Source model: Qwen/Qwen3-ASR-0.6B Source URL: https://huggingface.co/Qwen/Qwen3-ASR-0.6B Source artifact(s) and SHA-256: model.safetensors (upstream checkpoint, sharded) SHA-256: not recoverable from conversion records Source revision: not recoverable from conversion records Source license: Apache License 2.0 Tokenizer source (exported from a separate upstream repository) -------------------------------------------------------------- model: Qwen/Qwen3-0.6B URL: https://huggingface.co/Qwen/Qwen3-0.6B artifact: llm.tokenizer.gguf — produced by the RKNN export tool from the upstream tokenizer; the ASR repository itself is not the source of the LLM vocabulary. SHA-256: not recoverable from conversion records The files in this repository are RKNN-converted derivatives, not unmodified upstream weights. Artifact names and formats differ from upstream on purpose: the upstream checkpoint is a single safetensors file per stage and is split here into RKNN graphs plus their companion .weight files, with the token embedding table and the GPT-2 BPE vocabulary exported alongside them. The SHA-256 values above identify the **upstream** artifacts, not the files in this repository — use SHA256SUMS for the latter. Files added by this repository (not part of the upstream model) -------------------------------------------------------------- qwen3-asr.json wdnwebui worker manifest — the "manual" of this distribution unit. Deliberately plain text so it can be diffed, reviewed and previewed on the Hub. config.json Release metadata and file manifest. This is NOT the upstream architecture config.json. SHA256SUMS SHA-256 checksums of every model file in this repository. README.md, README_CN.md Documentation. .gitattributes Git LFS rules for this repository.