ToolACE-2.5-Llama-3.1-8B-heretic

ToolACE-2.5-Llama-3.1-8B — Abliterated

Abliterated version of Team-ACE/ToolACE-2.5-Llama-3.1-8B.

Base: Llama 3.1 8B fine-tuned for function/tool calling (ToolACE-2.5 dataset).

Abliteration

Performed with heretic — Optuna multi-objective optimization.

• Trials: 500 (50 × 10 parallel GPUs)
• Result: 0 refusals on eval set

Tool Calling Format

Uses JSON function calling via the Llama 3.1 chat template:

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "nitrox/ToolACE-2.5-Llama-3.1-8B-heretic",
    device_map="auto",
    torch_dtype="bfloat16",
)
tokenizer = AutoTokenizer.from_pretrained("nitrox/ToolACE-2.5-Llama-3.1-8B-heretic")

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get current weather for a location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "City name"}
                },
                "required": ["location"]
            }
        }
    }
]

messages = [{"role": "user", "content": "What's the weather in Paris?"}]
inputs = tokenizer.apply_chat_template(
    messages, tools=tools, add_generation_prompt=True, return_tensors="pt"
).to(model.device)
outputs = model.generate(inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True))

Disclaimer

Refusal mechanisms have been removed. Use responsibly and in accordance with applicable laws.