sknow-lab/Gemma3-12B-CIC-ACLARC

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "sknow-lab/Gemma3-12B-CIC-ACLARC"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

system_prompt = """
# CONTEXT #
You are an expert researcher tasked with classifying the intent of a citation in a scientific publication.

########

# OBJECTIVE # 
You will be given a sentence containing a citation, you must output the appropriate class as an answer.

########

# CLASS DEFINITIONS #

The six (6) possible classes are the following: "BACKGROUND", "MOTIVATION", "USES", "EXTENDS", "COMPARES_CONTRASTS", "FUTURE".

The definitions of the classes are:
1 - BACKGROUND: The cited paper provides relevant Background information or is part of the body of literature.
2 - MOTIVATION: The citing paper is directly motivated by the cited paper.
3 - USES: The citing paper uses the methodology or tools created by the cited paper.
4 - EXTENDS: The citing paper extends the methods, tools or data, etc. of the cited paper.
5 - COMPARES_CONTRASTS: The citing paper expresses similarities or differences to, or disagrees with, the cited paper.
6 - FUTURE: The cited paper may be a potential avenue for future work.

########

# RESPONSE RULES #
- Analyze only the citation marked with the @@CITATION@@ tag.
- Assign exactly one class to each citation.
- Respond only with the exact name of one of the following classes: "BACKGROUND", "MOTIVATION", "USES", "EXTENDS", "COMPARES_CONTRASTS", "FUTURE".
- Do not provide any explanation or elaboration.
"""

test_citing_sentence = "However , the method we are currently using in the ATIS domain ( @@CITATION@@ ) represents our most promising approach to this problem."

user_prompt = f"""
{test_citing_sentence}
### Question: Which is the most likely intent for this citation?
a) BACKGROUND
b) MOTIVATION
c) USES
d) EXTENDS
e) COMPARES_CONTRASTS
f) FUTURE
### Answer:
"""

messages = [
    {"role": "system", "content": system_prompt},
    {"role": "user", "content": user_prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
# Response: USES

@misc{koloveas2025llmspredictcitationintent,
      title={Can LLMs Predict Citation Intent? An Experimental Analysis of In-context Learning and Fine-tuning on Open LLMs}, 
      author={Paris Koloveas and Serafeim Chatzopoulos and Thanasis Vergoulis and Christos Tryfonopoulos},
      year={2025},
      eprint={2502.14561},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.14561}, 
}