LM Eval Tasks not found

Hi it seems that any of the tasks from the lm_eval benchmarks trigger "Task not found". I have attached the warning log and used the format from the README.md (I think we can ignore the fact I have no access to the underlying Mistral model as I have tired this with other models I can run and it warns me the same). I have been able to run other benchmarks such as GPQADiamond or MATH500, but anything from lm_eval seems to trigger this warning.

evalchemy                          0.1.0           /home/aahila/evalchemy
lm_eval                            0.4.7
vllm                               0.10.2

(myenv) aahila$ python -m eval.eval \
    --model hf \
    --tasks mmlu \
    --model_args "pretrained=mistralai/Mistral-7B-Instruct-v0.3" \
    --batch_size 2 \
    --output_path logs
INFO 09-18 21:50:10 [__init__.py:216] Automatically detected platform cuda.
WARNING: HF_HUB_CACHE environment variable is not set, using default cache directory ~/.cache/huggingface/hub for database utils
2025-09-18:21:50:12,384 WARNING  [task.py:155] OPENAI_API_KEY not set. Tasks requiring OpenAI will be skipped.
2025-09-18:21:50:20,720 INFO     [eval.py:381] Selected Tasks: ['mmlu']
2025-09-18:21:50:20,720 WARNING  [task.py:309] Task not found: mmlu
2025-09-18:21:50:20,974 INFO     [huggingface.py:133] Using device 'cuda'
2025-09-18:21:50:21,106 ERROR    [eval.py:404] Failed to initialize model: You are trying to access a gated repo.
Make sure to have access to it at https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3.
401 Client Error. (Request ID: Root=1-68cc7e9d-20e392fc208f834f5dd24e67;8fa548b6-be1c-486e-af1f-129a2fc3b1c1)

Cannot access gated repo for url https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/resolve/main/config.json.
Access to model mistralai/Mistral-7B-Instruct-v0.3 is restricted. You must have access to it and be authenticated to access it. Please log in.

Start with the README.md command and the eval.eval entry point, then inspect how the --tasks value is passed to lm_eval 0.4.7. Reproduce with an accessible model, compare the supported task names with mmlu, and trace why lm_eval tasks are reported as missing. Done means the reported lm_eval task can be selected or the incompatibility is documented with a clear supported invocation.