Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
C++
Release notes 0.2.2: DRAFT paragraphs for the merged streams
Niko12214aabf3b8f5b4
diff --git a/docs/release-notes-0.2.2.md b/docs/release-notes-0.2.2.mdindex 77499fea9..e9f04218c 100644--- a/docs/release-notes-0.2.2.md+++ b/docs/release-notes-0.2.2.md@@ -1,6 +1,17 @@-**0.2.2 (draft).** Only the items of the `f-022-rx6000` and `f-022-numa` branches are written here; the coordinator adds the rest.+**0.2.2 (DRAFT, not released).** Written from the notes of each merged stream; the coordinator edits it before the release. Items of the `f-022-iopf`, `f-022-flow`, `f-022-pkgspeed` and `f-022-modelsdir` branches are not merged and are not listed.## What is new- **An engine package for the RX 6000 series (built, not tested on a real RX 6000 card).** Linux has `rx6000` and Windows has `rdna2`, each with GPU code for gfx1030 and gfx1031 (RX 6800, 6900 and 6950 series, RX 6700 XT). They are offered to a PC whose card reports one of those two and to no other: the RX 7000, RX 9000 and Strix Halo packages are unchanged, and a PC with an RX 7000 or 9000 card next to an RX 6000 card keeps the package it had. The engine runs plain kernels on these cards (RDNA2 has no matrix cores, so none of the RX 7000 / RX 9000 or Radeon 780M prompt paths apply) and plain hipBLAS (hipBLASLt ships no kernels for gfx1030, so there is no tuning table). Setup and the app say once that this package was built and not tested on a real RX 6000 card. What was checked: the Linux package builds, every program in it has code for exactly gfx1030 and gfx1031, the rocBLAS kernel files for both are in it, every library it imports is in the package or belongs to the system, and the whole engine (all targets, tests included) compiles for gfx1030 and gfx1031. What was not: a model on an RX 6000 card, on either system, and the Windows package itself (the Windows build scripts take the new architecture list, but no Windows `rdna2` package has been built). Earlier reports on these cards came from people who built the engine from source ([docs/AMD_HIP.md](AMD_HIP.md#rdna2-gfx1030)). If something goes wrong, open an issue with the end of `strata-<model>.log`.- **The CPU expert pool splits per NUMA node on a machine with two or more NUMA nodes (built from the report in #3201, not tested on a two-socket machine).** Until now the pool kept its workers on the host thread's socket, because workers on the other socket read the experts across the link and were slower. The split gives each socket its own slice of the pool with its own task counter, and places each socket's half of every CPU-side expert in that socket's memory, so a slice reads local memory only. The reporter measured +10 to +14 % decode on a dual Xeon Gold 6226 (59.3 against 54.05 tok/s with the IOMMU in passthrough); a second reporter's dual Xeon E5-2670 v2 preferred more workers across both sockets, which the split also uses. It starts by itself when the machine has two or more NUMA nodes with CPUs and runs only when it can (Linux, two nodes, pinned threads, enough free memory per node); otherwise the pool is exactly what it was and the log says why. **A machine with one NUMA node runs the same code path as 0.2.1:** the identity gate (10 prompts, exact_pp 4K and 20K) is byte-identical and the single-request speed A/B is not slower on the one-node test boxes (numbers in the notes of the branch). `STRATA_NUMA_POOL=0` turns the split off; `STRATA_NUMA_SPLIT=1` still asks for it on any machine. The start-up log says what it did: `NUMA pool on: 2 NUMA nodes found ...` or `NUMA pool off: <reason>`. On Windows the node count is read but the split itself is Linux-only for now, so a two-node Windows PC logs the reason and runs as before. The logic runs in unit tests on made-up two-, three- and four-node machines; we have no two-socket machine, so the placement and the speed on real ones are the reporters' numbers, not ours. Do not wrap the server in `numactl --cpunodebind` or `--membind`: it pulls both halves onto one node. Thanks to baptisterajaut for the branch this is built on, and to xrip and Tworks25 for their numbers. Please send the start-up lines and a decode run with and without `STRATA_NUMA_POOL=0` from a two-socket machine.+- **The API key is optional.** The server, `strata launch`, the Terminal and the app no longer refuse to share without a key. Say so (`--no-api-key`, or `"api_key": "none"` in the server config) and it starts, with one plain warning that anyone who can reach the port can use the model and its tools; a public Cloudflare address gets a stronger one. The Terminal and the app still offer to make a key and do by default. In the app, Remove key now also works while local sharing or the tunnel runs. Fixes the refusals reported in #3198 and #3401.+- **Your own GGUF files load again.