Genesis-8B

Genesis-8B is a QLoRA fine-tune of Qwen3-8B for agentic tool-calling in drug-discovery workflows. It is trained to select the correct tool for a request, extract structured arguments (SMILES strings, property names, targets), and synthesize a final natural-language answer once a tool returns a result.

Model Details

Base model	unsloth/Qwen3-8B
Method	QLoRA (r=16, α=16), trained with Unsloth
Training data	lik98/TCDD — Tool-Calling Dataset for Drug Discovery, from the DrugPilot paper (arXiv:2505.13940)
Context length	8,192 tokens
Hardware	Single NVIDIA T4
License	CC-BY-NC-SA-4.0 (inherited from the training data) — non-commercial, share-alike

Format

Genesis-8B uses ChatML with Hermes-style tool calling. Tool schemas are declared in the system message inside <tools> tags; the model responds with a <tool_call> block, reads the result from a <tool_response> message, and produces a final answer.

<|im_start|>system
{system_prompt}
<tools>
[{tool_schema_json}]
</tools><|im_end|>
<|im_start|>user
{user_request}<|im_end|>
<|im_start|>assistant
<tool_call>
{"name": "...", "arguments": {...}}
</tool_call><|im_end|>
<|im_start|>user
<tool_response>
{tool_result_json}
</tool_response><|im_end|>
<|im_start|>assistant
{final_answer}<|im_end|>

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "Rumiii/Genesis-8B"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

system_prompt = (
    "You are a function calling AI model. You are provided with function "
    "signatures within <tools></tools> XML tags. You may call one or more "
    "functions to assist with the user query. Don't make assumptions about "
    "what values to plug into functions. For each function call return a "
    "json object with function name and arguments within <tool_call></tool_call> "
    "XML tags as follows: <tool_call>{\"name\": <function-name>,\"arguments\": "
    "<args-dict>}</tool_call>."
)

tools = [{
    "name": "drug_property_prediction",
    "description": "Predict properties of drugs based on SMILES",
    "parameters": {
        "type": "object",
        "properties": {
            "drug_smiles": {"type": "array", "items": {"type": "string"}},
            "property_name": {"type": "string"},
        },
    },
}]

import json
prompt = (
    f"<|im_start|>system\n{system_prompt}\n<tools>\n{json.dumps(tools)}\n</tools><|im_end|>\n"
    f"<|im_start|>user\nWhat is the free energy of dissolution for CCO?<|im_end|>\n"
    f"<|im_start|>assistant\n"
)

inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=False).to(model.device)
output = model.generate(**inputs, max_new_tokens=200, do_sample=False)
print(tokenizer.decode(output[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Intended Use

Research and portfolio use in agentic tool-calling for biomedical and drug-discovery workflows. Tool outputs seen during training were synthetic and not experimentally validated — this model is not intended for clinical, diagnostic, or production decision-making.

Citation

If you use this model, please also cite the underlying dataset and base model:

@article{li2025drugpilot,
  title   = {DrugPilot: LLM-based Parameterized Reasoning Agent for Drug Discovery},
  author  = {Li, Kun and Wu, Zhennan and Wang, Shoupeng and Wu, Jia and Pan, Shirui and Hu, Wenbin},
  journal = {arXiv preprint arXiv:2505.13940},
  year    = {2025}
}