diff --git a/.gitignore b/.gitignore
index e18333e7b4cc..9f7128d495a9 100644
--- a/.gitignore
+++ b/.gitignore
@@ -356,3 +356,4 @@ venv/
 
 # Log files
 *.log
+sweep/
diff --git a/benchmarks/dynamo/combine_csv.py b/benchmarks/dynamo/combine_csv.py
new file mode 100644
index 000000000000..b579e0a1bbbd
--- /dev/null
+++ b/benchmarks/dynamo/combine_csv.py
@@ -0,0 +1,49 @@
+# This script takes csvs produced by parse_logs.py and combines them
+# into a single CSV file
+
+import ast
+import csv
+import sys
+from collections import defaultdict
+
+assert len(sys.argv) == 3
+
+RESULTS = defaultdict(dict)
+
+for side, f in zip(["static", "dynamic"], sys.argv[1:]):
+    with open(f, "r") as f:
+        reader = csv.DictReader(f)
+        for row in reader:
+            RESULTS[(row["bench"], row["name"])][side] = row
+
+fields = ["frame_time", "graph_breaks"]
+
+out = csv.DictWriter(
+    sys.stdout,
+    ["bench", "name"] + [f"delta_{n}" for n in fields] + ["static_url", "dynamic_url"],
+    dialect="excel",
+)
+out.writeheader()
+
+for (bench, name), sides in RESULTS.items():
+    if "static" not in sides:
+        continue
+    if "dynamic" not in sides:
+        continue
+    if not name:
+        out.writerow(
+            {
+                "static_url": sides["static"]["explain"],
+                "dynamic_url": sides["dynamic"]["explain"],
+            }
+        )
+        continue
+    row = {"bench": bench, "name": name}
+    for f in fields:
+        try:
+            static = ast.literal_eval(sides["static"][f])
+            dynamic = ast.literal_eval(sides["dynamic"][f])
+        except SyntaxError:
+            continue
+        row[f"delta_{f}"] = dynamic - static
+    out.writerow(row)
diff --git a/benchmarks/dynamo/parse_logs.py b/benchmarks/dynamo/parse_logs.py
index a555c4d52c16..a82648d4dd77 100644
--- a/benchmarks/dynamo/parse_logs.py
+++ b/benchmarks/dynamo/parse_logs.py
@@ -1,7 +1,6 @@
 import csv
 import os
 import re
-import subprocess
 import sys
 
 # This script takes the logs produced by the benchmark scripts (e.g.,
@@ -24,11 +23,6 @@ m = re.search(r"https://gist.github.com/[a-f0-9]+", full_log)
 if m is not None:
     gist_url = m.group(0)
 
-# Record the current commit hash for ease of reproducibility
-hash = subprocess.check_output(
-    "git rev-parse HEAD".split(" "), encoding="utf-8"
-).rstrip()
-
 # Split the log into an entry per benchmark
 entries = re.split(
     r"(?:cuda (?:train|eval) +([^ ]+)|WARNING:root:([^ ]+) failed to load)", full_log
@@ -45,24 +39,26 @@ def chunker(seq, size):
 c = 0
 i = 0
 
-out = csv.writer(sys.stdout, dialect="excel")
-out.writerow(
+out = csv.DictWriter(
+    sys.stdout,
     [
-        "",
-        hash,
-        "",
-        "",
-        "",
-        "",
-        gist_url,
+        "bench",
+        "name",
+        "result",
+        "component",
+        "context",
+        "explain",
         "frame_time",
         "backend_time",
         "graph_count",
         "op_count",
         "graph_breaks",
         "unique_graph_breaks",
-    ]
+    ],
+    dialect="excel",
 )
+out.writeheader()
+out.writerow({"explain": gist_url})
 
 # Sometimes backtraces will be in third party code, which results
 # in very long file names.  Delete the absolute path in this case.
@@ -179,21 +175,20 @@ for name, name2, log in chunker(entries, 3):
         context = ""
 
     out.writerow(
-        [
-            bench,
-            name,
-            "",
-            r,
-            component,
-            context,
-            explain,
-            frame_time,
-            backend_time,
-            graph_count,
-            op_count,
-            graph_breaks,
-            unique_graph_breaks,
-        ]
+        {
+            "bench": bench,
+            "name": name,
+            "result": r,
+            "component": component,
+            "context": context,
+            "explain": explain,
+            "frame_time": frame_time,
+            "backend_time": backend_time,
+            "graph_count": graph_count,
+            "op_count": op_count,
+            "graph_breaks": graph_breaks,
+            "unique_graph_breaks": unique_graph_breaks,
+        }
     )
     i += 1
 
diff --git a/benchmarks/dynamo/run_delta.sh b/benchmarks/dynamo/run_delta.sh
new file mode 100755
index 000000000000..7ca5a881a284
--- /dev/null
+++ b/benchmarks/dynamo/run_delta.sh
@@ -0,0 +1,22 @@
+#!/bin/bash
+
+set -x
+
+# Some QoL for people running this script on Meta servers
+if getent hosts fwdproxy; then
+    export https_proxy=http://fwdproxy:8080 http_proxy=http://fwdproxy:8080 no_proxy=.fbcdn.net,.facebook.com,.thefacebook.com,.tfbnw.net,.fb.com,.fburl.com,.facebook.net,.sb.fbsbx.com,localhost
+fi
+
+WORK="$PWD"
+
+cd "$(dirname "$BASH_SOURCE")"/../..
+
+ROOT="$PWD"
+
+mkdir -p "$WORK/sweep/static"
+mkdir -p "$WORK/sweep/dynamic"
+
+(cd "$WORK/sweep/static" && "$ROOT/benchmarks/dynamo/run_all.sh" "$@")
+(cd "$WORK/sweep/dynamic" && "$ROOT/benchmarks/dynamo/run_all.sh" "$@" --dynamic-shapes)
+python benchmarks/dynamo/combine_csv.py "$WORK/sweep/static/final.csv" "$WORK/sweep/dynamic/final.csv" > "$WORK/delta.csv"
+gh gist create "$WORK/delta.csv"