voxedge-qwen3-06b-int4-ort

VoxEdge Qwen3 0.6B INT4 for ONNX Runtime GenAI

This is the compact text-generation stage for the VoxEdge local voice-agent stack, an offering from Resontra AI. It is prepared for private, low-latency ASR → LLM → TTS execution on consumer CPUs.

The package contains an INT4 ONNX Runtime GenAI graph, external weights, tokenizer, chat template, and runtime configuration. The canonical file hashes, compatibility policy, evidence lineage, and immutable Hub revision are published separately in ResontraAI/voxedge-model-index after upload verification.

Intended use

• Short, local voice-agent responses on consumer computers.
• Reproducible VoxEdge Lab benchmarks and VoxBridge prototypes.
• CPU execution through ONNX Runtime GenAI.

This package is not a safety system, medical or legal advisor, or proof of output quality. Application owners must evaluate their prompts, languages, latency targets, and failure handling.

Loading

Use the immutable commit and exact allow-list from the VoxEdge model index. VoxBridge downloads into staging, verifies every byte size and SHA-256, then activates the directory atomically. Remote code, symlinks, mutable branch revisions, and in-place replacement are rejected.

Evidence and limitations

The local package has structural, real-model, and task evidence recorded by VoxEdge Lab. Current committed device measurements do not represent physical x86 laptops or NVIDIA/CUDA hardware. Model outputs may be inaccurate, verbose, biased, or unsuitable for a requested task.

No user recordings, speaker embeddings, personalized voices, secrets, generated review audio, or authorization records are included.

Lineage and license

Derived from Qwen3-0.6B and distributed under Apache-2.0. Consult the upstream model card and the VoxEdge model index for exact source revisions and artifact lineage.