Add p6-b300 and p6e-gb200 GPU health-check instance profiles

[Zhenye-Na]

Purpose

The GPU health check compares each node's real hardware against a per-instance profile in instance-profiles.conf. If an instance type isn't in that file, the loader falls back to a GPU/EFA count of 0, and every check guarded by > 0 (GPU count, EFA enumeration, topology) just skips. So the suite happily reports PASS on an unrecognized node while barely checking anything.

The file currently stops at p6-b200, which means the newer Blackwell nodes fall into exactly that gap.

Changes

Added profiles for two Blackwell-generation instances:

• p6-b300.48xlarge → 8|16|true|efa — 8 GPUs, NVLink. The AWS EFA docs list 17 network cards: NCI 0 is ENA-only and NCI 1–16 are EFA-only, so 16 EFA devices.
• p6e-gb200.36xlarge → 4|4|true|efa — 4 GPUs (2× GB200 superchips), NVLink (NVL72 domain). GB200's attached EFA count is provisioning-dependent; 4 is the AWS-recommended max-EFA layout (4 × 400 Gbps). Called out in an inline comment.

I also added the AWS doc links to the file header so the counts are traceable.

This is a data-only change — no code touched, existing rows untouched, so nothing changes for already-supported instances.

Source: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/efa-acc-inst-types.html

Test Plan

It's a config change, so I checked that the rows are well-formed and that the loader can actually find them.

# Every data row has the expected 5 pipe-delimited fields
awk -F'|' '/^[a-z]/ {if (NF!=5) print "BAD: "$0}' \
  4.validation_and_observability/2.gpu-cluster-healthcheck/instance-profiles.conf

# The new types resolve via the same exact-match lookup load_instance_profile() uses
for t in p6-b300.48xlarge p6e-gb200.36xlarge; do
  awk -F'|' -v inst="$t" '$1==inst {print}' \
    4.validation_and_observability/2.gpu-cluster-healthcheck/instance-profiles.conf
done

Test Results

Both new types resolve to a non-zero profile, so the GPU/EFA/topology checks now run on these nodes instead of skipping:

p6-b300.48xlarge|8|16|true|efa
p6e-gb200.36xlarge|4|4|true|efa

I don't have p6-b300 / p6e-gb200 hardware on hand, so a live run of the suite on real nodes would be a good follow-up for anyone with fleet access.

Directory Structure

N/A — updates an existing config file, not a test case.

Checklist

• I have read the [contributing guidelines](https://github.com/awslabs/awsome-distributed-training/blob/main/CONTRIBUTING.md).
• I am working against the latest main branch.
• I have searched existing open and recently merged PRs to confirm this is not a duplicate.
• The contribution is self-contained with documentation and scripts.
• External dependencies are pinned to a specific version or tag (no latest). — N/A, no dependencies
• A README is included or updated with prerequisites, instructions, and known issues. — N/A, config-only
• New test cases follow the expected directory structure. — N/A, not a test case