Qwen3-30B-A3B-GRPO-LowPrecision-Control-Step410

Qwen3-30B-A3B GRPO Low-Precision Control — Step 410

This is a research checkpoint derived from Qwen/Qwen3-30B-A3B-Base with group-relative policy optimization on mathematical reasoning data. It is the retained step-410 checkpoint from the low-precision control run.

The checkpoint is exported as standard BF16 Hugging Face safetensors; it does not require custom low-precision inference kernels.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "shawnzzzzz/Qwen3-30B-A3B-GRPO-LowPrecision-Control-Step410"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

Use transformers>=4.51.0 for Qwen3-MoE support.

Notes

This run ended at step 410, so it is not step-matched to the two step-800 checkpoints in the accompanying comparison. This is an intermediate research checkpoint, not a production release. Users should independently evaluate correctness, safety, and suitability for their deployment setting. The base-model license applies.