Use range-checked crate

Cargo.toml

@@ -35,6 +35,7 @@ env_logger = "0.10.1"
 tracing = "0.1.40"
 tch = { git = "https://github.com/EricLBuehler/tch-rs.git", branch = "copy_derive", version = "0.14.0" } # pending on LaurentMazare/tch-rs#823
 torch-sys = { git = "https://github.com/EricLBuehler/tch-rs.git", branch = "copy_derive", version = "0.14.0" } # pending on LaurentMazare/tch-rs#823
+range-checked = { git = "https://github.com/EricLBuehler/range-checked.git", version = "0.1.0" }
 
 [features]
 default = []
@@ -43,4 +44,4 @@ cuda = ["candle-core/cuda", "candle-nn/cuda", "candle-transformers/cuda", "dep:c
 cudnn = ["candle-core/cudnn"]
 flash-attn = ["cuda", "candle-transformers/flash-attn"]
 mkl = ["dep:intel-mkl-src", "candle-core/mkl", "candle-nn/mkl", "candle-transformers/mkl"]
-nccl = ["cuda", "cudarc/nccl", "dep:half"]
+nccl = ["cuda", "cudarc/nccl", "dep:half"]

src/paged_attention/cache_engine.rs

@@ -1,5 +1,7 @@
 use candle_core::{DType, Device, Tensor};
 
+use range_checked::F64Bounded;
+
 use crate::openai::{models::ConfigLike, responses::APIError};
 
 const _GB: usize = 1 << 30;
@@ -16,24 +18,18 @@ pub(crate) struct CacheConfig {
 impl CacheConfig {
  pub(crate) fn new(
  block_size: usize,
- gpu_mem_utilization: f64,
+ gpu_mem_utilization: F64Bounded<0, 1, false>,
  swap_space_bytes: usize,
  sliding_window: Option<usize>,
- ) -> Result<Self, APIError> {
- if gpu_mem_utilization > 1.0 {
- return Err(APIError::new(format!(
- "GPU memory utilization must be less that 1.0. Got {gpu_mem_utilization}"
- )));
- }
-
- Ok(Self {
+ ) -> Self {
+ Self {
  block_size,
- gpu_mem_utilization,
+ gpu_mem_utilization: *gpu_mem_utilization,
  swap_space_bytes: swap_space_bytes * _GB,
  sliding_window,
  num_gpu_blocks: None,
  num_cpu_blocks: None,
- })
+ }
  }
 }