenizer.decode(x) + '\'\n')
+            # elif x == 2564:
+            #     f.write(str(x) + ' \' ' + tokenizer.decode(x) + '\'\n')
+            # elif x == 758:
+            #     f.write(str(x) + ' \' ' + tokenizer.decode(x) + '\'\n')
+            # elif x == 949:
+            #     f.write(str(x) + ' \' ' + tokenizer.decode(x) + '\'\n')
+            # elif x == 5354:
+            #     f.write(str(x) + ' \' ' + tokenizer.decode(x) + '\'\n')
+            # else:
+            #     f.write(str(x) + ' \'' + tokenizer.decode(x) + '\'\n')
+            # f.write(str(x) + ' \'' + tokenizer.decode(x).strip() + '\'\n')
+            f.write(str(x) + '\n')
+    print('len(res): ', len(res))
+    print('len(lines): ', len(lines))
+print('results written to: ', fname_out)

tests/test-tokenizer-0.cpp

@@ -0,0 +1,34 @@
+#!/bin/bash
+#
+# Usage:
+#
+#   test-tokenizer-0.sh <name> <input>
+#
+
+if [ $# -ne 2 ]; then
+    printf "Usage: $0 <name> <input>\n"
+    exit 1
+fi
+
+name=$1
+input=$2
+
+make -j tests/test-tokenizer-0
+
+printf "Testing %s on %s ...\n" $name $input
+
+python3 ./tests/test-tokenizer-0.py ./models/tokenizers/$name --fname-tok $input > /tmp/test-tokenizer-0-$name-py.log 2>&1
+cat /tmp/test-tokenizer-0-$name-py.log | grep "tokenized in"
+
+./tests/test-tokenizer-0 ./models/ggml-vocab-$name.gguf $input > /tmp/test-tokenizer-0-$name-cpp.log 2>&1
+cat /tmp/test-tokenizer-0-$name-cpp.log | grep "tokenized in"
+
+diff $input.tok $input.tokcpp > /dev/null 2>&1
+
+if [ $? -eq 0 ]; then
+    printf "Tokenization is correct!\n"
+else
+    diff $input.tok $input.tokcpp | head -n 32
+
+    printf "Tokenization differs!\n"
+fi