From 825e0da374cd73bc0058502019ac2c8995dbf9fc Mon Sep 17 00:00:00 2001
From: Eric Buehler <65165915+EricLBuehler@users.noreply.github.com>
Date: Mon, 2 Sep 2024 11:04:31 -0400
Subject: [PATCH] [Breaking] Bump version to v0.3.0 (#736)

* Move into constructor

* Bump version to 0.3.0
---
 Cargo.lock                             | 16 ++++++++--------
 Cargo.toml                             |  2 +-
 README.md                              |  2 +-
 mistralrs-bench/Cargo.toml             |  2 +-
 mistralrs-core/Cargo.toml              |  6 +++---
 mistralrs-core/src/engine/mod.rs       |  9 ++-------
 mistralrs-core/src/lib.rs              | 11 ++++-------
 mistralrs-pyo3/Cargo.toml              |  2 +-
 mistralrs-pyo3/Cargo_template.toml     |  2 +-
 mistralrs-pyo3/pyproject.toml          |  2 +-
 mistralrs-pyo3/pyproject_template.toml |  2 +-
 mistralrs-server/Cargo.toml            |  2 +-
 mistralrs/Cargo.toml                   |  2 +-
 13 files changed, 26 insertions(+), 34 deletions(-)

diff --git a/Cargo.lock b/Cargo.lock
index ef186b69c..4f74c8438 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2094,7 +2094,7 @@ dependencies = [
 
 [[package]]
 name = "mistralrs"
-version = "0.2.5"
+version = "0.3.0"
 dependencies = [
  "anyhow",
  "candle-core",
@@ -2112,7 +2112,7 @@ dependencies = [
 
 [[package]]
 name = "mistralrs-bench"
-version = "0.2.5"
+version = "0.3.0"
 dependencies = [
  "anyhow",
  "candle-core",
@@ -2127,7 +2127,7 @@ dependencies = [
 
 [[package]]
 name = "mistralrs-core"
-version = "0.2.5"
+version = "0.3.0"
 dependencies = [
  "accelerate-src",
  "akin",
@@ -2195,7 +2195,7 @@ dependencies = [
 
 [[package]]
 name = "mistralrs-paged-attn"
-version = "0.2.5"
+version = "0.3.0"
 dependencies = [
  "anyhow",
  "bindgen_cuda 0.1.6",
@@ -2205,7 +2205,7 @@ dependencies = [
 
 [[package]]
 name = "mistralrs-pyo3"
-version = "0.2.5"
+version = "0.3.0"
 dependencies = [
  "accelerate-src",
  "anyhow",
@@ -2229,7 +2229,7 @@ dependencies = [
 
 [[package]]
 name = "mistralrs-quant"
-version = "0.2.5"
+version = "0.3.0"
 dependencies = [
  "bindgen_cuda 0.1.5",
  "candle-core",
@@ -2244,7 +2244,7 @@ dependencies = [
 
 [[package]]
 name = "mistralrs-server"
-version = "0.2.5"
+version = "0.3.0"
 dependencies = [
  "accelerate-src",
  "anyhow",
@@ -2273,7 +2273,7 @@ dependencies = [
 
 [[package]]
 name = "mistralrs-vision"
-version = "0.2.5"
+version = "0.3.0"
 dependencies = [
  "candle-core",
  "image",
diff --git a/Cargo.toml b/Cargo.toml
index f67176c31..9bd4a85c0 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -14,7 +14,7 @@ exclude = [
 resolver = "2"
 
 [workspace.package]
-version = "0.2.5"
+version = "0.3.0"
 edition = "2021"
 description = "Fast and easy LLM serving."
 homepage = "https://github.com/EricLBuehler/mistral.rs"
diff --git a/README.md b/README.md
index 1be543896..a1383f574 100644
--- a/README.md
+++ b/README.md
@@ -210,7 +210,7 @@ Please submit more benchmarks via raising an issue!
 > Note: You can use our [Docker containers here](https://github.com/EricLBuehler/mistral.rs/pkgs/container/mistral.rs).
 > Learn more about running Docker containers: https://docs.docker.com/engine/reference/run/
 
-> Note: You can use pre-built `mistralrs-server` binaries [here](https://github.com/EricLBuehler/mistral.rs/releases/tag/v0.2.5)
+> Note: You can use pre-built `mistralrs-server` binaries [here](https://github.com/EricLBuehler/mistral.rs/releases/tag/v0.3.0)
 
 - Install the [Python package here](mistralrs-pyo3/README.md).
 
