--- library_name: transformers pipeline_tag: text-generation base_model: Qwen/Qwen3-30B-A3B-Base tags: - qwen3 - moe - dapo - grpo - bf16 - research --- # Qwen3-30B-A3B-DAPO-BF16-step-0620 This repository contains training checkpoint **step 620**, converted from a distributed training checkpoint into standard Hugging Face safetensors. It is part of the [Qwen3-30B-A3B W4A4-QAT vs BF16 Checkpoints](https://huggingface.co/collections/shawnzzzzz/qwen3-30b-a3b-w4a4-qat-vs-bf16-checkpoints-6a5349a8477c95ea76c4cb17) series. ## Checkpoint metadata - Architecture: `Qwen3MoeForCausalLM` - Model type: `qwen3_moe` - Base model: `Qwen/Qwen3-30B-A3B-Base` - Training trajectory: **BF16 baseline** - Public trajectory label: `BF16 baseline` - Source checkpoint: `global_step_620` - Tensor storage: **BF16** - Matched counterpart: [shawnzzzzz/Qwen3-30B-A3B-DAPO-FFN-W4A4-QAT-BF16Master-step-0620](https://huggingface.co/shawnzzzzz/Qwen3-30B-A3B-DAPO-FFN-W4A4-QAT-BF16Master-step-0620) - Matched source step: `620` - Absolute step difference: `0` ## Important quantization note This is the BF16 baseline trajectory. QAT was disabled, and the repository contains standard BF16 Hugging Face weights. ## Intended use These checkpoints are research artifacts for comparing approximately step-matched W4A4-QAT and BF16 training trajectories. They have not been evaluated here as general-purpose production models. ## Validation The export was checked for: - required Hugging Face model and tokenizer metadata; - readable safetensors headers and complete shard index; - exact index-to-shard key consistency; - BF16 tensor dtype throughout; - exact key and tensor-shape match against the native Qwen3-MoE architecture; - full source-artifact content comparison against the validated export; - Hugging Face path, byte-size, LFS SHA256, and metadata-download integrity. `SHA256SUMS` covers every published file in this repository except the checksum manifest itself. Internal execution provenance is intentionally omitted from this public release.