Decision Workspace
kaio-candle vs atomr-accel vs baracuda-cutlass
Side-by-side comparison of Rust crates
Candle bridge for KAIO — CustomOp bindings for 12 GPU ops (matmul_tc, matmul_tc_bf16, matmul_tc_async, matmul_tc_bf16_async, matmul_int4, matmul_int8, attention_tc, attention_tc_causal, attention_flash, attention_flash_causal, qkv_project_int8, qkv_project_int4). All four matmul TC variants (f16+bf16, sync+async) support backward (autograd); FlashAttention (plain+causal) supports backward via dedicated PTX kernels. Build with `cargo build --features cuda`.
Backend-agnostic compute-acceleration core. Defines the AccelBackend trait, AccelRef<T> typed pointers, AccelError enum, and CompletionStrategy — the abstraction layer that lets atomr-accel-cuda (NVIDIA), and future ROCm / Metal / oneAPI / Vulkan backends plug into the same actor surface.
Safe Rust wrapper for compiled CUTLASS kernels: plan-based GEMM and grouped GEMM with caller-supplied workspace, typed device-buffer arguments, and capture-safe launch.
Core Metrics
| kaio-candle | atomr-accel | baracuda-cutlass | |
|---|---|---|---|
| Health Score | 48 | 59 | 59 |
| Total Downloads | 65 | 2.5K | 3.1K |
| 30d Downloads | 0 | 0 | 0 |
| Dependents | 0 | 33 | 55 |
| Releases | 3 | 4 | 58 |
| Last Updated | 119d ago | 152d ago | 3d ago |
| Age | 5m | 5m | 5m |
Health Breakdown
Technical Details
| kaio-candle | atomr-accel | baracuda-cutlass | |
|---|---|---|---|
| Version | 0.2.0 | 0.10.0 | 0.0.1-alpha.86 |
| Stable (≥1.0) | ✗ No | ✗ No | ✗ No |
| License | MIT OR Apache-2.0 | Apache-2.0 | MIT OR Apache-2.0 |
| Dependencies | 7 | 6 | 10 |
| Crate Size | 42KB | 28KB | 78KB |
| Features | 2 | 4 | 5 |
| Yanked % | 0.0% | 0.0% | 0.0% |
| Edition | 2024 | 2021 | 2024 |
| MSRV | 1.94 | 1.78 | — |
| Owners | 1 | 1 | 1 |
Links
Quick Verdict
- •atomr-accel leads with a health score of 59/100, but none of the options score above 80.
- •baracuda-cutlass has the most downloads (3.1K), suggesting wider adoption.