(moss_config: dict)
| 63 | |
| 64 | |
| 65 | def build_qwen3_config(moss_config: dict) -> dict: |
| 66 | lang = dict(moss_config["language_config"]) |
| 67 | lang["architectures"] = ["Qwen3ForCausalLM"] |
| 68 | lang["model_type"] = "qwen3" |
| 69 | lang.pop("_name_or_path", None) |
| 70 | lang.setdefault("torch_dtype", "bfloat16") |
| 71 | lang.setdefault("transformers_version", moss_config.get("transformers_version", "4.57.1")) |
| 72 | return lang |
| 73 | |
| 74 | |
| 75 | MAX_SHARD_SIZE = 5 * 1024**3 |