#2053 Community benchmark: RTX 5070 Ti 16 GB / 64 GB RAM, v0.1.43 IQ2_XS and IQ3_XXS

Description

Adds Sonny's Labs community measurements for **Strata v0.1.43** (`7695859aa13afb248bdb4d36f9b2e95707db079d`) on one **RTX 5070 Ti 16 GB**, Ryzen 5 5600G, **64 GB installed RAM** (58.8 GiB OS-usable), PCIe Gen3 x16 and NVMe.

Both **IQ2_XS** and **IQ3_XXS** have standard 32,768-context and optional 65,536-context (`--long --clients 2`) reports from the unmodified upstream tool. All eight original README/results files are byte-identical to the generated outputs, with verification hashes; separate supplements provide source/image/model pins, hardware, settings and cache-state details. Engine compiled before runtime using CUDA 13.0.0 / SM120; LOW_RAM resident mode, int8 KV and requested spec 4.

Standard chat decode medians: **81.9 / 66.4 tok/s** (IQ2 / IQ3). Standard fresh 16K prefill: **2176.8 / 1945.4 tok/s**. Three measured repeats and original ranges/per-request data are included. Warm process/filesystem; no storage-cold claim. Two-client results include queueing on one serial engine slot. Official speed outputs do not establish correctness; separately scored workloads and further limitations are linked in the [technical report](https://report5.firetussin.com/).

Validation: original-output SHA256 agreement, all submission checksum entries verified, sensitive-identifier/credential-pattern review, and results-only diff review. No engine/source changes, model files, packs, container archives or large telemetry. Hardware/software descriptions remain for reproducibility. Generated upstream RAM wording is clarified in the umbrella report without changing original files.