alphadl/R1-Distill-0.6B-Qwen

Model Card for R1-Distill-0.6B

This model is a fine-tuned version of Qwen/Qwen3-0.6B-Base on the open-r1/Mixture-of-Thoughts dataset. It has been trained using TRL.

Quick start

from transformers import pipeline

question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="alphadl/R1-Distill-0.6B", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Training procedure

This model was trained with SFT.

Framework versions

• TRL: 0.18.0
• Transformers: 4.52.3
• Pytorch: 2.6.0
• Datasets: 3.2.0
• Tokenizers: 0.21.1

Evaluation

• lighteval: 0.10.0

Model	Qwen3-0.6B-Base	R1-Distill-0.6B (Ours)
Math-500	38.2	41.0(+2.8)
GPQA Diamond	24.2	28.3(+4.1)

export VLLM_WORKER_MULTIPROC_METHOD=spawn # Required for vLLM
export NUMEXPR_MAX_THREADS=128 # Utilize all 128 cores for numerical computations

MODEL=data/R1-Distill-0.6B

# Evaluate the base model
# MODEL=Qwen/Qwen3-0.6B-Base

MODEL_ARGS="model_name=$MODEL,dtype=bfloat16,max_model_length=32768,gpu_memory_utilization=0.8,generation_parameters={max_new_tokens:8192,temperature:0.6,top_p:0.95}"
OUTPUT_DIR=data/evals/$MODEL

# Math 500  
TASK=math_500
lighteval vllm $MODEL_ARGS "lighteval|$TASK|0|0" \
    --use-chat-template \
    --output-dir $OUTPUT_DIR

# GPQA Diamond
TASK=gpqa:diamond
lighteval vllm $MODEL_ARGS "lighteval|$TASK|0|0" \
    --use-chat-template \
    --output-dir $OUTPUT_DIR