A high-performance inference engine for AI models
Misc cleanup (#646)
uuuvn721aedbcfcea
diff --git a/crates/backend-uzu/build/main.rs b/crates/backend-uzu/build/main.rsindex 51828c1e2..0f3456948 100644--- a/crates/backend-uzu/build/main.rs+++ b/crates/backend-uzu/build/main.rs@@ -55,7 +55,6 @@ async fn main() -> anyhow::Result<ExitCode> {let enum_paths = EnumPaths::from_gpu_types(&gpu_types).context("Failed to build enum path map")?;- #[allow(unused_mut)]let mut compilers: Vec<Box<dyn Compiler>> = Vec::new();if backend_cpu {diff --git a/crates/backend-uzu/src/backends/common/encoder.rs b/crates/backend-uzu/src/backends/common/encoder.rsindex 99bb99c76..1710f3fc3 100644--- a/crates/backend-uzu/src/backends/common/encoder.rs+++ b/crates/backend-uzu/src/backends/common/encoder.rs@@ -192,15 +192,14 @@ impl<B: Backend> Pending<B> {pub fn wait_until_completed(self) -> Result<Completed<B>, B::Error> {Ok(Completed {command_buffer: self.command_buffer.wait_until_completed()?,- allocation_pool: self.allocation_pool,+ _allocation_pool: self.allocation_pool,})}}pub struct Completed<B: Backend> {command_buffer: <B::CommandBuffer as CommandBuffer>::Completed,- #[allow(unused)]- allocation_pool: Arc<AllocationPool<B>>,+ _allocation_pool: Arc<AllocationPool<B>>,}impl<B: Backend> Completed<B> {diff --git a/crates/backend-uzu/src/backends/cpu/kernel/embedding/full_precision_embedding.rs b/crates/backend-uzu/src/backends/cpu/kernel/embedding/full_precision_embedding.rsindex 3245146fb..2920a3654 100644--- a/crates/backend-uzu/src/backends/cpu/kernel/embedding/full_precision_embedding.rs+++ b/crates/backend-uzu/src/backends/cpu/kernel/embedding/full_precision_embedding.rs@@ -7,13 +7,13 @@ use crate::array::ArrayElement;#[kernel(FullPrecisionEmbeddingLookup)]#[variants(T, f32, f16, bf16)]pub fn full_precision_embedding_lookup<T: ArrayElement + Float>(- #[allow(unused)] token_ids: *const u32,- #[allow(unused)] weights: *const T,- #[allow(unused)] output: *mut T,- #[allow(unused)] batch_size: u32,- #[allow(unused)] vocab_size: u32,- #[allow(unused)] model_dim: u32,- #[allow(unused)] input_scale: f32,+ token_ids: *const u32,+ weights: *const T,+ output: *mut T,+ batch_size: u32,+ vocab_size: u32,+ model_dim: u32,+ input_scale: f32,) {for batch_idx in 0..batch_size {let token_id = unsafe { *token_ids.add(batch_idx as usize) };diff --git a/crates/backend-uzu/src/backends/cpu/kernel/embedding/quant_embedding.rs b/crates/backend-uzu/src/backends/cpu/kernel/embedding/quant_embedding.rsindex 9d97c24e1..8a071e503 100644--- a/crates/backend-uzu/src/backends/cpu/kernel/embedding/quant_embedding.rs+++ b/crates/backend-uzu/src/backends/cpu/kernel/embedding/quant_embedding.rs@@ -11,35 +11,23 @@ use crate::{#[kernel(QuantizedEmbeddingLookup)]#[variants(T, f32, f16, bf16)]pub fn quantized_embedding_lookup<T: ArrayElement + Float>(- #[allow(unused)] token_ids: *const u32,- #[allow(unused)] weights: *const u8,- #[allow(unused)] scales: *const T,- #[allow(unused)]- #[optional(quantization_method == QuantizationMethod::ScaleZeroPoint)]- zero_points: Option<*const u8>,- #[allow(unused)]- #[optional(quantization_method == QuantizationMethod::ScaleBias)]- biases: Option<*const T>,- #[allow(unused)] output: *mut T,+ token_ids: *const u32,+ weights: *const u8,+ scales: *const T,+ #[optional(quantization_method == QuantizationMethod::ScaleZeroPoint)] zero_points: Option<*const u8>,+ #[optional(quantization_method == QuantizationMethod::ScaleBias)] biases: Option<*const T>,+ output: *mut T,#[allow(unused)]#[optional(use_hadamard)]output_hadamard_factors: Option<*const i32>,- #[allow(unused)] batch_size: u32,- #[allow(unused)] vocab_size: u32,- #[allow(unused)] model_dim: u32,- #[allow(unused)] input_scale: f32,- #[allow(unused)]- #[specialize]- group_size: u32,- #[allow(unused)]- #[specialize]- quantization_mode: QuantizationMode,- #[allow(unused)]- #[specialize]- quantization_method: QuantizationMethod,- #[allow(unused)]- #[specialize]- use_hadamard: bool,+ batch_size: u32,+ vocab_size: u32,+ model_dim: u32,+ input_scale: f32,+ #[specialize] group_size: u32,+ #[specialize] quantization_mode: QuantizationMode,+ #[specialize] quantization_method: QuantizationMethod,+ #[specialize] use_hadamard: bool,) {if use_hadamard {unimplemented!("not supported yet");diff --git a/crates/backend-uzu/src/backends/cpu/kernel/gdn/tree_verify/out.rs b/crates/backend-uzu/src/backends/cpu/kernel/gdn/tree_verify/out.rsindex c18492cc3..7bb84ebb5 100644--- a/crates/backend-uzu/src/backends/cpu/kernel/gdn/tree_verify/out.rs+++ b/crates/backend-uzu/src/backends/cpu/kernel/gdn/tree_verify/out.rs@@ -28,9 +28,7 @@ pub fn build_tree_out<QKT: ArrayElement + Float, OutputT: ArrayElement + Float>(#[allow(unused)]#[specialize]transposed_h0: bool,- #[allow(unused)]- #[specialize]- use_h0: bool,+ #[specialize] use_h0: bool,) {let batch_size = batch_size as usize;let tree_size = tree_size as usize;diff --git a/crates/backend-uzu/src/backends/cpu/kernel/tensor_add_bias/tensor_add_bias.rs b/crates/backend-uzu/src/backends/cpu/kernel/tensor_add_bias/tensor_add_bias.rsindex 86da05d80..a8bf72eb0 100644--- a/crates/backend-uzu/src/backends/cpu/kernel/tensor_add_bias/tensor_add_bias.rs+++ b/crates/backend-uzu/src/backends/cpu/kernel/tensor_add_bias/tensor_add_bias.rs@@ -8,16 +8,12 @@ use crate::array::ArrayElement;#[variants(T, f32, f16, bf16)]#[variants(BiasT, f32, f16, bf16)]pub fn tensor_add_bias<T: ArrayElement + Float, BiasT: ArrayElement + Float>(- #[allow(unused)]- #[optional(!in_place)]- input: Option<*const T>,- #[allow(unused)] bias: *const BiasT,- #[allow(unused)]- #[optional(indexed)]- bias_row_indices: Option<*const u32>,- #[allow(unused)] output: *mut T,- #[allow(unused)] num_cols: u32,- #[allow(unused)] length: u32,+ #[optional(!in_place)] input: Option<*const T>,+ bias: *const BiasT,+ #[optional(indexed)] bias_row_indices: Option<*const u32>,+ output: *mut T,+ num_cols: u32,+ length: u32,#[specialize] in_place: bool,#[specialize] indexed: bool,) {diff --git a/crates/backend-uzu/src/backends/cpu/kernel/tensor_add_swap/tensor_add_swap.rs b/crates/backend-uzu/src/backends/cpu/kernel/tensor_add_swap/tensor_add_swap.rsindex 26f840797..5ad4589e0 100644--- a/crates/backend-uzu/src/backends/cpu/kernel/tensor_add_swap/tensor_add_swap.rs+++ b/crates/backend-uzu/src/backends/cpu/kernel/tensor_add_swap/tensor_add_swap.rs@@ -7,9 +7,9 @@ use crate::array::ArrayElement;#[kernel(TensorAddSwap)]#[variants(T, f32, f16, bf16)]pub fn tensor_add_swap<T: ArrayElement + Float>(- #[allow(unused)] skip_buffer: *mut T,- #[allow(unused)] main_buffer: *mut T,- #[allow(unused)] length: u32,+ skip_buffer: *mut T,+ main_buffer: *mut T,+ length: u32,) {unsafe {for i in 0..length {diff --git a/crates/backend-uzu/src/backends/cpu/kernel/tensor_copy/tensor_copy.rs