[SergiioB]

Adds the Strata runtime and separate IQ3_XXS and IQ3_S quant boards for Qwen3.8-Flash-Next. Source artifacts: https://huggingface.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF/tree/ed59f92082b1e93c0e96d60a8b11aab089b52f09 . XeStrata is a Strata fork, not llama.cpp. These entries prevent incorrectly mapping IQ3 results to Q2_0. I withdrew my incorrectly classified result 240; replacement matched measurements will use the correct boards after catalog sync. Verification: catalog:validate passes (52 hardware, 51 models, 26 quants, 9 runtimes).

[vercel]

@SergiioB is attempting to deploy a commit to the Community Labs Team on Vercel.

A member of the Team first needs to authorize it.

[github-actions]

No submission files to import.

Site / sign up · Workflow details and retry

[SergiioB]

Checks are green on this one (test, results, validate, Result ingestion). It currently blocks two measured results at the API door: POST /api/v1/results/batch answers Unknown quant: iq3_xxs. Read /catalog for available IDs.

What is waiting to publish once this lands and the catalog sync runs:

quant	decode	coordinates
iq3_xxs	86.8 tok/s	C1, P34/G512, n=3 median, thinking off, 262K ctx
iq3_s	86.7 tok/s	same protocol, matched A/B

Evidence, raw responses and engine SHA256: https://github.com/SergiioB/intel-arc-pro-b70-inference-cookbook/blob/master/docs/qwen38-flash-next/QWEN38-FLASH-NEXT-XESTRATA.md

Context for the two additions: the GSQ-RCO Flash-Next files carry GGUF general.file_type 19/20 (IQ3_XXS / IQ3_S mixed tensors), so the existing q2_0 board is the wrong home for them — I withdrew result 240 for exactly that mislabel. The strata runtime row exists so XeStrata (a Strata fork, not llama.cpp) can carry its own badge instead of hiding under a custom llama.cpp runtime. Happy to split the PR if you would rather take the quant rows and the runtime separately.