Runs On
What it takes to serve EventLens-4B (4.8B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
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.
EventLens-4B on every accelerator the SAVRN Index prices, at every precision →
Model Card
EventLens-4B is a video-language model developed through event-structure reinforcement learning and multi-teacher on-policy distillation (MT-OPD). Built on Qwen3-VL-4B-Instruct, it learns to recognize event boundaries, progression, and relations, alongside temporal grounding and video question answering. The model is part of EventLens: Event-Structure Reinforcement Learning for Video Understanding, developed by AIM3 Lab, Renmin University of China, and Kling AI. Selected results reported in the EventLens paper, compared with the corresponding pretrained backbone under the shared 256-frame protocol. QVHighlights, ActivityNet, and Charades-STA use mIoU. E.T. Bench and VinoGround use their…
Excerpt from the card by Guowen Zhang, licensed cc-by-nc-sa-4.0.
Configuration
- Architecture
- Qwen3VLForConditionalGeneration
- Context length (tokens)
- 262,144
- Layers
- 36
- Hidden size
- 2,560
- Feed-forward size
- 9,728
- Attention heads
- 32
- Key/value heads
- 8
- Head dimension
- 128
- Vocabulary size
- 151,936
- RoPE base
- 5,000,000
- Model type
- qwen3_vl
Identity and Version
- Repository
- Lostgreen/EventLens-4B
- Publisher
- Guowen Zhang
- Task
- Video text to text
- Modality
- Other
- Library
- transformers
- Parameters
- 4.8B parameters
- Languages
- Not stated by the source
- Revision
- 55de171038f3df61816d08c5cbc0009e1f609a40
- First published
- 2026-10-13
- Last updated
- 2026-10-13
Files and Weights
20 files, 9.7 GB in total. The weights are 2 files totalling 9.7 GB in safetensors.
Weights2 files · 9.7 GB
Configuration6 files · 72.9 KB
Tokenizer4 files · 11.5 MB
Documentation4 files · 19.0 KB
Other3 files · 333.4 KB
Repository1 file · 1.6 KB
Every file
License and Download
- License
- cc-by-nc-sa-4.0
- Access
- Open weights, no gate
- Download size
- 9.7 GB
Memory Requirements
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About EventLens-4B
How much GPU memory does EventLens-4B need?
About 11.6 GB at 16-bit and 2.9 GB at 4-bit: the weights (4.8B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run EventLens-4B 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.
Can I use EventLens-4B commercially?
Not without separate permission. EventLens-4B is released under Creative Commons Attribution-NonCommercial-ShareAlike 4.0. CC BY-NC-SA 4.0 permits non-commercial sharing and adapting with credit, and requires adaptations to use the same license. Commercial use needs separate permission.
What is EventLens-4B's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.
Similar Models
EventLens-8B is a video-language model developed through event-structure reinforcement learning and multi-teacher on-policy distillation (MT-OPD). Built on Qwen3-VL-8B-Instruct, it learns to recognize event boundaries, progression, and relations, alongside temporal grounding and video question answering. The model is part of EventLens: Event-Structure Reinforcement Learning for Video Understanding, developed by AIM3 Lab, Renmin University of China, and Kling AI. Selected results reported in the EventLens paper, compared with the corresponding pretrained backbone under the shared 256-frame protocol. QVHighlights, ActivityNet, and Charades-STA use mIoU. E.T. Bench and VinoGround use their…
Open weights
cc-by-nc-sa-4.0
8.8B parameters
262,144 tokens
transformers
C/C++ — https://github.com/ggml-org/llama.cpp
Open weights
apache-2.0
transformers
C/C++ — https://github.com/ggml-org/llama.cpp
Open weights
apache-2.0
transformers
GGUF quantization of OpenMOSS-Team/MOSS-VL-Realtime-NF4 (11B cross-attention VLM: Qwen3 text trunk with 12 pure cross-attn layers + SigLIP tower with deepstack), repackaged for the SkyVision.cpp inference engine. - NF4 weights are dequantized bit-exactly against bitsandbytes CUDA first (checkmossvlnf4.py); - the fused cross-KV projection weight is built at load time (engine-side); 12 GB GPU deployments: --n-ctx 4096 (and MOSSXCTX=6144 for long videos). Full chat, multi-turn sessions, realtime streaming (proactive narration with ) and a web UI are available via --serve. Follows the base model's license (see the OpenMOSS-Team/MOSS-VL-Realtime-NF4 page).
Open weights
apache-2.0
SkyVision.cpp
