Minor improvements in GPT2 tokenizer (#3567)

* Fixing minor bugs in bpe_gpt2_preprocess

* Don't add bos token in test

4 files changed, 13 insertions, 15 deletions

diff --git a/tests/test-tokenizer-0-falcon.cpp b/tests/test-tokenizer-0-falcon.cpp
index 0f3c50bc..a4e9d2b9 100644
--- a/tests/test-tokenizer-0-falcon.cpp
+++ b/tests/test-tokenizer-0-falcon.cpp
@@ -36,6 +36,8 @@ static const std::map<std::string, std::vector<llama_token>> & k_tests() {
         { "   Hello"              , {     258,  23090, }, },
         { "    Hello"             , {     466,  23090, }, },
         { "    Hello\n    Hello"  , {     466,  23090,    742,  23090, }, },
+        { "\n ="                  , {    1212,     40, }, },
+        { "' era"                 , {      18,   4932, }, },
     };
 
     return _k_tests;
@@ -155,7 +157,7 @@ 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, true);
+        const std::vector<llama_token> res = llama_tokenize(ctx, text, false);
 
         fprintf(stderr, "%s : tokens: %zu\n", __func__, res.size());
 
@@ -169,10 +171,8 @@ int main(int argc, char **argv) {
             }
 
             for (const auto & tok : res) {
-                ofs << tok << " ";
+                ofs << tok << " '" << llama_detokenize_bpe(ctx, std::vector<int>{tok}) << "'" << std::endl;
             }
-
-            ofs << "\n";
         }
 
         fprintf(stderr, "%s : tokens written to '%s'\n", __func__, (fname_text + ".tokcpp").c_str());
diff --git a/tests/test-tokenizer-0-falcon.py b/tests/test-tokenizer-0-falcon.py
index 9c8c1c7d..cf65a3f6 100644
--- a/tests/test-tokenizer-0-falcon.py
+++ b/tests/test-tokenizer-0-falcon.py
@@ -41,6 +41,8 @@ tests = [
         "   Hello",
         "    Hello",
         "    Hello\n    Hello",
+        "\n =",
+        "' era",
     ]
 
 for text in tests:
@@ -69,15 +71,14 @@ fname_tok = args.fname_tok
 if fname_tok:
     print('tokenizing file: ', fname_tok)
     fname_out = fname_tok + '.tok'
-    with open(fname_tok, 'r') as f:
+    with open(fname_tok, 'r', encoding='utf-8') as f:
         lines = f.readlines()
         s = ''.join(lines)
         res = tokenizer.encode(s)
         # write to file
-        with open(fname_out, 'w') as f:
+        with open(fname_out, 'w', encoding='utf-8') as f:
             for x in res:
-                f.write(str(x) + ' ')
-            f.write('\n')
+                f.write(str(x) + ' \'' + tokenizer.decode(x) + '\'\n')
         print('len(res): ', len(res))
         print('len(lines): ', len(lines))
     print('results written to: ', fname_out)
diff --git a/tests/test-tokenizer-0-llama.cpp b/tests/test-tokenizer-0-llama.cpp
index 91c841f7..39c8d188 100644
--- a/tests/test-tokenizer-0-llama.cpp
+++ b/tests/test-tokenizer-0-llama.cpp
@@ -174,10 +174,8 @@ int main(int argc, char **argv) {
             }
 
             for (const auto & tok : res) {
-                ofs << tok << " ";
+                ofs << tok << " '" << llama_detokenize_spm(ctx, std::vector<int>{tok}) << "'" << std::endl;
             }
-
-            ofs << "\n";
         }
 
         fprintf(stderr, "%s : tokens written to '%s'\n", __func__, (fname_text + ".tokcpp").c_str());
diff --git a/tests/test-tokenizer-0-llama.py b/tests/test-tokenizer-0-llama.py
index bc164ee2..078f680b 100644
--- a/tests/test-tokenizer-0-llama.py
+++ b/tests/test-tokenizer-0-llama.py
@@ -81,15 +81,14 @@ fname_tok = args.fname_tok
 if fname_tok:
     print('tokenizing file: ', fname_tok)
     fname_out = fname_tok + '.tok'
-    with open(fname_tok, 'r') as f:
+    with open(fname_tok, 'r', encoding='utf-8') as f:
         lines = f.readlines()
         s = ''.join(lines)
         res = tokenizer.encode(s, add_bos=True)
         # write to file
-        with open(fname_out, 'w') as f:
+        with open(fname_out, 'w', encoding='utf-8') as f:
             for x in res:
-                f.write(str(x) + ' ')
-            f.write('\n')
+                f.write(str(x) + ' \'' + tokenizer.decode(x) + '\'\n')
         print('len(res): ', len(res))
         print('len(lines): ', len(lines))
     print('results written to: ', fname_out)