newsbang/Homer-v1.0-Qwen2.5-7B

Homer-v1.0-Qwen2.5-7B is a fine-tuned version of Qwen2.5-7B using a large amount of instruction-based data.
We released the math subset of our dataset (https://huggingface.co./datasets/newsbang/homer_math_v0.1), and we also analyzed the data leakage of current open-source math datasets on the benchmark (https://huggingface.co./datasets/newsbang/math_benbench_data_leak_analysis).

How to use

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "newsbang/Homer-v1.0-Qwen2.5-7B"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

messages = [
    {"role": "system", "content": "You are a very helpful assistant."},
    {"role": "user", "content": "Hello"}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	32.15
IFEval (0-Shot)	63.93
BBH (3-Shot)	37.81
MATH Lvl 5 (4-Shot)	30.36
GPQA (0-shot)	9.62
MuSR (0-shot)	11.88
MMLU-PRO (5-shot)	39.27