Profile-guided layout optimization for GGUF MoE models (BOLT/PGO for LLM weights)
Project description
mbolt
Profile-guided layout optimization for GGUF MoE models — BOLT/PGO, but for weights.
Trace which experts your workload co-activates, rewrite the GGUF so those bytes sit together on disk, decode with explicit-read prefetch: 1.55–1.63× faster memory-squeezed MoE streaming in llama.cpp, 2.23× at the storage level. Weights byte-exact, +0.003% size.
pip install mbolt
mbolt-sim cluster route.bin -o perms.json # co-activation -> permutations
mbolt model.gguf perms.json -o model.opt.gguf --layout interleave
mbolt-sim gate model.gguf route.bin perms.json -o gate.json # predict before rewriting
Trace capture and the explicit-read prefetcher need a small llama.cpp patch
(MBOLT_TRACE / MBOLT_PREFETCH env vars) shipped in the repo.
Full docs, benchmark reports, and the llama.cpp patch: https://github.com/doramirdor/mbolt
MIT license.
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file mbolt-0.1.0.tar.gz.
File metadata
- Download URL: mbolt-0.1.0.tar.gz
- Upload date:
- Size: 40.0 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: uv/0.11.20 {"installer":{"name":"uv","version":"0.11.20","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
9558d278444f1b2dabef3727b95f4a5ce9281b33bb9cde180ed75c507672542d
|
|
| MD5 |
d4579c1b2ead65c87303931f0e1f3a77
|
|
| BLAKE2b-256 |
656ad33bd3166bbeb154d2278f06d69683152988515018b7a4843e311540a82d
|
File details
Details for the file mbolt-0.1.0-py3-none-any.whl.
File metadata
- Download URL: mbolt-0.1.0-py3-none-any.whl
- Upload date:
- Size: 20.1 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: uv/0.11.20 {"installer":{"name":"uv","version":"0.11.20","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
4f3466146e98ebf0793cd379912a0796d02a63b77cb3dacb8b5d754db6477be7
|
|
| MD5 |
e179831328a3db398d6db867fdf1cc97
|
|
| BLAKE2b-256 |
70df76aa073f3bf7ac853a8a87ab5142ba4eb48a5bfadff9e31fa162cfbd2e27
|
