From 609b63211625a3485f1f94050b722a96e64ecf43 Mon Sep 17 00:00:00 2001
From: webbrain-one <295484252+webbrain-one@users.noreply.github.com>
Date: Sat, 1 Aug 2026 13:39:28 +0300
Subject: [PATCH] Add support for Qwen3 dense models

Leverages architectural similarity with Qwen2/Qwen2.5 to integrate
Qwen3 dense models with minimal changes. Updates benchmark scripts
to include Qwen3 evaluations.
---
 benchmarks/benchmark_e2e_arc.py | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/benchmarks/benchmark_e2e_arc.py b/benchmarks/benchmark_e2e_arc.py
index d80ba4a..53001c6 100644
--- a/benchmarks/benchmark_e2e_arc.py
+++ b/benchmarks/benchmark_e2e_arc.py
@@ -74,6 +74,16 @@ class ModelConfig:
             attention_bias=True,
             mlp_bias=True
         ),
+    "qwen3-8b":
+        ModelConfig(
+            name='qwen3-8b',
+            num_layers=36,
+            num_heads=28,
+            hidden_size=4096,
+            intermediate_size=14336,
+            attention_bias=False,
+            mlp_bias=False
+        ),
 }