Published October 10, 2023
| Version v0.3.2
Software
Open
stanford-futuredata/megablocks: v0.3.2
- 1. MosaicML
- 2. Databricks
- 3. Stanford '22
- 4. NVIDIA
Description
What's Changed
- Support for bfloat16
- Optimizations for top_k > 1
- Support for fully-sharded data parallelism
- Support tensor model parallelism when expert_parallel_world_size > num_experts
- Optimizations for activation memory
- Support activation quantization (thanks @dblalock!)
- Optimizations for SM90 (Hopper)
- Lots of bug fixes, cleanup and small optimizations
- @vchiley made their first contribution in https://github.com/stanford-futuredata/megablocks/pull/9
- @deepakn94 made their first contribution in https://github.com/stanford-futuredata/megablocks/pull/16
- @b-chu made their first contribution in https://github.com/stanford-futuredata/megablocks/pull/19
Full Changelog: https://github.com/stanford-futuredata/megablocks/compare/v0.1...v0.3.2
Files
stanford-futuredata/megablocks-v0.3.2.zip
Files
(1.1 MB)
Name | Size | Download all |
---|---|---|
md5:c6f223a848edb0b4359c4e4ca31a6781
|
1.1 MB | Preview Download |
Additional details
Related works
- Is supplement to
- https://github.com/stanford-futuredata/megablocks/tree/v0.3.2 (URL)