copy

_posts/2024-05-22-swe-bench-lite.md

@@ -1,12 +1,15 @@
 ---
 title: Aider scores SOTA 26.3% on SWE Bench Lite
 excerpt: Aider scored 26.3% on SWE Bench Lite, achieving a state of the art result.
+highlight_image: /assets/swe_bench_lite.jpg
 draft: true
 ---
 
+[![SWE Bench Lite results](/assets/swe_bench_lite.svg)](https://aider.chat/assets/swe_bench_lite.svg)
+
 # Aider scores SOTA 26.3% on SWE Bench Lite
  
-[Aider scored 26.3%]()
+Aider scored 26.3%
 on the
 [SWE Bench Lite benchmark](https://www.swebench.com), achieving a state of the art result. 
 The current top leaderboard entry is 20.33%
@@ -14,6 +17,8 @@ from Amazon Q Developer Agent.
 The best result reported elsewhere online seems to be
 [22.3% from AutoCodeRover](https://github.com/nus-apr/auto-code-rover).
 
+## Interactive, not agentic
+
 Aider achieved this result mainly through its focus on static code analysis,
 reliable LLM code editing
 and pragmatic workflows for interactive pair programming with AI.
@@ -33,6 +38,8 @@ When a user asks aider for a change, they see the edits performed in real-time.
 Aider may also then offer additional
 help like fixing lint or test errors.
 
+## Methodology
+
 For the benchmark, 
 aider was launched in each problem's git repository
 with the problem statement
@@ -113,7 +120,7 @@ Some noteworthy observations:
 | 6 | Opus   |  1 |  1.3 | 100.0
 |**Total**|   | **79** | **100%** | **100%** |
 
-If we just look at which models produced correct solutions,
+If we breakdown correct solutions purely by model,
 we can see that GPT-4o dominates.
 This isn't a fair comparison, because GPT-4o always took the first
 attempt at solving.
@@ -145,8 +152,7 @@ to provide a compact and powerful summary of the entire code base.
 The map is constantly
 tailored to show
 repo context that is relevant to the current state of the chat conversation.
-
-by performing a graph optimization on the code's call graph.
+This is done by performing a graph optimization on the code's call graph.
 
 When the user asks for a change to their code, the LLM uses the repo map
 to decide which files to edit.

benchmark/swe_bench_lite.py

@@ -1,6 +1,7 @@
 import matplotlib.pyplot as plt
-from matplotlib import rc
 from imgcat import imgcat
+from matplotlib import rc
+
 
 def plot_swe_bench_lite(data_file):
     with open(data_file, "r") as file:
@@ -22,30 +23,37 @@ def plot_swe_bench_lite(data_file):
 
     rc("font", **{"family": "sans-serif", "sans-serif": ["Helvetica"], "size": 10})
 
-    fig, ax = plt.subplots(figsize=(10, 5))
+    fig, ax = plt.subplots(figsize=(10, 6))
     ax.grid(axis="y", zorder=0, lw=0.2)
     for spine in ax.spines.values():
         spine.set_edgecolor("#DDDDDD")
         spine.set_linewidth(0.5)
 
-    colors = [
-        "#b3e6a8" if "Aider" in model else "#b3d1e6" for model in models
-    ]
+    colors = ["#b3e6a8" if "Aider" in model else "#b3d1e6" for model in models]
     bars = ax.bar(models, pass_rates, color=colors, alpha=0.75, zorder=3)
 
     for bar in bars:
         yval = bar.get_height()
-        ax.text(bar.get_x() + bar.get_width()/2, yval + 0.5, f'{yval}%', ha='center', va='bottom', fontsize=12, alpha=0.75)
+        ax.text(
+            bar.get_x() + bar.get_width() / 2,
+            yval + 0.5,
+            f"{yval}%",
+            ha="center",
+            va="bottom",
+            fontsize=12,
+            alpha=0.75,
+        )
 
     # ax.set_xlabel("Models", fontsize=18)
     ax.set_ylabel("Pass rate (%)", fontsize=18)
-    ax.set_title("SWE Bench Lite pass rates", fontsize=20)
+    ax.set_title("SWE Bench Lite", fontsize=20)
     ax.set_ylim(0, 30)
-    plt.xticks(rotation=45, ha='right', fontsize=16)
+    plt.xticks(rotation=45, ha="right", fontsize=16)
     plt.tight_layout(pad=3.0)
-    plt.savefig("swe_bench_lite.png")
+    plt.savefig("swe_bench_lite.jpg")
     plt.savefig("swe_bench_lite.svg")
     imgcat(fig)
 
+
 # Example usage
 plot_swe_bench_lite("benchmark/tmp.txt")