Mungert/SkyCaptioner-V1-GGUF

SkyCaptioner-V1 GGUF Models

Model Generation Details

This model was generated using llama.cpp at commit 5787b5da.

Quantization beyond the IMatrix

Testing a new quantization method using rules to bump important layers above what the standard imatrix would use.

I have found that the standard IMatrix does not perform very well at low bit quantiztion and for MOE models. So I am using llama.cpp --tensor-type to bump up selected layers. See Layer bumping with llama.cpp

This does create larger model files but increases precision for a given model size.

Please provide feedback on how you find this method performs

Choosing the Right Model Format

Selecting the correct model format depends on your hardware capabilities and memory constraints.

BF16 (Brain Float 16) – Use if BF16 acceleration is available

• A 16-bit floating-point format designed for faster computation while retaining good precision.
• Provides similar dynamic range as FP32 but with lower memory usage.
• Recommended if your hardware supports BF16 acceleration (check your device's specs).
• Ideal for high-performance inference with reduced memory footprint compared to FP32.

📌 Use BF16 if:
✔ Your hardware has native BF16 support (e.g., newer GPUs, TPUs).
✔ You want higher precision while saving memory.
✔ You plan to requantize the model into another format.

📌 Avoid BF16 if:
❌ Your hardware does not support BF16 (it may fall back to FP32 and run slower).
❌ You need compatibility with older devices that lack BF16 optimization.

---

F16 (Float 16) – More widely supported than BF16

• A 16-bit floating-point high precision but with less of range of values than BF16.
• Works on most devices with FP16 acceleration support (including many GPUs and some CPUs).
• Slightly lower numerical precision than BF16 but generally sufficient for inference.

📌 Use F16 if:
✔ Your hardware supports FP16 but not BF16.
✔ You need a balance between speed, memory usage, and accuracy.
✔ You are running on a GPU or another device optimized for FP16 computations.

📌 Avoid F16 if:
❌ Your device lacks native FP16 support (it may run slower than expected).
❌ You have memory limitations.

---

Hybrid Precision Models (e.g., bf16_q8_0, f16_q4_K) – Best of Both Worlds

These formats selectively quantize non-essential layers while keeping key layers in full precision (e.g., attention and output layers).

• Named like bf16_q8_0 (meaning full-precision BF16 core layers + quantized Q8_0 other layers).
• Strike a balance between memory efficiency and accuracy, improving over fully quantized models without requiring the full memory of BF16/F16.

📌 Use Hybrid Models if:
✔ You need better accuracy than quant-only models but can’t afford full BF16/F16 everywhere.
✔ Your device supports mixed-precision inference.
✔ You want to optimize trade-offs for production-grade models on constrained hardware.

📌 Avoid Hybrid Models if:
❌ Your target device doesn’t support mixed or full-precision acceleration.
❌ You are operating under ultra-strict memory limits (in which case use fully quantized formats).

---

Quantized Models (Q4_K, Q6_K, Q8, etc.) – For CPU & Low-VRAM Inference

Quantization reduces model size and memory usage while maintaining as much accuracy as possible.

• Lower-bit models (Q4_K) → Best for minimal memory usage, may have lower precision.
• Higher-bit models (Q6_K, Q8_0) → Better accuracy, requires more memory.

📌 Use Quantized Models if:
✔ You are running inference on a CPU and need an optimized model.
✔ Your device has low VRAM and cannot load full-precision models.
✔ You want to reduce memory footprint while keeping reasonable accuracy.

📌 Avoid Quantized Models if:
❌ You need maximum accuracy (full-precision models are better for this).
❌ Your hardware has enough VRAM for higher-precision formats (BF16/F16).

---

Very Low-Bit Quantization (IQ3_XS, IQ3_S, IQ3_M, Q4_K, Q4_0)

These models are optimized for very high memory efficiency, making them ideal for low-power devices or large-scale deployments where memory is a critical constraint.