diff --git a/mistralrs-bench/Cargo.toml b/mistralrs-bench/Cargo.toml
index 8e38eaa46..4763fedc4 100644
--- a/mistralrs-bench/Cargo.toml
+++ b/mistralrs-bench/Cargo.toml
@@ -18,7 +18,7 @@ candle-core.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 clap.workspace = true
-mistralrs-core = { version = "0.2.5", path = "../mistralrs-core" }
+mistralrs-core = { version = "0.3.0", path = "../mistralrs-core" }
 tracing.workspace = true
 tokio.workspace = true
 cli-table = "0.4.7"
diff --git a/mistralrs-core/Cargo.toml b/mistralrs-core/Cargo.toml
index 4f288ade5..a071b4839 100644
--- a/mistralrs-core/Cargo.toml
+++ b/mistralrs-core/Cargo.toml
@@ -64,13 +64,13 @@ tracing-subscriber.workspace = true
 derive-new = "0.6.0"
 itertools = "0.13.0"
 sysinfo = "0.30.12"
-mistralrs-vision = { version = "0.2.5", path = "../mistralrs-vision" }
+mistralrs-vision = { version = "0.3.0", path = "../mistralrs-vision" }
 csv = "1.3.0"
 reqwest.workspace = true
 base64.workspace = true
 bytemuck_derive = "1.7.0"
-mistralrs-paged-attn = { version = "0.2.5", path = "../mistralrs-paged-attn", optional = true }
-mistralrs-quant = { version = "0.2.0", path = "../mistralrs-quant" }
+mistralrs-paged-attn = { version = "0.3.0", path = "../mistralrs-paged-attn", optional = true }
+mistralrs-quant = { version = "0.3.0", path = "../mistralrs-quant" }
 uuid = { version = "1.10.0", features = ["v4"] }
 schemars = "0.8.21"
 serde_yaml = "0.9.34"
diff --git a/mistralrs-core/src/engine/mod.rs b/mistralrs-core/src/engine/mod.rs
index 808257afc..d993472a9 100644
--- a/mistralrs-core/src/engine/mod.rs
+++ b/mistralrs-core/src/engine/mod.rs
@@ -62,6 +62,7 @@ impl Engine {
         no_prefix_cache: bool,
         prefix_cache_n: usize,
         disable_eos_stop: bool,
+        throughput_logging_enabled: bool,
     ) -> Self {
         let device = get_mut_arcmutex!(pipeline).device().clone();
         let is_xlora = get_mut_arcmutex!(pipeline).get_metadata().is_xlora;
@@ -84,16 +85,10 @@ impl Engine {
             ),
             is_debug: DEBUG.load(Ordering::Relaxed),
             disable_eos_stop,
-            throughput_logging_enabled: false,
+            throughput_logging_enabled,
         }
     }
 
-    // TODO(EricLBuehler): On v0.3.0 move this into the Engine constructor
-    /// Enable throughput logging.
-    pub fn enable_throughput_logging(&mut self) {
-        self.throughput_logging_enabled = true;
-    }
-
     pub async fn run(&mut self) {
         let rng = Arc::new(std::sync::Mutex::new(Isaac64Rng::seed_from_u64(SEED)));
         let mut last_completion_ids: Vec<usize> = vec![];
diff --git a/mistralrs-core/src/lib.rs b/mistralrs-core/src/lib.rs
index 0a7825c6a..1fd4291ad 100644
--- a/mistralrs-core/src/lib.rs
+++ b/mistralrs-core/src/lib.rs
@@ -293,6 +293,7 @@ impl MistralRs {
         let no_prefix_cache = no_prefix_cache.unwrap_or(false);
         let prefix_cache_n = prefix_cache_n.unwrap_or(16);
         let disable_eos_stop = disable_eos_stop.unwrap_or(false);
+        let throughput_logging_enabled = throughput_logging_enabled.is_some();
 
         let reboot_state = RebootState {
             pipeline: pipeline.clone(),
@@ -302,7 +303,7 @@ impl MistralRs {
             no_prefix_cache,
             prefix_cache_n,
             disable_eos_stop,
-            throughput_logging_enabled: throughput_logging_enabled.is_some(),
+            throughput_logging_enabled,
         };
 
