prioritise mlx jaccl coordinator ip (en0 -> en1 -> non-TB5 -> other)

35be4c55c30bd180119bb45c49fec682921be8a3 · 2025-12-30 17:15:48 +0000 · Alex Cheema

Files touched

M src/exo/master/placement_utils.py

Diff

commit 35be4c55c30bd180119bb45c49fec682921be8a3
Author: Alex Cheema <alexcheema123@gmail.com>
Date:   Tue Dec 30 17:15:48 2025 +0000

    prioritise mlx jaccl coordinator ip (en0 -> en1 -> non-TB5 -> other)
---
 src/exo/master/placement_utils.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/src/exo/master/placement_utils.py b/src/exo/master/placement_utils.py
index f7bff6f8..04f5f7a9 100644
--- a/src/exo/master/placement_utils.py
+++ b/src/exo/master/placement_utils.py
@@ -385,13 +385,14 @@ def get_mlx_jaccl_coordinators(
     address in format "X.X.X.X:PORT" per node.
     """
     rank_0_node = selected_cycle[0]
-    logger.info(f"Selecting coordinator from rank 0 node: {rank_0_node.node_id}")
+    logger.debug(f"Selecting coordinator from rank 0 node: {rank_0_node.node_id}")
 
     def get_ip_for_node(n: NodeInfo) -> str:
         if n.node_id == rank_0_node.node_id:
             return "0.0.0.0"
 
-        for ip, _ in _find_connection_ip(n, rank_0_node, cycle_digraph):
+        ip = _find_ip_prioritised(n, rank_0_node, cycle_digraph)
+        if ip:
             return ip
 
         logger.warning(

← 31d4cd84 set KV_CACHE_BITS to None to disable quantized kv cache · back to Exo · Placement filters for tensor parallel supports_tensor, tens 283c0e39 →