Add benchmark manifest runner

Author: zinyando

benchmarks/manifests/smoke.toml

@@ -0,0 +1,23 @@
+# Smoke benchmark matrix for local validation.
+#
+# Override the server at runtime with `izwi --server <URL> bench run ...`,
+# or set `server` here for a fixed target.
+
+[[benchmarks]]
+name = "chat-short-c1"
+command = "chat"
+model = "Qwen3.5-4B"
+iterations = 1
+concurrent = 1
+warmup = false
+prompt = "Summarize why batching helps transformer prefill in one sentence."
+max_tokens = 64
+
+[[benchmarks]]
+name = "asr-short-c1"
+command = "asr"
+model = "parakeet-tdt-0.6b-v3"
+file = "../../data/test.wav"
+iterations = 1
+concurrent = 1
+warmup = false

crates/izwi-cli/README.md

@@ -147,6 +147,7 @@ izwi bench tts --model qwen3-tts-0.6b-base --iterations 5 --concurrent 2
 izwi bench asr --model parakeet-tdt-0.6b-v3 --file data/test.wav --iterations 3 --concurrent 2
 izwi bench throughput --duration 10 --concurrent 2
 izwi --output-format json bench chat --iterations 5
+izwi --output-format json bench run benchmarks/local.toml
 ```
 
 ### Config

crates/izwi-cli/src/app/cli.rs

@@ -537,6 +537,13 @@ pub enum BenchCommands {
         #[arg(short, long, default_value = "1")]
         concurrent: u32,
     },
+
+    /// Run a benchmark manifest
+    Run {
+        /// Benchmark manifest path (TOML)
+        #[arg(value_name = "PATH")]
+        manifest: PathBuf,
+    },
 }
 
 #[derive(Subcommand)]