Upgrading our rustc version. (#2908)

* Upgrading our rustc version.

* Fixing the rust tests to proper version.

* Clippy everything.

Author: Narsil

.github/workflows/tests.yaml

@@ -31,7 +31,7 @@ jobs:
         with:
           # Released on: 02 May, 2024
           # https://releases.rs/docs/1.78.0/
-          toolchain: 1.80.0
+          toolchain: 1.84.0
           override: true
           components: rustfmt, clippy
       - name: Install Protoc

Dockerfile

@@ -1,5 +1,5 @@
 # Rust builder
-FROM lukemathwalker/cargo-chef:latest-rust-1.80.1 AS chef
+FROM lukemathwalker/cargo-chef:latest-rust-1.84.0 AS chef
 WORKDIR /usr/src
 
 ARG CARGO_REGISTRIES_CRATES_IO_PROTOCOL=sparse

Dockerfile_amd

@@ -1,5 +1,5 @@
 # Rust builder
-FROM lukemathwalker/cargo-chef:latest-rust-1.80.1 AS chef
+FROM lukemathwalker/cargo-chef:latest-rust-1.84.0 AS chef
 WORKDIR /usr/src
 
 ARG CARGO_REGISTRIES_CRATES_IO_PROTOCOL=sparse

Dockerfile_intel

@@ -1,6 +1,6 @@
 ARG PLATFORM=xpu
 
-FROM lukemathwalker/cargo-chef:latest-rust-1.80.1 AS chef
+FROM lukemathwalker/cargo-chef:latest-rust-1.84.0 AS chef
 WORKDIR /usr/src
 
 ARG CARGO_REGISTRIES_CRATES_IO_PROTOCOL=sparse

Dockerfile_trtllm

@@ -2,7 +2,7 @@ ARG CUDA_ARCH_LIST="75-real;80-real;86-real;89-real;90-real"
 ARG OMPI_VERSION="4.1.7rc1"
 
 # Build dependencies resolver stage
-FROM lukemathwalker/cargo-chef:latest AS chef
+FROM lukemathwalker/cargo-chef:latest-rust-1.84.0 AS chef
 WORKDIR /usr/src/text-generation-inference/backends/trtllm
 
 FROM chef AS planner

backends/grpc-metadata/src/lib.rs

@@ -8,7 +8,7 @@ use tracing_opentelemetry::OpenTelemetrySpanExt;
 /// Inject context in the metadata of a gRPC request.
 struct MetadataInjector<'a>(pub &'a mut tonic::metadata::MetadataMap);
 
-impl<'a> Injector for MetadataInjector<'a> {
+impl Injector for MetadataInjector<'_> {
     /// Set a key and value in the MetadataMap.  Does nothing if the key or value are not valid inputs
     fn set(&mut self, key: &str, value: String) {
         if let Ok(key) = tonic::metadata::MetadataKey::from_bytes(key.as_bytes()) {

backends/v2/src/queue.rs

@@ -213,8 +213,7 @@ impl State {
         }
 
         // Pad prefill_token_budget to be a multiple of block size
-        let prefill_token_budget =
-            ((prefill_token_budget + self.block_size - 1) / self.block_size) * self.block_size;
+        let prefill_token_budget = prefill_token_budget.div_ceil(self.block_size) * self.block_size;
 
         // Create span for this batch to add context to inference calls
         let next_batch_span = info_span!(parent: None, "batch", batch_size = tracing::field::Empty);
@@ -245,9 +244,8 @@ impl State {
                 prefill_tokens = (batch_requests.len() + 1) as u32 * max_input_length
             } else {
                 // pad to block size
-                prefill_tokens += ((entry.request.input_length + self.block_size - 1)
-                    / self.block_size)
-                    * self.block_size;
+                prefill_tokens +=
+                    entry.request.input_length.div_ceil(self.block_size) * self.block_size;
             }
 
             if self.requires_padding {
@@ -262,8 +260,7 @@ impl State {
                 };
 
                 // pad to block size
-                decode_tokens +=
-                    ((max_new_tokens + self.block_size - 1) / self.block_size) * self.block_size;
+                decode_tokens += max_new_tokens.div_ceil(self.block_size) * self.block_size;
             }
 
             if prefill_tokens > prefill_token_budget

backends/v3/src/block_allocator.rs

@@ -165,13 +165,13 @@ impl Allocator for SimpleAllocator {
             let (tokens, repeats) = match self.window_size {
                 None => (tokens, 1),
                 Some(window_size) => {
-                    let repeats = (tokens + window_size - 1) / window_size;
+                    let repeats = tokens.div_ceil(window_size);
                     let tokens = core::cmp::min(tokens, window_size);
                     (tokens, repeats as usize)
                 }
             };
             // Pad to a multiple of block size
-            let required_blocks = (tokens + self.block_size - 1) / self.block_size;
+            let required_blocks = tokens.div_ceil(self.block_size);
             (required_blocks, repeats)
         };
 

backends/v3/src/queue.rs

@@ -257,8 +257,7 @@ impl State {
         }
 
         // Pad prefill_token_budget to be a multiple of block size
-        let prefill_token_budget =
-            ((prefill_token_budget + self.block_size - 1) / self.block_size) * self.block_size;
+        let prefill_token_budget = prefill_token_budget.div_ceil(self.block_size) * self.block_size;
 
         // Create span for this batch to add context to inference calls
         let next_batch_span = info_span!(parent: None, "batch", batch_size = tracing::field::Empty);

backends/v3/src/radix.rs

@@ -103,7 +103,7 @@ impl Allocator for RadixAllocator {
         let prefix_len = blocks.len() * self.block_size as usize;
         let suffix_len = tokens - prefix_len as u32;
 
-        let suffix_blocks = (suffix_len + self.block_size - 1) / self.block_size;
+        let suffix_blocks = suffix_len.div_ceil(self.block_size);
 
         tracing::info!("Prefix {prefix_len} - Suffix {suffix_len}");