GemmAMR-fr-v1

Model Card for Model ID

GemmAMR-fr-v1 is an AMR semantic parser for the French language.

Model Details

Model Description

• Developed by: Laboratoire ERIC
• Funded by: ANR-25-CE45-0054 IMAGE-TEXTE-AVC
• Languages: input: French, English; output: AMR
• Finetuned from model: google/translategemma-4b-it

Model Sources

• Paper: A Decoder for French AMR Parsing. T Checchin, J Jacques, A Guille. TALN 2026
• Demo: Notebook

How to Get Started with the Model

Use the code below to get started with the model.

from transformers import AutoTokenizer, AutoModelForCausalLM

# load model
tokenizer = AutoTokenizer.from_pretrained("adrienguille/gemmamr-fr-v1")
model = AutoModelForCausalLM.from_pretrained(
    "adrienguille/gemmamr-fr-v1",
    dtype="auto",
    device_map="auto",
    trust_remote_code=True
)

# preprocess input
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "text",
                "source_lang_code": "fr",
                "text": "Un caillot a obstrué une artère cérébrale." # <-- input a sentence here
            }
        ],
    }
]
formatted_input = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

# generate AMR
inputs = tokenizer(
    [formatted_input],
    max_length=512,
    padding=True,
    truncation=True,
    return_tensors="pt"
).to("cuda")
output_ids = model.generate(
    inputs["input_ids"],
    attention_mask=inputs["attention_mask"],
    max_length=512
)
decoded_amr = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].split("\nmodel\n")[-1]
print(decoded_amr)

Citation

Thomas Checchin, Julien Jacques, Adrien Guille. Un décodeur pour l'analyse sémantique AMR en français. Actes des 33ème Conférence sur le Traitement Automatique des Langues Naturelles (TALN 2026)

BibTeX:

@inproceedings{Checchin-Jacques-Guille:CORIA-TALN:2026, author = "Checchin, Thomas and Jacques, Julien and Guille, Adrien", title = "Un décodeur pour l'analyse sémantique AMR en français", booktitle = "Actes de CORIA-TALN 2026. Actes des 33ème Conférence sur le Traitement Automatique des Langues Naturelles. Volume 1 : articles scientifiques originaux", month = "6", year = "2026", address = "Nantes, France", publisher = "Association pour le Traitement Automatique des Langues", pages = "610-620", url = "https://talnarchives.atala.org/TALN/TALN-2026/19.pdf" }