** "Use a local GGUF", `strata launch --model-path` and `strata models import` now accept what `setup.py --gguf-dir` always accepted: a folder with the files, the folder above it (a whole Hugging Face download, or an old Strata data folder), or one `.gguf` file. Unsloth sets whose first file holds no tensors, linked files (Hugging Face cache, Ollama) and folders with several sizes work; when a file is unusable, the message names the file and says why. Setup no longer deletes your own file when its checksum differs. The download and prepare steps use plain words, the checksum pass shows its own progress instead of a stuck 99 % download bar, and a model folder you pick (a whole drive becomes `D:\Strata`) is accepted by the installer whatever it is called. Reports #3316, #3292, #3190.+- **Problems with the computer's own setup are repaired or explained, not refused.** If the Windows runtime folder lists another account, Strata locks it down itself instead of stopping. A Documents folder that cannot be used (Controlled folder access, an offline OneDrive) falls back to the Strata home folder and Settings > Storage says so. A port that is taken or reserved gets one sentence with the next step. Local requests ignore a system proxy. File-in-use errors and a full disk say what happened. `strata launch` asks plain questions in a terminal that reports a tty late. `strata update` copes with a slow `--version`, a busy graphics card is not held against an engine, pre-release Control builds hand over correctly, and a wrong PC clock is named as the reason a feed or certificate looks bad. install.sh and install.ps1 give clearer messages (ARM, TLS 1.2, proxy credentials, no `tar.exe`).+- **One speed number everywhere.** Answer speed is every token the model produced (thinking, answer, tool calls; with MTP the kept tokens) divided by decode time. The chat footer, the monitor, the model API and the server console used different definitions: an agent turn read about 632 tok/s in 0.2.0-beta, then fell to the real 50 to 75 in 0.2.1, and some places divided by the whole request, which reads about half. The footer now shows the last reply's own speed, the popover has a "Prompt reading" row, and the console no longer prints a million tok/s for a one-token reply.+- **`strata bench run`, first real runs.** Run on two test PCs with the 0.2.1 release, it showed that the release archive carried no commit (every run was `nonconforming`), that the drive lookup failed on LVM, that the CUDA runtime was missing from the nvidia-smi 615 output, and that the engine log in the result was empty. All four are fixed, the protocol text now says 2 to 4 minutes for the required part, and results get a Capped column for answers that stopped at the token limit. Package building refuses to make a distributable archive without a commit.+- **`strata update` takes effect on the next command.** When an update installs a newer Control, the Terminal hands over to it by itself. Terminals 0.1.46 and 0.1.46.1 cannot do this (the code is in the Terminal): reinstall the Terminal once.+- **A stale update notice goes away.** After a version change the app and the Terminal no longer keep showing an offer for a version you already have.+- **Idle GPU time was measured, with no default change.** On a two-card test PC decode leaves about half of the card time idle (batch 51 %), on a one-card PC 5 %. Overlap gave +0.3, +2.1 and -0.3 % against 2.8 % noise; pipeline windows and batch groups help in one case and cost 5 to 15 % in others, so nothing is switched on. Raw numbers are in `bench/results/2026-10-20-0.2.1-idle-gpu/`.+- **Small fixes.** `--mtp 2` is no longer checked as a folder (#3298); owner-only checks understand Windows account aliases (#3302, thanks jackbapa); `strata bench` with its own server sets its port (#3188); `calibrate` checks the report folder first (#3187); the CUDA 12.0 and 12.1 build error `__half_as_ushort` (#3368, thanks yudhanjaya, built with CUDA 12.6 only); the AMD picture reader builds position independent (#3359); the Mac setup installs cmake and ninja before it builds (#3403); the picture reader recreates its work folder if a temp cleaner removed it (#3370); a list of GPUs such as `"0,1"` is accepted (#3275, thanks MEsfits); `thinking_budget_tokens` is read (#2141) and the Anthropic usage numbers add up to the prompt (#2289, thanks xrip); an engine that cannot start is a 503 with the reason (#2496, thanks devYRPauli); the serve web page no longer scrolls sideways on a phone (#2971, thanks ctr-137); folders in a Trash or Recycle Bin are not offered as installs (#3410). Ported pull requests keep their authors.+- **From the issues labelled needs-format.** The app numbers Radeon cards the way HIP does when an integrated GPU is present (#3418). The hardware report counts physical cores, so the Work profile asks for 6 pool workers on an 8-core CPU instead of 12 (#3417; the reporter measured 27.9 against 38.1 tok/s). Three trellis CPU tests skip on a ggml without trellis types, and the ISA guard knows the AVX-512 files (#3358, tests and build only; the memory fault on gfx1011 is not touched).+- **Docker image build.** The terminal stage no longer needs git in the image (it failed with `FileNotFoundError`); the commit comes in as a build argument.