Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
43ee84c
feat(inferencex_atom): add multinode params and scaling efficiency me…
hnimra-amd Jul 6, 2026
81eeebd
feat(inferencex_atom): run distributed ATOM serve across cluster ranks
hnimra-amd Jul 6, 2026
0738fcd
feat(inferencex_atom): add MI300X 2-node cluster and W1 perf_multi va…
hnimra-amd Jul 6, 2026
ad4f982
test(inferencex_atom): cover multinode job routing and config loading
hnimra-amd Jul 6, 2026
c148c6e
docs(inferencex_atom): document multinode cluster and variant params
hnimra-amd Jul 6, 2026
5523491
feat(orchestrator): install openssh-server when image lacks sshd
hnimra-amd Jul 6, 2026
fe6fe02
fix(inference): restore test_launch_container in inference_suite_life…
hnimra-amd Jul 6, 2026
d1e777d
fix(inferencex_atom): use ATOM-native multinode serve instead of vLLM…
hnimra-amd Jul 6, 2026
4925068
fix(inferencex_atom): isolate ATOM multinode argv from vLLM distribut…
hnimra-amd Jul 6, 2026
d7248b7
config(inferencex_atom): expand multinode sweep to 5 shapes x conc 16…
hnimra-amd Jul 6, 2026
4dfa5b0
feat(inferencex_atom): enforce multinode scaling thresholds in gates …
hnimra-amd Jul 7, 2026
239a496
fix(inference): repair syntax error in test_launch_container
hnimra-amd Jul 7, 2026
c33fe45
fix(inference): tolerate CollectReport in html_metric_table_row
hnimra-amd Jul 7, 2026
3b88011
refactor(inferencex_atom): rename suite from inferencex_atom_single
hnimra-amd Jul 7, 2026
9d384fc
docs(inferencex_atom): run make install before activating venv
hnimra-amd Jul 7, 2026
33b0e4a
style(inferencex_atom): fix ruff format and lint on multinode changes
hnimra-amd Jul 13, 2026
480b192
feat(inferencex_atom): add MI355X multinode cluster and perf_multi va…
hnimra-amd Jul 13, 2026
a541ad3
feat(inferencex_atom): add M4 vLLM parity suite and W1 variant
hnimra-amd Jul 7, 2026
b1b765b
refactor(inferencex_atom): rename vLLM parity suite to inferencex_ato…
hnimra-amd Jul 8, 2026
251baa3
feat(inferencex_atom): add vLLM parity multinode W1 variants
hnimra-amd Jul 13, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion cvs/core/orchestrators/container.py
Original file line number Diff line number Diff line change
Expand Up @@ -477,11 +477,16 @@ def setup_sshd(self):
"chown -R root:root /root/.ssh",
"bash -c 'chmod 700 /root/.ssh && chmod 600 /root/.ssh/*'",
"mkdir -p /run/sshd", # Create privilege separation directory for sshd
"bash -c 'which sshd > /dev/null 2>&1 || (apt-get update -qq && apt-get install -y -q openssh-server iproute2)'",
"/usr/sbin/sshd -p2224",
]

timeouts = {
"bash -c 'which sshd > /dev/null 2>&1 || (apt-get update -qq && apt-get install -y -q openssh-server iproute2)'": 120,
}

for cmd in ssh_setup_commands:
result = self.exec(cmd, timeout=10, detailed=True)
result = self.exec(cmd, timeout=timeouts.get(cmd, 10), detailed=True)
# Check if command succeeded on all hosts
for hostname, output in result.items():
if output['exit_code'] != 0:
Expand Down
34 changes: 34 additions & 0 deletions cvs/input/cluster_file/mi300x_atom_multi.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,34 @@
{
"_comment": "2-node MI300X example for inferencex_atom multinode (M5). Replace mgmt_ip / node_dict with lab IPs. head_node_dict.mgmt_ip should match the rank-0 (master) node VPC IP used in params.master_addr.",
"orchestrator": "container",
"username": "{user-id}",
"priv_key_file": "/home/{user-id}/.ssh/id_rsa",
"head_node_dict": {
"mgmt_ip": "{head-node-ip}"
},
"env_vars": {},
"node_dict": {
"{head-node-ip}": {
"bmc_ip": "NA",
"vpc_ip": "{head-node-ip}"
},
"{worker-node-ip}": {
"bmc_ip": "NA",
"vpc_ip": "{worker-node-ip}"
}
},
"container": {
"lifetime": "per_run",
"image": "rocm/atom-dev:latest",
"name": "inferencex_atom_mi300x_multi",
"runtime": {
"name": "docker",
"args": {
"network": "host",
"ipc": "host",
"privileged": true,
"shm_size": "128G"
}
}
}
}
2 changes: 1 addition & 1 deletion cvs/input/cluster_file/mi300x_atom_single.json
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
{
"_comment": "Single-node MI300X example for inferencex_atom_single (ATOM driver). Replace mgmt_ip / node_dict with lab IPs.",
"_comment": "Single-node MI300X example for inferencex_atom (ATOM driver). Replace mgmt_ip / node_dict with lab IPs.",
"orchestrator": "container",
"username": "{user-id}",
"priv_key_file": "/home/{user-id}/.ssh/id_rsa",
Expand Down
34 changes: 34 additions & 0 deletions cvs/input/cluster_file/mi355x_atom_multi.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,34 @@
{
"_comment": "2-node MI355X example for inferencex_atom multinode (M5). Replace mgmt_ip / node_dict with lab IPs. head_node_dict.mgmt_ip should match the rank-0 (master) node VPC IP used in params.master_addr.",
"orchestrator": "container",
"username": "{user-id}",
"priv_key_file": "/home/{user-id}/.ssh/id_rsa",
"head_node_dict": {
"mgmt_ip": "{head-node-ip}"
},
"env_vars": {},
"node_dict": {
"{head-node-ip}": {
"bmc_ip": "NA",
"vpc_ip": "{head-node-ip}"
},
"{worker-node-ip}": {
"bmc_ip": "NA",
"vpc_ip": "{worker-node-ip}"
}
},
"container": {
"lifetime": "per_run",
"image": "rocm/atom-dev:nightly_202606211542",
"name": "inferencex_atom_mi355x_multi",
"runtime": {
"name": "docker",
"args": {
"network": "host",
"ipc": "host",
"privileged": true,
"shm_size": "128G"
}
}
}
}
2 changes: 1 addition & 1 deletion cvs/input/cluster_file/mi355x_atom_single.json
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
{
"_comment": "Single-node MI355X example for inferencex_atom_single (ATOM driver). Pin image to match variant config / ATOM CI run.",
"_comment": "Single-node MI355X example for inferencex_atom (ATOM driver). Pin image to match variant config / ATOM CI run.",
"orchestrator": "container",
"username": "{user-id}",
"priv_key_file": "/home/{user-id}/.ssh/id_rsa",
Expand Down
Loading