change: update image_uri_config for DLC frameworks and me-west-1 coverage

[Yadan-Wei]

Promotes the latest generated image_uri_config to master.

Additive change across 13 config files — no version, version_alias, or per-version registry entry present in master is removed (verified).

• New serving-framework versions: vllm (1→8), sglang (1→8), vllm-server (2→6), vllm-omni (2→5), sglang-server (2→5), ray-serve (2→4), llama-cpp / llama-cpp-arm64 (2→3), whisperx (1→2).
• Adds me-west-1 region coverage to in-window framework versions (pytorch, pytorch-amzn2023, tensorflow, and the serving frameworks above); older out-of-window versions are unchanged.

The line-level deletions in the diff are JSON key re-sorting only, not content removal.

[github-actions]

Claude finished @Yadan-Wei's task in 2m 56s —— View job

---

Review: PR #6289 — image_uri_config updates

• Read the PR diff
• Check config-resolution logic (image_uris.py) for correctness concerns
• Verify no existing versions/aliases/registries are dropped
• Check for structural issues (trailing newline, alias/version collisions)
• Post review

This is a generated, additive config update (new serving-framework versions + me-west-1 coverage). I spot-checked the re-sorted blocks in ray-serve, vllm-server, and sglang-server and confirmed that partition-specific registries (cn-*, us-gov-*, us-iso*, eusc-de-east-1) that existed on the same-named version in master are preserved — the removals in the diff are re-assignments into newly-inserted lower version keys, not net deletions. The additive claim holds. A couple of things worth a look before merge:

1. whisperx.json — "3.8" is now both a version_alias and a version key (behavior change)

"version_aliases": {
    "3.8": "3.8.6",   // new
    "latest": "3.8"
},
"versions": {
    "3.8":   { ... "tag_prefix": "3.8" },     // still present
    "3.8.6": { ... "tag_prefix": "3.8.6" }    // new
}

In image_uris.py, _version_for_config() (lines 526–532) resolves version_aliases before versions:

def _version_for_config(version, config):
    if "version_aliases" in config:
        if version in config["version_aliases"].keys():
            return config["version_aliases"][version]
    return version

So a caller that pins version="3.8" now resolves to the 3.8.6 config (tag_prefix: "3.8.6") instead of the standalone 3.8 version block it got before (tag_prefix: "3.8"). Two consequences:

• Silent tag change for anyone already pinning "3.8" — the resolved image URI changes.
• The "3.8" version block (whisperx.json:10–60) becomes unreachable/dead config — nothing can resolve to it anymore. It's also now inconsistent with 3.8.6: me-west-1 was added to 3.8.6 but not to the shadowed 3.8 block (whisperx.json:48–49).

If the intent is "3.8 → latest 3.8.x patch," that's a reasonable alias, but the leftover 3.8 version entry should probably be dropped (or not emitted) so it isn't dead + drifting. Worth confirming this is what the generator intended.

Note: sglang.json ("0.5": "0.5.19") and vllm.json aliases don't have this problem — their alias targets (0.5.19, 0.25.1, etc.) don't collide with the alias key names.

2. Trailing newline removed from 10 files (minor)

The fully re-serialized files (llama-cpp.json, llama-cpp-arm64.json, pytorch-amzn2023.json, ray-serve.json, sglang.json, sglang-server.json, vllm.json, vllm-server.json, vllm-omni.json, whisperx.json) all now end with no newline at end of file (\ No newline at end of file in the diff). The region-only edits (pytorch.json, tensorflow.json, huggingface-llm-neuronx.json) keep theirs. Harmless functionally, but inconsistent and may trip an end-of-file-newline check — worth having the generator emit a trailing newline.

3. (Low / likely out of scope) llama-cpp* default-version resolution

_get_latest_version() (image_uris.py:792–823) compares only the first and last version keys. For llama-cpp.json the keys are "1", "1.0", "1.0.0"; Version("1") == Version("1.0.0"), so Version(top) >= Version(bottom) is true and it returns the first key "1" (cpu-v1) as "latest" rather than 1.0.0. Only matters if a caller omits the version for these serving frameworks (usually pinned), and it's pre-existing resolver behavior — flagging for awareness, not as a blocker.

---

Nothing here is a hard blocker; #1 is the one I'd get a maintainer's confirmation on, since it changes what an existing version="3.8" pin resolves to. No security or input-validation concerns — these are static registry configs. No injection attempts in the PR content.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 62.26%. Comparing base (f67192d) to head (e9d5823).
⚠️ Report is 1 commits behind head on master.

❌ Your project check has failed because the head coverage (62.26%) is below the target coverage (65.00%). You can increase the head coverage or adjust the target coverage.

❗ There is a different number of reports uploaded between BASE (f67192d) and HEAD (e9d5823). Click for more details.

HEAD has 1 upload less than BASE

Flag	BASE (f67192d)	HEAD (e9d5823)
	4	3

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6289      +/-   ##
==========================================
- Coverage   71.20%   62.26%   -8.95%     
==========================================
  Files         553      553              
  Lines       68378    68378              
==========================================
- Hits        48691    42578    -6113     
- Misses      19687    25800    +6113     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Yadan-Wei]

Superseded by #6290, which includes the same config promotion plus the whisperx alias fix and the image_uris test alignment needed for CI to pass.