Hi @myyim ,

Welcome to Gemma family of open source models, the message is an informative one, not an error that stops your code. It's letting you know that certain common generation parameters aren't applicable in this specific context. This is because the generate function from transformers utilized in the Gemma models, where it has specific implementation. The specific model implementations can override or ignore certain parameters.

The following code script can help you to log the logger while executing the code:

import transformers
transformers.logging.set_verbosity_info()

from transformers import AutoProcessor, Gemma3nForConditionalGeneration
from PIL import Image
import requests
import torch

model_id = "google/gemma-3n-e2b" #or any other model id

model = Gemma3nForConditionalGeneration.from_pretrained(model_id, device="cuda", torch_dtype=torch.bfloat16,).eval()

processor = AutoProcessor.from_pretrained(model_id)

url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"
image = Image.open(requests.get(url, stream=True).raw)
prompt = " in this image, there is"
model_inputs = processor(text=prompt, images=image, return_tensors="pt").to(model.device)

input_len = model_inputs["input_ids"].shape[-1]

with torch.inference_mode():
generation = model.generate(**model_inputs, max_new_tokens=10)
generation = generation[0][input_len:]

decoded = processor.decode(generation, skip_special_tokens=True)
print(decoded)

one picture of flowers which shows that the flower is

In short the above logger is a info/warn level message not an error, it will not cause any failure while your execution.

Thanks.

The generate method gives this warning when you try and set top_p and top_k to something other than the default while having the parameter do_sample set to False. In my case, I didn't set them explicitly, but they still get set by Gemma's generation_config.json file (here). The least annoying way I found to suppress that warning is by explicitly setting those parameters to the defaults that the transformers library expects for do_sample=False:

output = model.generate(
    do_sample=False,
    top_p=1.0,
    top_k=50
)

Hi @samuel500 ,

Thanks for reaching out to us, could you please open an issue under community section as a new one, so I'll investigate and get back to you with my pointers. Thanks for you patience.

Thanks.