Skip to content

qwen3 models map to wrong tokenizer class on the hub #49116

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
itazap merged 2 commits into from
Sep 28, 2026
Merged

qwen3 models map to wrong tokenizer class on the hub #49116

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1b596ca
qwen3 models map to wrong tokenizer class on the hub
itazap Sep 25, 2026
462bf5a
Merge branch 'main' into qwen3_tok
itazap Sep 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions src/transformers/models/auto/tokenization_auto.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -293,6 +293,7 @@
("qwen3", "Qwen2Tokenizer" if is_tokenizers_available() else None),
("qwen3_5", "Qwen3_5Tokenizer" if is_tokenizers_available() else None),
("qwen3_5_moe", "Qwen3_5Tokenizer" if is_tokenizers_available() else None),
("qwen3_5_moe_text", "Qwen3_5Tokenizer" if is_tokenizers_available() else None),
("qwen3_5_text", "Qwen3_5Tokenizer" if is_tokenizers_available() else None),
("qwen3_asr", "Qwen2Tokenizer" if is_tokenizers_available() else None),
("qwen3_moe", "Qwen2Tokenizer" if is_tokenizers_available() else None),
Expand Down Expand Up @@ -414,6 +415,10 @@
"phi3_v",
"phimoe",
"qwen2",
"qwen3_5",
"qwen3_5_moe",
"qwen3_5_moe_text",
"qwen4_exp",
"step3p5",
"step3_vl",
"vipllava",
Expand Down
26 changes: 25 additions & 1 deletion tests/models/qwen3_5/test_tokenization_qwen3_5.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,9 @@

import unittest

from transformers import Qwen3_5Tokenizer
from parameterized import parameterized

from transformers import AutoTokenizer, Qwen3_5Tokenizer, TokenizersBackend
from transformers.testing_utils import require_tokenizers

from ...test_tokenization_common import TokenizerTesterMixin
Expand Down Expand Up @@ -63,3 +65,25 @@ def test_attention_mask_is_a_model_input(self):
tokenizer = self.get_tokenizer()

self.assertEqual(tokenizer.model_input_names, ["input_ids", "attention_mask"])

@parameterized.expand(
[
"Qwen/Qwen3.8-Flash-Next", # model_type: qwen4_exp
"Qwen/Qwen3.6-27B", # model_type: qwen3_5
"Qwen/Qwen3.6-35B-A3B", # model_type: qwen3_5_moe
"Qwen/Qwen3.8-2.4T-A95B", # model_type: qwen3_5_moe_text
]
)
def test_auto_skips_incorrect_hub_tokenizer_class(self, repo_id):
"""These checkpoints' hub tokenizer_config.json incorrectly sets tokenizer_class=Qwen2Tokenizer.
AutoTokenizer should use the registered model_type mapped Qwen3_5Tokenizer instead,
matching the hub's tokenizer.json."""
tokenizer_auto = AutoTokenizer.from_pretrained(repo_id)
tokenizer_tok = TokenizersBackend.from_pretrained(repo_id)

self.assertIsInstance(tokenizer_auto, Qwen3_5Tokenizer)

text = "नमस्ते दुनिया"
auto_ids = tokenizer_auto.encode(text, add_special_tokens=False)
tok_ids = tokenizer_tok.encode(text, add_special_tokens=False)
self.assertEqual(auto_ids, tok_ids)
Loading