AVC-Reward is an open-weight model from Yuruo Wu. It has 31.7B parameters. At 16-bit it needs about 76.1 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index. It draws 8 downloads a month.
Event-Centric Audio-Video Consistency Evaluation · Full reward-model checkpoint. From the GitHub repository, after setting up the environment
Runs On
What it takes to serve AVC-Reward (31.7B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 63.4 GB | 76.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 31.7 GB | 38.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 15.9 GB | 19.0 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 18, 2026.
AVC-Reward on every accelerator the SAVRN Index prices, at every precision
Model Card
Event-Centric Audio-Video Consistency Evaluation · Full reward-model checkpoint. From the GitHub repository, after setting up the environment
Excerpt from the card by Yuruo Wu.
Configuration
- Architecture
- Qwen3OmniMoeForConditionalGeneration
- Hidden size
- 1,024
- Vocabulary size
- 152,064
- Model type
- qwen3_omni_moe
Identity and Version
- Repository
- Tsuki0512/AVC-Reward
- Publisher
- Yuruo Wu
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- 31.7B parameters
- Languages
- Not stated by the source
- Revision
- 816ff0c7adf825656c7701fe99c4f687749ac621
- First published
- 2026-10-13
- Last updated
- 2026-10-16
Files and Weights
26 files, 63.5 GB in total. The weights are 13 files totalling 63.4 GB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model-00001-of-00013.safetensors | Weights | 5.0 GB | 8f446c290bc6 |
| model-00002-of-00013.safetensors | Weights | 5.0 GB | a6b7516fb48f |
| model-00003-of-00013.safetensors | Weights | 5.0 GB | 62158aaf8802 |
| model-00004-of-00013.safetensors | Weights | 5.0 GB | 1bb07850b82a |
| model-00005-of-00013.safetensors | Weights | 5.0 GB | a7aa0e4662e6 |
| model-00006-of-00013.safetensors | Weights | 5.0 GB | d89933a07091 |
| model-00007-of-00013.safetensors | Weights | 5.0 GB | a0cac7d5a74c |
| model-00008-of-00013.safetensors | Weights | 5.0 GB | d2bb4864bb71 |
| model-00009-of-00013.safetensors | Weights | 5.0 GB | 9e765757bdd9 |
| model-00010-of-00013.safetensors | Weights | 5.0 GB | 57d295006bba |
| model-00011-of-00013.safetensors | Weights | 5.0 GB | 45f0b304f6b4 |
| model-00012-of-00013.safetensors | Weights | 5.0 GB | a8f5d9ff3457 |
| model-00013-of-00013.safetensors | Weights | 3.5 GB | b840e744e578 |
| added_tokens.json | Configuration | 904 B | — |
| config.json | Configuration | 15.0 KB | — |
| model.safetensors.index.json | Configuration | 1.9 MB | — |
| preprocessor_config.json | Configuration | 603 B | — |
| special_tokens_map.json | Configuration | 1.0 KB | — |
| video_preprocessor_config.json | Configuration | 1.2 KB | — |
| README.md | Documentation | 1.3 KB | — |
| chat_template.jinja | Other | 5.7 KB | — |
| .gitattributes | Repository | 1.6 KB | — |
| merges.txt | Tokenizer | 1.7 MB | — |
| tokenizer.json | Tokenizer | 11.4 MB | 09267689b836 |
| tokenizer_config.json | Tokenizer | 7.4 KB | — |
| vocab.json | Tokenizer | 2.8 MB | — |
License and Download
- License
- Not stated by the source
- Access
- Open weights, no gate
- Download size
- 63.4 GB
Released by Yuruo Wu through its official repository on Hugging Face.
Built From
- Derived from Qwen/Qwen3-Omni-30B-A3B-Thinking
- Described by arXiv:2610.22321
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 63.4 GB |
| 16-bit | 63.4 GB |
| 8-bit | 31.7 GB |
| 4-bit | 15.9 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About AVC-Reward
How much GPU memory does AVC-Reward need?
About 76.1 GB at 16-bit and 19 GB at 4-bit: the weights (31.7B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run AVC-Reward on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.