fix: Transformers Model no template cast stop_sequences to list

[mrsndmn]

Hi!

Running Transformers models with no template leads to following error. Can be fixed with casting batch[0].stop_sequences to list.

# src/lighteval/models/transformers/transformers_model.py
            for batch in tqdm(
                dataloader, desc="Greedy generation", position=1, leave=False, disable=self.disable_tqdm
            ):
                contexts = [self.prompt_manager.prepare_prompt(doc) for doc in batch]

                # For chat models, generation stops with EOS token, so we don't need to specify stop tokens
                if self.use_chat_template:
                    stop_tokens = [self.tokenizer.eos_token]
                else:
                    # NOTE: we are assuming all items in a batch behave similarly (same
                    # stop_tokens and max_tokens genrated) which is not necessarily
                    # the case! Because of that we only use batch size of 1
>                   stop_tokens = [self.tokenizer.eos_token] + batch[0].stop_sequences
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E                   TypeError: can only concatenate list (not "tuple") to list

How to reproduce bug

from transformers import AutoModelForCausalLM

from lighteval.logging.evaluation_tracker import EvaluationTracker
from lighteval.models.transformers.transformers_model import TransformersModel, TransformersModelConfig
from lighteval.pipeline import ParallelismManager, Pipeline, PipelineParameters


MODEL_NAME = "unsloth/Llama-3.2-1B" # no chat template model
BENCHMARKS = "mmlu_pro"

evaluation_tracker = EvaluationTracker(output_dir="./results")
pipeline_params = PipelineParameters(
  launcher_type=ParallelismManager.NONE,
  max_samples=2
)

model = AutoModelForCausalLM.from_pretrained(
MODEL_NAME, device_map="auto"
)
config = TransformersModelConfig(model_name=MODEL_NAME, batch_size=1)
model = TransformersModel.from_model(model, config)

pipeline = Pipeline(
  model=model,
  pipeline_parameters=pipeline_params,
  evaluation_tracker=evaluation_tracker,
  tasks=BENCHMARKS,
)

results = pipeline.evaluate()
pipeline.show_results()
results = pipeline.get_results()