Spaces:

---

FINAL-Bench

/

Darwin-35B-A3B-Opus

like 20

Sleeping

Hi, thanks for reporting this. The issue was caused by a tokenizer type mismatch in tokenizer_config.json. It has been fixed now. Please try again!

This error occurs due to tokenizer type mismatch.
Please load the tokenizer with use_fast=True:

tokenizer = AutoTokenizer.from_pretrained(
"MODEL_PATH",
trust_remote_code=True,
use_fast=True
)

Or use the parent tokenizer directly:

tokenizer = AutoTokenizer.from_pretrained(
"Qwen/Qwen3.5-9B",
trust_remote_code=True
)

I'm not using it programmatically but through this web interface on my phone:
https://huggingface.co/spaces/FINAL-Bench/Darwin-35B-A3B-Opus

So the problem is in the interface.

Hi, thanks for the detailed report. You were right — the tokenizer_config.json had an incorrect tokenizer_class setting. This has been fixed now. It should work directly from the web interface without any workarounds. Please give it another try!

Hi, thanks for the detailed report. You were right — the tokenizer_config.json had an incorrect tokenizer_class setting. This has been fixed now. It should work directly from the web interface without any workarounds. Please give it another try!

Unfortunately it still happens. I also tried incognito mode to make sure that it's not that my browser has cached the old version.

Hi, thanks for the detailed report. You were right — the tokenizer_config.json had an incorrect tokenizer_class setting. This has been fixed now. It should work directly from the web interface without any workarounds. Please give it another try!

Unfortunately it still happens. I also tried incognito mode to make sure that it's not that my browser has cached the old version.

Thanks for letting me know. The model repository has been updated, but the backend inference server needs to be restarted to pick up the changes. We're working on restarting it now — please try again in a little while. Sorry for the inconvenience!

Retry, after 5 minute