[Feature]: Add support for Qwen3 0.6B (Qwen3-0.6B-Instruct)

[gauravstead]

Component

Other

Problem Description

Feature Request: Support for Qwen3 0.6B

Description

Qwen3 series has been released (from Alibaba) and the 0.6B variant is an extremely compelling small model:

• Only ~0.6 billion parameters → very low memory footprint (~400–700 MB quantized)
• Strong performance on many benchmarks (often beats or matches older 1.5B–3B models)
• Excellent multilingual support (including Indic languages, which is great for users in India)
• Apache 2.0 license

It would be a fantastic addition to RunAnywhere's on-device lineup, sitting nicely between SmolLM2-360M / Qwen2.5-0.5B (faster but weaker) and Qwen2.5-1.5B / 3B (stronger but heavier).

Many users are already looking for the smallest possible Qwen3 variant that still gives good instruction-following and reasoning.

Why this model?

• Memory + speed sweet spot for mid-range Android/iOS devices
• Noticeably better coherence and instruction following than Qwen2.5-0.5B in many early user reports
• Would give developers more granularity in model choices (0.36B → 0.6B → 1B → 1.5B → 3B+)
• Complements the existing Qwen2.5 support nicely

Proposed Solution

Proposed implementation

1. Add GGUF-quantized variants of Qwen3-0.6B-Instruct to the model registry / recommended list
   • Most useful quants: Q4_K_M, Q5_K_M, Q6_K (maybe Q8_0 for quality testing)
2. Add the correct chat template / tokenizer config if it differs from Qwen2.5
3. Update example code / model picker UI/docs with something like:

   await TextGeneration.loadModel(
     '/models/qwen3-0.6b-instruct-q5_K_M.gguf',
     'qwen3-0.6b-instruct'
   );

Alternatives Considered

No response

Additional Context

Hugging Face: https://huggingface.co/Qwen/Qwen3-0.6B-Instruct (or search "Qwen3 0.6B GGUF" for community quants)
Official repo: https://github.com/QwenLM/Qwen3 (announcement + technical report)

[Suvam-paul145]

@mikeger @polarbeargo assign me this issues

[shubhammalhotra28]

Hey @gauravstead
Given this is GGML/GGUF format - it should already be supported. https://huggingface.co/ggml-org/Qwen3-0.6B-GGUF/tree/main
Have u tried adding this or just opened the issue since our sample apps are not having this model ?

But will def add this model in our sample apps.

[gauravstead]

Earlier I messed up somehow and it did not work. Anyway it works now , no issues.
Thank you.