Fast-Math-Qwen3-14B

By applying SFT and GRPO on difficult math problems, we enhanced the performance of DeepSeek-R1-Distill-Qwen-14B and developed Fast-Math-R1-14B, which achieves approx. 30% faster inference on average, while maintaining accuracy.

In addition, we trained and open-sourced Fast-Math-Qwen3-14B, an efficiency-optimized version of Qwen3-14B`, following the same approach.

Compared to Qwen3-14B, this model enables approx. 65% faster inference on average, with minimal loss in performance.

Technical details can be found in our github repository.

Note: This model likely inherits the ability to perform inference in TIR mode from the original model. However, all of our experiments were conducted in CoT mode, and its performance in TIR mode has not been evaluated.

Evaluation

		AIME 2024		AIME 2025
Model	Token budget	Pass@1 (avg. 64)	Output tokens	Pass@1 (avg. 64)	Output tokens
Qwen3-14B	32000	79.3	13324	69.5	15165
	16000	65.5	9179	51.5	9724
	8000	29.7	5926	20.1	5484
Fast-Math-Qwen3-14B	32000	77.6	9668	66.6	11950
	16000	72.8	7161	60.7	7874
	8000	51.6	4778	36.9	4531

Inference

vLLM

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer
model_path = 'RabotniKuma/Fast-Math-Qwen3-14B'
vllm_engine = LLM(
    model=model_path,
    max_model_len=16000,
    gpu_memory_utilization=0.9,
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(model_path)
sampling_params = SamplingParams(
    temperature=1.0,
    top_p=0.90,
    min_p=0.05,
    max_tokens=8192,
    stop='</think>',  # For even faster inference, applying early stopping at the </think> tag and extracting the final boxed content is recommended.
)
messages = [
    {
        'role': 'user', 
        'content': (
            'Solve the problem, and put the answer in \boxed{{}}. '
            'Sarah is twice as old as her youngest brother. If the difference between their ages is 15 years. How old is her youngest brother?'
        )
    }
]
messages = tokenizer.apply_chat_template(
    conversation=messages,
    tokenize=False,
    add_generation_prompt=True
)
response = vllm_engine.generate(messages, sampling_params=sampling_params)