• IQ3_XS: Ultra-low-bit quantization (3-bit) with very high memory efficiency.

  • Use case: Best for ultra-low-memory devices where even Q4_K is too large.
  • Trade-off: Lower accuracy compared to higher-bit quantizations.

• IQ3_S: Small block size for maximum memory efficiency.

  • Use case: Best for low-memory devices where IQ3_XS is too aggressive.

• IQ3_M: Medium block size for better accuracy than IQ3_S.

  • Use case: Suitable for low-memory devices where IQ3_S is too limiting.

• Q4_K: 4-bit quantization with block-wise optimization for better accuracy.

  • Use case: Best for low-memory devices where Q6_K is too large.

• Q4_0: Pure 4-bit quantization, optimized for ARM devices.

  • Use case: Best for ARM-based devices or low-memory environments.

Ultra Low-Bit Quantization (IQ1_S IQ1_M IQ2_S IQ2_M IQ2_XS IQ2_XSS)

• *Ultra-low-bit quantization (1 2-bit) with extreme memory efficiency.
  • Use case: Best for cases were you have to fit the model into very constrained memory
  • Trade-off: Very Low Accuracy. May not function as expected. Please test fully before using.

---

Summary Table: Model Format Selection

Model Format	Precision	Memory Usage	Device Requirements	Best Use Case
BF16	Very High	High	BF16-supported GPU/CPU	High-speed inference with reduced memory
F16	High	High	FP16-supported GPU/CPU	Inference when BF16 isn’t available
Q4_K	Medium-Low	Low	CPU or Low-VRAM devices	Memory-constrained inference
Q6_K	Medium	Moderate	CPU with more memory	Better accuracy with quantization
Q8_0	High	Moderate	GPU/CPU with moderate VRAM	Highest accuracy among quantized models
IQ3_XS	Low	Very Low	Ultra-low-memory devices	Max memory efficiency, low accuracy
IQ3_S	Low	Very Low	Low-memory devices	Slightly more usable than IQ3_XS
IQ3_M	Low-Medium	Low	Low-memory devices	Better accuracy than IQ3_S
Q4_0	Low	Low	ARM-based/embedded devices	Llama.cpp automatically optimizes for ARM inference
Ultra Low-Bit (IQ1/2_*)	Very Low	Extremely Low	Tiny edge/embedded devices	Fit models in extremely tight memory; low accuracy
Hybrid (e.g., bf16_q8_0)	Medium–High	Medium	Mixed-precision capable hardware	Balanced performance and memory, near-FP accuracy in critical layers

---

SkyCaptioner-V1: A Structural Video Captioning Model

📑 Technical Report · 👋 Playground · 💬 Discord · 🤗 Hugging Face · 🤖 ModelScope · 🌐 GitHub

---

Welcome to the SkyCaptioner-V1 repository! Here, you'll find the structural video captioning model weights and inference code for our video captioner that labels the video data efficiently and comprehensively.

🔥🔥🔥 News!!

• Apr 21, 2025: 👋 We release the vllm batch inference code for SkyCaptioner-V1 Model and caption fusion inference code.
• Apr 21, 2025: 👋 We release the first shot-aware video captioning model SkyCaptioner-V1 Model. For more details, please check our paper.

📑 TODO List

• SkyCaptioner-V1

  • Checkpoints
  • Batch Inference Code
  • Caption Fusion Method
  • Web Demo (Gradio)

🌟 Overview

SkyCaptioner-V1 is a structural video captioning model designed to generate high-quality, structural descriptions for video data. It integrates specialized sub-expert models and multimodal large language models (MLLMs) with human annotations to address the limitations of general captioners in capturing professional film-related details. Key aspects include:

1. ​​Structural Representation​: Combines general video descriptions (from MLLMs) with sub-expert captioner (e.g., shot types,shot angles, shot positions, camera motions.) and human annotations.
2. ​​Knowledge Distillation​: Distills expertise from sub-expert captioners into a unified model.
3. ​​Application Flexibility​: Generates dense captions for text-to-video (T2V) and concise prompts for image-to-video (I2V) tasks.

