Omar SobhandClaude Sonnet 4.6
228137555f
G0 (Critical): Replace 45 unimplemented!() panics across three GPU backends - rtx-backend-cuda: sin/cos/tanh via PTX, relu/sigmoid/leaky_relu/elu via activation.rs, pow/clamp/gt_scalar via unary.rs, var/var_dim host-side, conv2d/max_pool2d/avg_pool2d CPU fallback in new ops/conv.rs; new PTX kernels in element_wise.cu - rtx-backend-rocm: all 15 ops via CPU round-trip (to_vec → compute → from_slice) - rtx-backend-sycl: all 15 ops via CPU round-trip (to_host → compute → from_data) G2 (High): Re-add rtx-distributed to workspace - Vendor 4 minimal RNCCL stub crates at crates/vendor/rnccl/* - Update rtx-distributed RNCCL path deps to point at stubs (../../../../RNCCL/* → ../../vendor/rnccl/*) - Remove rtx-distributed from workspace exclude list, add to members G5 (Medium): Re-enable rtx-tts (213 tests restored) - Fix 15 rtx-nn API drift issues: LayerNorm::new, Conv1d::from_config, Conv1dPadding::Zeros, Dropout::new(p, device), tensor methods (relu/tanh/sigmoid/cat/stack), squeeze(Some(n)), to_vec() turbofish removal, Tensor::randn with &[...] slices G8 (Low): Quantum stubs + multimodal forward bug - rtx-timeseries: remove dead quantum/neuromorphic TODO comment blocks (no module files exist) - rtx-multimodal/fusion/transformer.rs: wire TransformerBlock loop in forward() - rtx-multimodal/fusion/strategies.rs: wire bottleneck_layers loop in forward() - rtx-transformers/architectures/transformer_block.rs: add forward() method (pre-norm residuals; full attention+FFN pending when those sub-layers are wired) Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
89 lines
2.7 KiB
TOML
89 lines
2.7 KiB
TOML
[package]
|
|
name = "rtx-distributed"
|
|
version = "1.0.0"
|
|
edition = "2024"
|
|
rust-version = "1.92"
|
|
authors = ["RustyTorch Team"]
|
|
license = "MIT OR Apache-2.0"
|
|
repository = "https://github.com/rustytorch/rustytorch"
|
|
description = "Distributed training support for RustyTorch++"
|
|
|
|
[dependencies]
|
|
# Core dependencies (explicit versions - crate excluded from workspace)
|
|
anyhow = "1.0"
|
|
thiserror = "2"
|
|
tracing = "0.1"
|
|
tokio = { version = "1.0", features = ["full"] }
|
|
serde = { version = "1.0", features = ["derive"] }
|
|
bincode = "1.3"
|
|
serde_json = "1.0"
|
|
|
|
# HPC-AI integration (optional) - path dep removed: hpc-channels not in this workspace
|
|
# hpc-channels = { path = "../../../../hpc-channels", features = ["rkyv-codec"], optional = true }
|
|
|
|
# RNCCL - Pure Rust collective communications (optional)
|
|
rnccl-core = { path = "../../vendor/rnccl/rnccl-core", optional = true }
|
|
rnccl-collectives = { path = "../../vendor/rnccl/rnccl-collectives", default-features = false, features = ["metal", "transport"], optional = true }
|
|
rnccl-transport = { path = "../../vendor/rnccl/rnccl-transport", optional = true }
|
|
rnccl-bootstrap = { path = "../../vendor/rnccl/rnccl-bootstrap", optional = true }
|
|
|
|
# Local crates
|
|
rtx-runtime = { path = "../../core/rtx-runtime" }
|
|
rtx-tensor = { path = "../../core/rtx-tensor" }
|
|
|
|
# NCCL binding via cudarc - safer and more comprehensive (optional for non-CUDA builds)
|
|
cudarc = { version = "0.18.2", features = ["std", "driver", "runtime", "nvrtc", "cublas", "cublaslt", "nccl", "cudnn", "cusparse", "cusolver", "cufile", "curand", "cuda-13000", "f16"], optional = true }
|
|
|
|
# Network and process management
|
|
mpi = { version = "0.6", optional = true }
|
|
crossbeam = "0.8"
|
|
parking_lot = "0.12"
|
|
uuid = { version = "1.0", features = ["v4", "serde"] }
|
|
|
|
# Configuration
|
|
config = "0.14"
|
|
clap = { version = "4.0", features = ["derive"] }
|
|
|
|
# Async trait support
|
|
async-trait = "0.1"
|
|
|
|
# Additional dependencies for multi-node support
|
|
rand = "0.8"
|
|
futures = "0.3"
|
|
|
|
# Half-precision floating point
|
|
half = "2.3"
|
|
|
|
# System info
|
|
hostname = "0.4"
|
|
|
|
# Time and date handling
|
|
chrono = { version = "0.4", features = ["serde"] }
|
|
|
|
[dev-dependencies]
|
|
proptest = "1.4"
|
|
criterion = { version = "0.5", features = ["html_reports", "csv_output"] }
|
|
tempfile = "3.8"
|
|
tracing-subscriber = "0.3"
|
|
|
|
[features]
|
|
default = []
|
|
cuda = ["dep:cudarc"]
|
|
nccl = ["cuda", "cudarc/nccl"]
|
|
rccl = [] # RCCL for AMD ROCm GPUs - requires librccl.so
|
|
rnccl = ["dep:rnccl-core", "dep:rnccl-collectives", "dep:rnccl-transport", "dep:rnccl-bootstrap"] # Rust-native NCCL from rnccl
|
|
mpi = ["dep:mpi"]
|
|
# HPC-AI low-latency channels (optional) - removed: dep not available
|
|
# hpc-channels = ["dep:hpc-channels"]
|
|
|
|
[lib]
|
|
name = "rtx_distributed"
|
|
path = "src/lib.rs"
|
|
|
|
[[bin]]
|
|
name = "benchmark_backends"
|
|
path = "src/bin/benchmark_backends.rs"
|
|
|
|
[lints]
|
|
workspace = true
|