ttrl-opt

ttrl-opt is a Qwen3-4B-Instruct-2507 checkpoint specialized for operations-research modeling. It serves as the TTRL baseline for generating structured reasoning, linear-programming formulations, and solver-oriented Python code from natural-language optimization problems.

This release is the checkpoint from training step 125. It was trained with GRPO and TTRL-style majority voting using Gurobi-oriented nine-step optimization outputs and solver-based reward signals.

Quick start

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "JR-James-0125/ttrl-opt"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "user", "content": "Formulate and solve this optimization problem: ..."}
]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt",
).to(model.device)

outputs = model.generate(inputs, max_new_tokens=2048)
print(tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True))

Intended use and limitations

This model is released for research on mathematical optimization and operations-research reasoning. Generated formulations, coefficients, constraints, solver code, and claimed solutions may be incorrect, infeasible, or unsafe to use without review. Validate outputs with an appropriate solver and independent checks before using them in consequential settings.

No standalone evaluation results are included in this initial model card.