         let (tx, rx) = channel(10_000);
@@ -322,10 +323,8 @@ impl MistralRs {
                     no_prefix_cache,
                     prefix_cache_n,
                     disable_eos_stop,
+                    throughput_logging_enabled,
                 );
-                if throughput_logging_enabled.is_some() {
-                    engine.enable_throughput_logging();
-                }
                 engine.run().await;
             });
         });
@@ -375,10 +374,8 @@ impl MistralRs {
                         reboot_state.no_prefix_cache,
                         reboot_state.prefix_cache_n,
                         reboot_state.disable_eos_stop,
+                        reboot_state.throughput_logging_enabled,
                     );
-                    if reboot_state.throughput_logging_enabled {
-                        engine.enable_throughput_logging();
-                    }
                     engine.run().await;
                 });
             });
diff --git a/mistralrs-pyo3/Cargo.toml b/mistralrs-pyo3/Cargo.toml
index fc09d788f..b71e0e563 100644
--- a/mistralrs-pyo3/Cargo.toml
+++ b/mistralrs-pyo3/Cargo.toml
@@ -17,7 +17,7 @@ doc = false
 
 [dependencies]
 pyo3.workspace = true
-mistralrs-core = { version = "0.2.5", path = "../mistralrs-core", features = ["pyo3_macros"] }
+mistralrs-core = { version = "0.3.0", path = "../mistralrs-core", features = ["pyo3_macros"] }
 serde.workspace = true
 serde_json.workspace = true
 candle-core.workspace = true
diff --git a/mistralrs-pyo3/Cargo_template.toml b/mistralrs-pyo3/Cargo_template.toml
index 087ff7708..c0339b48f 100644
--- a/mistralrs-pyo3/Cargo_template.toml
+++ b/mistralrs-pyo3/Cargo_template.toml
@@ -17,7 +17,7 @@ doc = false
 
 [dependencies]
 pyo3.workspace = true
-mistralrs-core = { version = "0.2.5", path = "../mistralrs-core", features=["pyo3_macros","$feature_name"] }
+mistralrs-core = { version = "0.3.0", path = "../mistralrs-core", features=["pyo3_macros","$feature_name"] }
 serde.workspace = true
 serde_json.workspace = true
 candle-core = { git = "https://github.com/EricLBuehler/candle.git", version = "0.6.0", rev = "3c8e120", features=["$feature_name"] }
diff --git a/mistralrs-pyo3/pyproject.toml b/mistralrs-pyo3/pyproject.toml
index caa863b3a..b9da18c6c 100644
--- a/mistralrs-pyo3/pyproject.toml
+++ b/mistralrs-pyo3/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "maturin"
 
 [project]
 name = "mistralrs"
-version = "0.2.5"
+version = "0.3.0"
 requires-python = ">=3.8"
 classifiers = [
     "Programming Language :: Rust",
diff --git a/mistralrs-pyo3/pyproject_template.toml b/mistralrs-pyo3/pyproject_template.toml
index 99479c192..4c2199ef3 100644
--- a/mistralrs-pyo3/pyproject_template.toml
+++ b/mistralrs-pyo3/pyproject_template.toml
@@ -4,7 +4,7 @@ build-backend = "maturin"
 
 [project]
 name = "$name"
-version = "0.2.5"
+version = "0.3.0"
 requires-python = ">=3.8"
 classifiers = [
     "Programming Language :: Rust",
diff --git a/mistralrs-server/Cargo.toml b/mistralrs-server/Cargo.toml
index bd08f6576..62560af12 100644
--- a/mistralrs-server/Cargo.toml
+++ b/mistralrs-server/Cargo.toml
@@ -22,7 +22,7 @@ axum = { version = "0.7.4", features = ["tokio"] }
 tower-http = { version = "0.5.1", features = ["cors"]}
 utoipa = { version = "4.2", features = ["axum_extras"] }
 utoipa-swagger-ui = { version = "7.1.0", features = ["axum"]}
-mistralrs-core = { version = "0.2.5", path = "../mistralrs-core" }
+mistralrs-core = { version = "0.3.0", path = "../mistralrs-core" }
 indexmap.workspace = true
 accelerate-src = { workspace = true, optional = true }
 intel-mkl-src = { workspace = true, optional = true }
diff --git a/mistralrs/Cargo.toml b/mistralrs/Cargo.toml
index 04fb460b7..36071e008 100644
--- a/mistralrs/Cargo.toml
+++ b/mistralrs/Cargo.toml
@@ -12,7 +12,7 @@ license.workspace = true
 homepage.workspace = true
 
 [dependencies]
-mistralrs-core = { version = "0.2.5", path = "../mistralrs-core" }
+mistralrs-core = { version = "0.3.0", path = "../mistralrs-core" }
 anyhow.workspace = true
 tokio.workspace = true
 candle-core.workspace = true