huggingface / huggingface/ratchet
CPU Backend
- Dominant language
- Rust
- Stars
- 771
- Forks
- 44
- Avg merge
- 40m
- Merged PRs (30d)
- 1
Description
The Ratchet CPU backend is nearly here!
Some remaining work to be done:
1. We have a "conflicting" execution model now between CPU and GPU. GPU is applied lazily (i.e we construct a compute graph of all operations and then dispatch to the GPU, for CPU we apply as we build the graph).
2. Many operations not yet implemented.
3. Matmul optimizations required for performance to be acceptable. Candle did well here with some `wasm128` magic.
Contributor guide
Research direction
Start by locating the Ratchet CPU backend and comparing its execution model with the GPU backend. Then inventory the unimplemented operations and investigate the matmul path, including the mentioned wasm128 optimization. Done means the execution model is resolved, required operations are implemented, and matmul performance is acceptable.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust, wasm
- Domain
- machine-learning, performance
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100