Point at the repository this is a companion to

The card and the app footer both linked to jmurray10/peas_agents, which is
retired and private. The public repository is jmurray10/agent_design_peas.
Two lines; nothing else in the Space changes.

README.md

@@ -13,7 +13,7 @@ short_description: Where a GPU stops being worth the transfer
 
 # The parallelization floor, measured on real CUDA
 
-Companion to [jmurray10/peas_agents](https://github.com/jmurray10/peas_agents), which
+Companion to [jmurray10/agent_design_peas](https://github.com/jmurray10/agent_design_peas), which
 argues that classical agent architectures survive the arrival of LLMs and that an LLM
 replaces one component inside them rather than the architecture.
 

app.py

@@ -31,7 +31,7 @@ with gr.Blocks(title="The parallelization floor, on real CUDA") as demo:
         "spend, unless your data already lives on the device and stays there.\n\n"
         "The same measurement runs in Colab on whatever GPU you are given, and the CPU "
         "half of the argument runs anywhere with no setup: "
-        "[jmurray10/peas_agents](https://github.com/jmurray10/peas_agents).\n\n"
+        "[jmurray10/agent_design_peas](https://github.com/jmurray10/agent_design_peas).\n\n"
         "Nothing here is a published figure and no model is called."
     )
     button = gr.Button("Measure", variant="primary")