[jayleaton]

What this changes

The public Flash-1017 EXL3 pack declares quantization_config as dsq-1 metadata with a producer annotation, so main refuses it as UnsupportedCheckpoint before reading any tensor. This admits the supported format independently of the producer name. Geometry, tensor-layout, scale and marker checks stay as they are.

Regression fixtures: DeepSeek's public 1017 configuration, and the pack's published metadata. Host admission tests cover the real Config.read metadata path. Malformed-storage tests check the native refusals.

Issue: none open upstream. Stacked on: #1640 (asset generators), whose Config.parseMetadata the admission test uses. Review only ddde40de8812..69537d14db7c.

Receipt

• Environment, with the TensorFold commit, MLX or PyTorch version, machine and GPU, checkpoint and revision: head 69537d14db7c on main 3f0a715c4; Zig 0.17.0; host checks on a Linux aarch64 DGX Spark (GB10, CUDA 13.0). Real-pack boot on two GB10s, TP=2 over RoCE, checkpoint jayleaton/DeepSeek-V4.1-Flash-1017-EXL3-2.8bpw revision 859396591c34, TF_EXPERIMENTAL=deepseek.
• Exactness: n/a; only admission changes.
• Decode speed before and after: n/a; main refuses this checkpoint.
• Prompt speed before and after: n/a; main refuses this checkpoint.
• Real-pack evidence: the series' TP=2 boot below reads this pack's metadata through the new path.
• Tests run, with pass, fail and skip counts:

  zig fmt --check build.zig zig                      exit 0
  git diff --check 3f0a715c4..HEAD                    exit 0
  python3 tools/zig/lean_check.py                     base 425 problems, head 425, new 0 (lean_new.py)
  zig build test-cuda-host --summary all              exit 0, 271/271 steps, 3906/3915 tests passed, 9 skipped (with test-server-cpu)
  zig build test-dsv41-shared test-dsv41-state test-dsv41-forward test-dsv41-kv test-dsv41-device-host test-dsv41-draft test-dsv41-cuda-host test-draft-dsv41-host test-dsv41-assets-host test-checkpoint-security  exit 0, 62/62 steps, 622/624 tests passed, 2 skipped
  control (neg-dsq, fix reverted, tests kept): zig build test-dsv41-shared  exit 1 (fails, as required)
  

• Real-pack TP=2 boot of main + every PR in this series (and the DeepSeek image path that is not yet proposed), at 83ba53298924: passed. Both ranks reached readiness; the recipe smoke, the canary and a strict tool call passed; MissingTritonVariant 0 in both rank logs; 2048 greedy tokens in 51.1 s with token_sha256 7538b20557b1…, the same as the downstream build of this series.
• Not run, and why: Metal.

Checklist

• Every token still goes through the lane rounds. No serial path, and nothing that needs drafts off.
• Drafted output equals "draft": false, a resumed prompt equals a fresh one, and concurrent equals solo. (Not measured for this change alone.)
• No precision traded for speed. If the bits change, the description says which sums change and why.
• A new or changed GPU kernel is registered with the kernel contract (docs/kernel-contract.md); each backend's SUMMARY line is in the receipt. (No kernel changes.)
• Prompt processing is no slower than the last release. (Admission only.)
• The description says which platforms I ran, Metal M1 to M5 and CUDA, and which I could not.
• New tests fail before the change, pass after it, and skip cleanly without their dependency. (Control: the old admission restored with the tests kept fails test-dsv41-shared.)
• Comments and docstrings are one line. No measurements or history in the source.
• No personal data, machine names, internal hosts or local paths. No AI attribution lines.

[ashhart]

Thanks, this looks good. Keeping the dsq-1 check to the format identifier plus the codebook and output-scale policy, and leaving the trellis, scale and marker checks with the loader, is the right split, and the refusal and legacy tests cover it well. It builds on #1640 (its admission test uses Config.parseMetadata), so we'll land it straight after #1640 rather than on its own. Nothing is needed from you for now; we'll follow up here if #1640 needs any changes.