b/crates/backend-uzu/src/backends/cpu/kernel/tensor_copy/tensor_copy.rsindex 4460b6877..ba4e982d6 100644--- a/crates/backend-uzu/src/backends/cpu/kernel/tensor_copy/tensor_copy.rs+++ b/crates/backend-uzu/src/backends/cpu/kernel/tensor_copy/tensor_copy.rs@@ -7,9 +7,9 @@ use crate::array::ArrayElement;#[kernel(TensorCopy)]#[variants(T, f32, f16, bf16)]pub fn tensor_copy<T: ArrayElement + Float>(- #[allow(unused)] src_buffer: *const T,- #[allow(unused)] dst_buffer: *mut T,- #[allow(unused)] length: u32,+ src_buffer: *const T,+ dst_buffer: *mut T,+ length: u32,) {unsafe {for i in 0usize..(length as usize) {diff --git a/crates/backend-uzu/src/encodable_block/mixer/attention/mode.rs b/crates/backend-uzu/src/encodable_block/mixer/attention/mode.rsindex bcbde99d4..cc77db861 100644--- a/crates/backend-uzu/src/encodable_block/mixer/attention/mode.rs+++ b/crates/backend-uzu/src/encodable_block/mixer/attention/mode.rs@@ -214,7 +214,6 @@ impl<B: Backend> Attention<B> {self.out_projection.encode(attention_output, batch_dim.size(), encoder)}- #[allow(dead_code)]pub(crate) fn append_kv(&self,hidden: Allocation<B>,diff --git a/crates/backend-uzu/src/encodable_block/mixer/attention/state.rs b/crates/backend-uzu/src/encodable_block/mixer/attention/state.rsindex 3150c43a7..3b482ae36 100644--- a/crates/backend-uzu/src/encodable_block/mixer/attention/state.rs+++ b/crates/backend-uzu/src/encodable_block/mixer/attention/state.rs@@ -135,7 +135,6 @@ impl<B: Backend> AttentionState<B> {})}- #[allow(dead_code)]pub(super) fn append_full(&mut self,length: usize,diff --git a/crates/backend-uzu/src/tests/mod.rs b/crates/backend-uzu/src/tests/mod.rsindex bd7370a90..4721d1dd5 100644--- a/crates/backend-uzu/src/tests/mod.rs+++ b/crates/backend-uzu/src/tests/mod.rs@@ -1,5 +1,3 @@-#![allow(dead_code)] // TODO-#[cfg(not(backend = "cpu"))]compile_error!("uzu tests require cpu backend");diff --git a/flake.lock b/flake.lockindex 05c424f67..23adf7853 100644--- a/flake.lock+++ b/flake.lock@@ -2,11 +2,11 @@"nodes": {"crane": {"locked": {- "lastModified": 1778106249,- "narHash": "sha256-cM/AuKy5tMhwOOQIbha8ZRRMHVfNf7cv2aljIw+qoCg=",+ "lastModified": 1784564969,+ "narHash": "sha256-TkTWNhJV/8Wk3czlbz4bwHZkFCaCHPsOy+oJe4PUiXg=","owner": "ipetkov","repo": "crane",- "rev": "6d015ea29630b7ad2402841386da2cb617a470a7",+ "rev": "7930f6c291de6f83c257839d434592aa085f290a","type": "github"},"original": {@@ -35,11 +35,11 @@},"nixpkgs": {"locked": {- "lastModified": 1777954456,- "narHash": "sha256-hGdgeU2Nk87RAuZyYjyDjFL6LK7dAZN5RE9+hrDTkDU=",+ "lastModified": 1784796856,+ "narHash": "sha256-wWFrV5/Qbm+lyt5x20E/bSbfJiGKMo4RCxZV8cl/WZI=","owner": "nixos","repo": "nixpkgs",- "rev": "549bd84d6279f9852cae6225e372cc67fb91a4c1",+ "rev": "e2587caef70cea85dd97d7daab492899902dbf5d","type": "github"},"original": {@@ -64,11 +64,11 @@]},"locked": {- "lastModified": 1778469574,- "narHash": "sha256-NTZzJ7xJvMXOonYqut3WLUhryeZj5QuuL0ANcqS7d30=",+ "lastModified": 1785044261,+ "narHash": "sha256-ehF/c4fLdgmNu7YXEClnmwhoBYddhYIVoSQpsZoorC4=","owner": "oxalica","repo": "rust-overlay",- "rev": "4852a8aa041c94af55e136cde5b8b6d42c3563e8",+ "rev": "fe2124391e739e7b3e8720cd8a73f06edd0aceba","type": "github"},"original": {diff --git a/flake.nix b/flake.nixindex f4345dc15..c8f7239c4 100644--- a/flake.nix+++ b/flake.nix@@ -72,6 +72,7 @@rustToolchaincargo-denycargo-nextest+ cargo-hackcargo-expandcargo-flamegraphcargo-show-asm