v1.1.0

Green Compress 1.1.0

Phase 1 export-gguf and Phase 2 foundation pack-model, with honest README scope.

Highlights

• export-gguf — preserves GGUF metadata and re-quants 2D weights to Q4_0 for llama.cpp (--verify optional)
• pack-model (experimental) — writes model.green/ with manifest.json, metadata.gguf, dense.gguf, checksums
• Honest README — compression and benchmarking today; full native runtime package planned

Quick start

git clone https://github.com/VeyrForge/GreenCompress.git && cd GreenCompress
make
bin/greencompress export-gguf --help
bin/greencompress pack-model --help

See README.md for methods and CLI reference.

License

Free to run and use; view source and submit suggested changes via GitHub. See LICENSE.

Full Changelog: https://github.com/VeyrForge/GreenCompress/commits/v1.1.0

Green Compress 1.0.0

Green Compress 1.0.0

First public release under the VeyrForge Source-Available License.

Highlights

• greencompress CLI — quantize, repair, benchmark, and infer compressed transformer layers
• green_optimal default — ~45% less RAM than FP32 at ~99.9% quality on real model layers
• AVX2 SIMD CPU matmul with portable x86-64 fallback
• Optional CUDA backend for faster GEMM (make rust-gpu)
• GGUF pipeline — scripts/compress_model.py for batch layer compression
• Sub-8-bit green_q7 — −12% RAM vs Q8 with validated perplexity impact

Quick start

git clone https://github.com/VeyrForge/GreenCompress.git && cd GreenCompress
make
bin/greencompress help
bash benchmarks/run.sh

See README.md for methods, CLI reference, and benchmark tables.

License

Free to run and use; view source and submit suggested changes via GitHub. No fork, redistribution, or competing products without permission. See LICENSE.

Full Changelog: https://github.com/VeyrForge/GreenCompress/commits/v1.0.0