You can deploy this model and start earning money today!

yashpratap/
Qwen3-Omni-30B-A3B-LoRA-test-r32

Name: Qwen3-Omni-30B-A3B-LoRA-test-r32
Availability: InStock

Qwen/Qwen3-Omni-30B-A3B-Instruct

adapter

Views

No views yet

Qwen3-Omni-30B-A3B LoRA Test Adapter (Random Weights)

This adapter contains RANDOM weights and is intended solely for testing LoRA loading/serving infrastructure in vLLM. It is NOT a trained model and will produce garbage outputs.

Details

Base model: Qwen/Qwen3-Omni-30B-A3B-Instruct
PEFT type: LoRA
Rank (r): 32
Alpha: 32
Target modules: thinker.model.*.(gate_proj|v_proj|o_proj|k_proj|up_proj|down_proj|q_proj)
Task type: CAUSAL_LM

Usage with vLLM

bash

vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct \
    --tensor-parallel-size 4 \
    --gpu-memory-utilization 0.85 \
    --enable-lora \
    --lora-modules test=yashpratap/Qwen3-Omni-30B-A3B-LoRA-test-r32

Purpose

Created for testing vLLM LoRA support for Qwen3OmniMoeThinkerForConditionalGeneration. All weight tensors have been replaced with torch.randn_like() to preserve tensor metadata (names, shapes, dtypes) without exposing any trained parameters.