gifted_maths_unsloth_remerged_full_gemma_4

Gifted Maths Gemma 4 — Verified Full Model

Verified 16-bit full-weight merge of Gifted Maths LoRA, a Gemma 4 model fine-tuned as maths_engine for clear, step-by-step mathematics solutions in LaTeX.

This is the authoritative full model and the conversion source for the verified GGUF release.

The earlier GGUF export is retained as a legacy artifact and is not recommended for the current maths_engine release.

Key facts

Prompt format

<start_of_turn>system
You are maths_engine, a local gifted mathematics reasoning model for students.
You solve maths problems step by step, clearly and calmly.
All variables, formulas, equations, transformations, and final answers must use LaTeX.
Use inline math with $...$ and display math with $$...$$.
Do not skip algebraic steps when the learner would benefit from seeing them.
End every solution with a clear line starting with: Final Answer:<end_of_turn>
<start_of_turn>user
{{ question }}<end_of_turn>
<start_of_turn>model

Use with Unsloth

pip install -U unsloth transformers accelerate safetensors

import torch
from unsloth import FastModel

MODEL_ID = "arabellako22/gifted_maths_unsloth_remerged_full_gemma_4"

model, tokenizer = FastModel.from_pretrained(
    model_name=MODEL_ID,
    max_seq_length=1024,
    dtype=None,
    load_in_4bit=False,
)
FastModel.for_inference(model)

messages = [
    {
        "role": "system",
        "content": """You are maths_engine, a local gifted mathematics reasoning model for students.
You solve maths problems step by step, clearly and calmly.
All variables, formulas, equations, transformations, and final answers must use LaTeX.
Use inline math with $...$ and display math with $$...$$.
Do not skip algebraic steps when the learner would benefit from seeing them.
End every solution with a clear line starting with: Final Answer:""",
    },
    {
        "role": "user",
        "content": "Find the coefficient of x^4 in (2x - 3)^6.",
    },
]
prompt = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.inference_mode():
    outputs = model.generate(
        **inputs,
        max_new_tokens=900,
        do_sample=False,
        pad_token_id=tokenizer.eos_token_id,
    )

new_tokens = outputs[0][inputs["input_ids"].shape[-1]:]
print(tokenizer.decode(new_tokens, skip_special_tokens=True))

Verification

These are narrow artifact-integrity checks, not a standardized mathematics benchmark.

Intended use

• Mathematics-assistant experiments
• Worked-example and LaTeX-formatting demonstrations
• Source model for the verified Gifted Maths GGUF release

Limitations

• The source adapter was trained on only 54 examples and validated on 7; no independent test set was created.
• The provenance and licensing of the original 50 dataset examples were not recorded.
• No standardized mathematics, factuality, safety, bias, or robustness benchmark is published.
• Generated algebra, arithmetic, proofs, notation, or LaTeX may be wrong.
• “Gifted” describes the project goal, not a validated learner classification.

Do not treat outputs as guaranteed answers or use them for high-stakes grading, admissions, or placement without independent human verification.