Document standalone image gen script in README

Author: jbschlosser

README.md

@@ -52,10 +52,21 @@ For hardware, we used a 96GB 700W H100 GPU. Some of the optimizations applied (B
 
 ## Run the optimized pipeline
 
-TODO
+```
+python gen_image.py --prompt "An astronaut standing next to a giant lemon" --output-file output.png --use-cached-model
+```
+
+This will include all optimizations and will attempt to use pre-cached binary models
+generated via `torch.export` + AOTI. To generate these binaries for subsequent runs, run
+the above command without the `--use-cached-model` flag.
 
 > [!IMPORTANT]
-> The binaries won't work for hardware that are different from the ones they were obtained on. For example, if the binaries were obtained on an H100, they won't work on A100.
+> The binaries won't work for hardware that is sufficiently different from the hardware they were
+> obtained on. For example, if the binaries were obtained on an H100, they won't work on A100.
+> Further, the binaries are currently Linux-only and include dependencies on specific versions
+> of system libs such as libstdc++; they will not work if they were generated in a sufficiently
+> different environment than the one present at runtime. The PyTorch Compiler team is working on
+> solutions for more portable binaries / artifact caching.
 
 ## Benchmarking
 [`run_benchmark.py`](./run_benchmark.py) is the main script for benchmarking the different optimization techniques.

gen_image.py

@@ -24,6 +24,4 @@ def main(args):
 if __name__ == "__main__":
     parser = create_parser()
     args = parser.parse_args()
-    # use the cached model to minimize latency
-    args.use_cached_model = True
     main(args)