When I'm downloading the weights, the cell keeps running and doesn't stop. I need to fine tune Mistral-Small-3.1-24B-Instruct-2503 model

from transformers import AutoTokenizer, MistralForCausalLM
import torch

model_id = “mistralai/Mistral-Small-3.1-24B-Instruct-2503”

tokenizer = AutoTokenizer.from_pretrained(model_id,
trust_remote_code=True,
cache_dir=“/content/huggingface_cache”)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = “right”

model = MistralForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.float16,
device_map=“auto”,
cache_dir=“/content/huggingface_cache”,
low_cpu_mem_usage=True,
offload_folder=“offload”,
)

I have used Older_version = transformers==4.49.0 and Current_version = transformers==4.52.0.dev0, I tried both version but didn’t gets the solution.
Please help us! Thanks

1 Like

Hmm… Gated model issue?

Ya, I already got the access for this model. But cell keeps running, is there any other way or option for it.

1 Like

Actually, model weights are downloaded but at the end cell keeps running.

1 Like

Hmm… “Cell” probably refers to some kind of notebook environment.

There may be cache-related issues occurring occasionally in from_pretrained. In that case, rebuilding the virtual environment may resolve the issue…

Additionally, to determine whether this is a specific issue with the Mistral model (repository), testing with a smaller model should help isolate the problem.

model_id = "HuggingFaceTB/SmolLM2-135M-Instruct"