Qwen3-1.7B-LOMO-q4f16_1-MLC2

MLC-LLM formatted weights for on-device inference (Android / Vulkan, CPU, etc.).

• Base model: Qwen/Qwen3-1.7B
• Quantization: q4f16_1 (group size 32, int4 weights + f16 scales)
• Conversation template: chatml ← match runtime prompt formatting with training
• Files
  • mlc-chat-config.json
  • params_shard_*.bin
  • tensor-cache.json
  • tokenizer files (tokenizer.json + vocab.json + merges.txt)

Quick test (CLI)

mlc_llm chat HF://raining-codes/Qwen3-1.7B-LOMO-q4f16_1-MLC2   --temperature 0.7 --top-p 0.9 --max-gen-len 256

Notes

• This repo contains MLC execution format. It is not directly loadable with HF Transformers.
• In apps (e.g., MLCChat for Android), use "Add remote model" with raining-codes/Qwen3-1.7B-LOMO-q4f16_1-MLC2.