Downloads · 30 days
8
9% of all-time downloads
orc-unify-hybrid-v0.11-4bit
amityrobotics/orc-unify-hybrid-v0.11-4bit
orc-unify-hybrid-v0.11-4bit is a machine learning model from amityrobotics. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as other.
4-bit (compressed-tensors W4A16, group 128) quantization of orc-unify-hybrid-v0.11 — the unified ASR + tool-call + answer-generation model (DPO variant) for the ARC edge omni-hybrid pipeline. Directly loadable by vLLM…
All-time downloads
90
Public
Parameters
35.5B
51.8 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors26.2 GB · 100%
Parameter types
How the weights are stored.
I3229.9B · 84%
Base models
Model card
From the Hugging Face model README
orc-unify-hybrid-v0.11-4bit
4-bit (compressed-tensors W4A16, group 128) quantization of orc-unify-hybrid-v0.11 — the unified ASR + tool-call + answer-generation model (DPO variant) for the ARC edge omni-hybrid pipeline. Directly loadable by vLLM-Omni.
- Quantized: thinker MoE + attention linears (W4A16 g128, RTN).
- Kept bf16 (per vLLM-omni requirements):
talker.*,code2wav.*,thinker.audio_tower.*,visual.*, all*.mlp.gate, andlm_head. - ~24 GB (vs 66 GB bf16). See the parent repo for the full model card, training recipe, and evaluation (AGN 93.3% correct / 0.0% hallucinate, ASR parity with arc-orc-v08).
Load in vLLM-Omni the same way as cyankiwi/Qwen3-Omni-30B-A3B-Instruct-AWQ-4bit
(the base the pipeline already runs), with the GB10 single-GPU deploy config.
- Type
- qwen3_omni_moe
- License
- other
- Created
- Jul 5, 2026
- Updated
- Jul 5, 2026