Paschalidis-NOC-Lab/PandemIQ-Llama

PandemIQ Llama

PandemIQ Llama is a domain-adapted LLM for pandemic intelligence, built by continuous pre-training of Llama-3.1-8B on 5.8 billion tokens of pandemic-specific text. It powers the BEACON platform (Biothreats Emergence, Analysis and Communications Network), an open-source informal surveillance program designed to revolutionize global biothreats surveillance and response.

Quick Start

Download Model from Hugging Face

Method 1: Using Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "Paschalidis-NOC-Lab/PandemIQ-Llama",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Paschalidis-NOC-Lab/PandemIQ-Llama")

Method 2: Using huggingface_hub

from huggingface_hub import snapshot_download

model_path = snapshot_download(
    repo_id="Paschalidis-NOC-Lab/PandemIQ-Llama",
    cache_dir="./models"
)

Method 3: Using CLI

huggingface-cli download Paschalidis-NOC-Lab/PandemIQ-Llama --local-dir ./PandemIQ-Llama

Serve with vLLM

from vllm import LLM, SamplingParams

# Initialize vLLM
llm = LLM(model="Paschalidis-NOC-Lab/PandemIQ-Llama")
sampling_params = SamplingParams(temperature=0.7, top_p=0.9, max_tokens=512)

# Generate
prompts = ["Input your question here"]
outputs = llm.generate(prompts, sampling_params)

for output in outputs:
    print(output.outputs[0].text)

Fine-tune with LoRA

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import LoraConfig, get_peft_model
from trl import SFTTrainer, SFTConfig
from datasets import load_dataset

# Load model and tokenizer
model = AutoModelForCausalLM.from_pretrained("Paschalidis-NOC-Lab/PandemIQ-Llama")
tokenizer = AutoTokenizer.from_pretrained("Paschalidis-NOC-Lab/PandemIQ-Llama")
tokenizer.pad_token = tokenizer.eos_token

# Configure LoRA
lora_config = LoraConfig(
    r=16,
    lora_alpha=32,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj"],
    lora_dropout=0.05,
    task_type="CAUSAL_LM"
)
model = get_peft_model(model, lora_config)

# Load dataset
train_dataset = load_dataset("json", data_files="train.json", split="train")

# Training configuration
training_args = SFTConfig(
    output_dir="./results",
    num_train_epochs=1,
    per_device_train_batch_size=1,
    learning_rate=5e-5,
    bf16=True
)

# Train
trainer = SFTTrainer(
    model=model,
    train_dataset=train_dataset,
    args=training_args,
    processing_class=tokenizer
)
trainer.train()

Citation

@inproceedings{pandemiqllama2026,
  title={PandemIQ Llama: A Domain-Adapted Foundation Model for Enhanced Pandemic Intelligence},
  author={Yang, Jingmei and Talaei, Mahtab and Lassmann, Britta and Bhadelia, Nahid and Paschalidis, Ioannis Ch.},
  booktitle={AAAI},
  year={2026}
}