🔑 Key Features

Structural Captioning Framework

Our Video Captioning model captures multi-dimensional details:

• ​​Subjects​: Appearance, action, expression, position, and hierarchical categorization.
• ​​Shot Metadata​: Shot type (e.g., close-up, long shot), shot angle, shot position, camera motion, environment, lighting, etc.

Sub-Expert Integration

• ​​Shot Captioner​: Classifies shot type, angle, and position with high precision.
• ​​Expression Captioner​: Analyzes facial expressions, emotion intensity, and temporal dynamics.
• ​​Camera Motion Captioner​: Tracks 6DoF camera movements and composite motion types,

Training Pipeline

• Trained on ~2M high-quality, concept-balanced videos curated from 10M raw samples.
• Fine-tuned on Qwen2.5-VL-7B-Instruct with a global batch size of 512 across 32 A800 GPUs.
• Optimized using AdamW (learning rate: 1e-5) for 2 epochs.

Dynamic Caption Fusion:

• Adapts output length based on application (T2V/I2V).
• Employs LLM Model to fusion structural fields to get a natural and fluency caption for downstream tasks.

📊 Benchmark Results

SkyCaptioner-V1 demonstrates significant improvements over existing models in key film-specific captioning tasks, particularly in ​shot-language understanding and ​​domain-specific precision​. The differences stem from its structural architecture and expert-guided training:

1. ​​Superior Shot-Language Understanding​:
   • ​Our Captioner model outperforms Qwen2.5-VL-72B with +11.2% in shot type, +16.1% in shot angle, and +50.4% in shot position accuracy. Because SkyCaptioner-V1’s specialized shot classifiers outperform generalist MLLMs, which lack film-domain fine-tuning.
   • ​+28.5% accuracy in camera motion vs. Tarsier2-recap-7B (88.8% vs. 41.5%): Its 6DoF motion analysis and active learning pipeline address ambiguities in composite motions (e.g., tracking + panning) that challenge generic captioners.
2. ​​High domain-specific precision​:
   • ​​Expression accuracy​: ​68.8% vs. 54.3% (Tarsier2-recap-7B), leveraging temporal-aware S2D frameworks to capture dynamic facial changes.

Metric	Qwen2.5-VL-7B-Ins.	Qwen2.5-VL-72B-Ins.	Tarsier2-recap-7B	SkyCaptioner-V1
Avg accuracy	51.4%	58.7%	49.4%	76.3%
shot type	76.8%	82.5%	60.2%	93.7%
shot angle	60.0%	73.7%	52.4%	89.8%
shot position	28.4%	32.7%	23.6%	83.1%
camera motion	62.0%	61.2%	45.3%	85.3%
expression	43.6%	51.5%	54.3%	68.8%
TYPES_type	43.5%	49.7%	47.6%	82.5%
TYPES_sub_type	38.9%	44.9%	45.9%	75.4%
appearance	40.9%	52.0%	45.6%	59.3%
action	32.4%	52.0%	69.8%	68.8%
position	35.4%	48.6%	45.5%	57.5%
is_main_subject	58.5%	68.7%	69.7%	80.9%
environment	70.4%	72.7%	61.4%	70.5%
lighting	77.1%	80.0%	21.2%	76.5%

📦 Model Downloads

Our SkyCaptioner-V1 model can be downloaded from SkyCaptioner-V1 Model. We use Qwen2.5-32B-Instruct as our caption fusion model to intelligently combine structured caption fields, producing either dense or sparse final captions depending on application requirements.

