Skip to content

Releases: pmclSF/DeepCompress

v2.0.0

05 Feb 23:20
@pmclSF pmclSF
v2
90c11c5
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
v2.0.0 Latest
Latest

DeepCompress v2.0.0 is a major release featuring:

  • Advanced entropy modeling

  • Significantly improved compression ratios

  • 3–5× faster processing

  • 50–80% less memory usage

✨ New Features

Advanced Entropy Models

Choose from six entropy modeling strategies to balance compression quality and speed:

Model | Compression Improvement | Best For -- | -- | -- gaussian | Baseline | Fast prototyping hyperprior | 15–25% smaller | Production workloads channel | 25–35% smaller | Quality-focused applications context | 30–40% smaller | Maximum compression attention | Learns global patterns | Complex repeating structures hybrid | Combines all approaches | State-of-the-art quality

🧠 Mixed Precision Support

New PrecisionManager enables float16 training for modern GPUs:

from precision_config import PrecisionManager

PrecisionManager.configure("mixed_float16")

Benefits:

  • ~50% memory reduction

  • 1.5–2× speedup

🧩 New Modules

src/entropy_model.py        Patched Gaussian conditional with optimized scale quantization
src/entropy_parameters.py   Network for predicting entropy parameters
src/context_model.py        Autoregressive spatial context modeling
src/channel_context.py      Channel-wise context for parallel decoding
src/attention_context.py    Sparse and windowed 3D attention mechanisms
src/precision_config.py     Mixed precision training configuration
src/benchmarks.py           Performance benchmarking utilities
src/quick_benchmark.py      Synthetic compression benchmark
src/constants.py            Pre-computed mathematical constants

✅ Compatibility

  • TensorFlow: 2.11+

  • Python: 3.8+

  • Keras: 3.x compatible

  • Backward compatible with V1 models

📚 Documentation Improvements

  • Rewritten README for accessibility

  • Step-by-step guides

  • Real-world analogies

  • Architecture diagrams

  • Troubleshooting section

🔄 Upgrading from V1

V1 code continues to work unchanged.

V1 (still works)

model = DeepCompressModel(config)

V2 (new)

model = DeepCompressModelV2(
    config,
    entropy_model="hyperprior"
)
Assets 2
Loading