mirror of
https://github.com/storytold/effectcraft.git
synced 2026-11-01 14:08:38 -06:00
Kristian MyramandClaude Sonnet 5.5
5c7eb3c2ae
Strided batched matmul with bias, GELU/ReLU and residual in the epilogue, layer norm, softmax, window partition, pooling, patch embedding, RoPE and neck kernels (WGSL, all tested against CPU definitions). GpuEncoder keeps activations on the GPU and returns the 64x64 level and the decoder's skip features. On the RX 7900 GRE one frame encodes in 0.29 s (debug build) against 2.1 s on the CPU in release; errors 6.8e-6 vs PyTorch, 1.2e-5 vs the CPU path. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
25 lines
817 B
TOML
25 lines
817 B
TOML
[package]
|
|
name = "effectcraft-sam2"
|
|
description = "SAM 2.1 (Hiera) video object tracker for the Roto Brush: a Rust port of the Hugging Face transformers implementation, native only, weights downloaded by the user"
|
|
version.workspace = true
|
|
edition.workspace = true
|
|
license = "Apache-2.0"
|
|
rust-version.workspace = true
|
|
|
|
[lints]
|
|
workspace = true
|
|
|
|
[dependencies]
|
|
effectcraft-segment = { workspace = true }
|
|
rayon = { workspace = true }
|
|
log = { workspace = true }
|
|
|
|
# Native only (D1): candle is not built for wasm32, where this crate is empty.
|
|
[target.'cfg(not(target_arch = "wasm32"))'.dependencies]
|
|
candle-core = "=0.9.2"
|
|
candle-nn = "=0.9.2"
|
|
# The optional GPU path (wgpu compute, any vendor); the CPU path stays the reference.
|
|
wgpu = { workspace = true }
|
|
pollster = { workspace = true }
|
|
bytemuck = { workspace = true }
|