# download SkyCaptioner-V1
huggingface-cli download Skywork/SkyCaptioner-V1 --local-dir /path/to/your_local_model_path
# download Qwen2.5-32B-Instruct
huggingface-cli download Qwen/Qwen2.5-32B-Instruct --local-dir /path/to/your_local_model_path2

🛠️ Running Guide

Begin by cloning the repository:

git clone https://github.com/SkyworkAI/SkyReels-V2
cd skycaptioner_v1

Installation Guide for Linux

We recommend Python 3.10 and CUDA version 12.2 for the manual installation.

pip install -r requirements.txt

Running Command

Get Structural Caption by SkyCaptioner-V1

export SkyCaptioner_V1_Model_PATH="/path/to/your_local_model_path"

python scripts/vllm_struct_caption.py \
    --model_path ${SkyCaptioner_V1_Model_PATH} \
    --input_csv "./examples/test.csv" \
    --out_csv "./examepls/test_result.csv" \
    --tp 1 \
    --bs 4

T2V/I2V Caption Fusion by Qwen2.5-32B-Instruct Model

export LLM_MODEL_PATH="/path/to/your_local_model_path2"

python scripts/vllm_fusion_caption.py \
    --model_path ${LLM_MODEL_PATH} \
    --input_csv "./examples/test_result.csv" \
    --out_csv "./examples/test_result_caption.csv" \
    --bs 4 \
    --tp 1 \
    --task t2v

Note:

• If you want to get i2v caption, just change the --task t2v to --task i2v in your Command.

Acknowledgements

We would like to thank the contributors of Qwen2.5-VL, tarsier2 and vllm repositories, for their open research and contributions.

Citation

@misc{chen2025skyreelsv2infinitelengthfilmgenerative,
author = {Guibin Chen and Dixuan Lin and Jiangping Yang and Chunze Lin and Juncheng Zhu and Mingyuan Fan and Hao Zhang and Sheng Chen and Zheng Chen and Chengchen Ma and Weiming Xiong and Wei Wang and Nuo Pang and Kang Kang and Zhiheng Xu and Yuzhe Jin and Yupeng Liang and Yubing Song and Peng Zhao and Boyuan Xu and Di Qiu and Debang Li and Zhengcong Fei and Yang Li and Yahui Zhou},
title = {Skyreels V2:Infinite-Length Film Generative Model},
year = {2025},
eprint={2504.13074},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2504.13074}
}

🚀 If you find these models useful

Help me test my AI-Powered Quantum Network Monitor Assistant with quantum-ready security checks:

👉 Quantum Network Monitor

The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : Source Code Quantum Network Monitor. You will also find the code I use to quantize the models if you want to do it yourself GGUFModelBuilder

💬 How to test:
Choose an AI assistant type:

• TurboLLM (GPT-4.1-mini)
• HugLLM (Hugginface Open-source models)
• TestLLM (Experimental CPU-only)

What I’m Testing

I’m pushing the limits of small open-source models for AI network monitoring, specifically:

• Function calling against live network services
• How small can a model go while still handling:
  • Automated Nmap security scans
  • Quantum-readiness checks
  • Network Monitoring tasks

🟡 TestLLM – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):

• ✅ Zero-configuration setup
• ⏳ 30s load time (slow inference but no API costs) . No token limited as the cost is low.
• 🔧 Help wanted! If you’re into edge-device AI, let’s collaborate!

Other Assistants

🟢 TurboLLM – Uses gpt-4.1-mini :

• **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
• Create custom cmd processors to run .net code on Quantum Network Monitor Agents
• Real-time network diagnostics and monitoring
• Security Audits
• Penetration testing (Nmap/Metasploit)

🔵 HugLLM – Latest Open-source models:

• 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.

💡 Example commands you could test:

1. "Give me info on my websites SSL certificate"
2. "Check if my server is using quantum safe encyption for communication"
3. "Run a comprehensive security audit on my server"
4. '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code from. This is a very flexible and powerful feature. Use with caution!

Final Word

I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is open source. Feel free to use whatever you find helpful.

If you appreciate the work, please consider buying me a coffee ☕. Your support helps cover service costs and allows me to raise token limits for everyone.

I'm also open to job opportunities or sponsorship.

Thank you! 😊