---
license: apache-2.0
language:
- en
base_model:
- aaditya/Llama3-OpenBioLLM-8B
---

# OpenBioLLM-Text2Graph-8B

This model is a biomedical annotation model designed to generate named entity annotations from unlabeled biomedical text.  
It was introduced in the paper [GLiNER-BioMed: A Suite of Efficient Models for Open Biomedical Named Entity Recognition](https://arxiv.org/abs/2504.00676).  

This model enables **high-throughput, cost-efficient synthetic biomedical NER data generation**, serving as the synthetic annotation backbone for [GLiNER-BioMed models](https://huggingface.co/collections/knowledgator/gliner-biomed-67ecf1b7cc62e673dbc8b57f).

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "Ihor/OpenBioLLM-Text2Graph-8B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
tokenizer.chat_template = "{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] | trim + '<|end_of_text|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{% if add_generation_prompt %}{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}{% endif %}"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    device_map="auto",
    torch_dtype=torch.bfloat16
)

MESSAGES = [
    {
        "role": "system",
        "content": (
            "You are an advanced assistant trained to process biomedical text for Named Entity Recognition (NER) and Relation Extraction (RE). "
            "Your task is to analyze user-provided text and provided entities selecting all unique and contextually relevant entities, and infer directed relationships. "
            "between these entities based on the context. Ensure that all relations exist only between annotated entities. "
            "Entities and relationships should be human-readable and natural, reflecting real-world concepts and connections. "
            "Output the annotated data in JSON format, structured as follows:\n\n"
            """{"entities": [{"id": 0, "text": "ner_string_0", "type": "ner_type_string_0"}, {"id": 1, "text": "ner_string_1", "type": "ner_type_string_1"}], "relations": [{"head": 0, "tail": 1, "type": "re_type_string_0"}]}"""
            "\n\nEnsure that the output captures all significant entities and their directed relationships in a clear and concise manner."
        ),
    },
    {
        "role": "user",
        "content": (
            'Here is a text input: "John received 3mg of aspirin." '
            """Here is the list of input entities: ['John', '3mg', 'aspirin']"""
            "Analyze this text, select and classify the entities, and extract their relationships as per your instructions."
        ),
    }
]

chat_prompt = tokenizer.apply_chat_template(
    MESSAGES, tokenize=False, add_generation_prompt=True
)

inputs = tokenizer(chat_prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=3000,
    do_sample=False,
    eos_token_id=tokenizer.eos_token_id,
    pad_token_id=tokenizer.eos_token_id,
    return_dict_in_generate=True
)

prompt_len = inputs["input_ids"].shape[-1]
generated_ids = outputs.sequences[0][prompt_len:]
response = tokenizer.decode(generated_ids, skip_special_tokens=True)
print(response)
```

## Citation

If you use this model, please cite:

```bibtex
@misc{yazdani2025glinerbiomedsuiteefficientmodels,
      title={GLiNER-BioMed: A Suite of Efficient Models for Open Biomedical Named Entity Recognition}, 
      author={Anthony Yazdani and Ihor Stepanov and Douglas Teodoro},
      year={2025},
      eprint={2504.00676},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2504.00676}, 
}
```