From d39b8ab55d6d2da65c2fa82b8ea39cd3f5d25be6 Mon Sep 17 00:00:00 2001 From: Yong Wu Date: Thu, 26 Jun 2025 12:14:24 -0500 Subject: [PATCH] Update autoscalers --- terraform/vars/tvm-ci-prod.auto.tfvars | 45 ++++++++++++++++++++++++++ 1 file changed, 45 insertions(+) diff --git a/terraform/vars/tvm-ci-prod.auto.tfvars b/terraform/vars/tvm-ci-prod.auto.tfvars index 3987262..4118674 100644 --- a/terraform/vars/tvm-ci-prod.auto.tfvars +++ b/terraform/vars/tvm-ci-prod.auto.tfvars @@ -47,6 +47,15 @@ autoscaler_types = { on_demand_percentage_above_base_capacity = 100 on_demand_base_capacity = 0 } + "Prod-Autoscaler-Jenkins-CPU-Large-SPOT" = { + image_family = "jenkins-stock-agent-x64" + agent_instance_type = "c4.4xlarge" + labels = "CPU-LARGE-SPOT" + min_size = 0 + max_size = 90 + on_demand_percentage_above_base_capacity = 100 + on_demand_base_capacity = 0 + } "Prod-Autoscaler-Jenkins-CPU-SPOT" = { image_family = "jenkins-stock-agent-x64" agent_instance_type = "c4.4xlarge" @@ -155,6 +164,42 @@ autoscaler_types = { on_demand_percentage_above_base_capacity = 0 on_demand_base_capacity = 0 } + "Flashinfer-Autoscaler-Jenkins-GPU-G5" = { + image_family = "jenkins-gpu-agent-x64" + agent_instance_type = "g5.xlarge" + labels = "GPU-G5" + min_size = 0 + max_size = 64 + on_demand_percentage_above_base_capacity = 100 + on_demand_base_capacity = 0 + } + "Flashinfer-Autoscaler-Jenkins-GPU-G5-SPOT" = { + image_family = "jenkins-gpu-agent-x64" + agent_instance_type = "g5.xlarge" + labels = "GPU-G5-SPOT" + min_size = 0 + max_size = 64 + on_demand_percentage_above_base_capacity = 100 + on_demand_base_capacity = 0 + } + "Flashinfer-Autoscaler-Jenkins-GPU-G6" = { + image_family = "jenkins-gpu-agent-x64" + agent_instance_type = "g6.xlarge" + labels = "GPU-G6" + min_size = 0 + max_size = 64 + on_demand_percentage_above_base_capacity = 100 + on_demand_base_capacity = 0 + } + "Flashinfer-Autoscaler-Jenkins-GPU-G6-SPOT" = { + image_family = "jenkins-gpu-agent-x64" + agent_instance_type = "g6.xlarge" + labels = "GPU-G6-SPOT" + min_size = 0 + max_size = 64 + on_demand_percentage_above_base_capacity = 100 + on_demand_base_capacity = 0 + } } ecr_repositories = [