Niko1221/Strata v0.2.1

Strata 0.2.1

on GitHub

Update 0.2.1.1 (2026-10-20): Windows Terminal fix. On Windows with long paths off (the default), the Terminal could not install an AMD engine: "The engine could not be installed. local file operation failed (FileNotFoundError)". The engine is now unpacked into a much shorter folder and with paths Windows accepts at any length, and a path that is still too long gets a plain message. If you use the Terminal on Windows, run strata update once; it installs Terminal 0.1.46.1 (Control), and strata launch works again. A new install gets it straight away. The app and the engines are unchanged.

0.2.1: a task board in the app, pictures on the Mac and on Intel Arc cards, strata bench run with a fixed standard benchmark, faster downloads, Docker images, and your own engine settings are no longer overwritten. It also fixes the crashes, stuck downloads and GPU-choice bugs reported on 0.2.0-beta. Each new default below says what it was measured on and whether the answers are identical to 0.2.0. The engine packages are version 0.1.46 (0.2.0 had 0.1.45.3) and the update feed has sequence number 102. Update with strata update (Terminal), Settings > Update (app), or UPDATE.bat / ./update.sh (source installs).

Which file to download

The Terminal installs with one line and picks its engine by itself (curl -fsSL https://github.com/Niko1221/Strata/releases/latest/download/install.sh | sh, or irm https://github.com/Niko1221/Strata/releases/latest/download/install.ps1 | iex on Windows). The other files in this release are the engine and Terminal packages, the signed update feed and the in-app update files; you do not need to download them. Docker images are separate (below).

What is new

• A task board in the app. On the Tasks screen, a Schedule | Board switch shows your scheduled tasks and loops as a board: what is running, what is waiting for your approval (from every chat), and what failed. Each run has one plain line, only the newest problem of a task is shown, and a problem can be dismissed. The sidebar shows how many tasks need you. All eight app languages are translated.
• Pictures on the Mac. The Mac app and Terminal now read pictures by default. The reader runs on the Mac's GPU (Metal), built into strata-vision and signed with the app. Measured on an M4 Mac: a 260-token picture 0.45 s on the GPU reader against 50.7 s on the CPU, a 972-token picture 2.4 s against 194 s; the GPU reader takes about 1.2 GB of memory. On the 972-token picture 8 of 972 picture tokens differ from the CPU reader (cosine down to 0.48, in the text area); the answers were correct. --vision cpu is still there and warns that it takes about 50 seconds per picture.
• Pictures on Intel Arc cards. Alchemist cards (A-series) read pictures by default through the CPU reader; Battlemage (B-series) stays opt-in (see Known limits). A new GPU picture reader for SYCL (--vision gpu) takes 0.16 s per picture on an A750 against 1.73 s on the CPU (its picture embeddings equal the CPU reader's to a cosine of 0.9959 to 0.9999). Ten varied pictures and a multi-picture chat all passed on the A750.
• strata bench run runs the standard benchmark. The default suite is standard-v1: fixed request files (34 of them, with a checksum list) in the Terminal archive, a scorer, a validator and a report that leaves out personal data. The Terminal archive now carries its commit, because a run without one does not count. The protocol text in docs/BENCHMARK_PROTOCOL.md is still a draft. See docs/COMMUNITY_BENCHMARKS.md for how to send a result.
• Faster downloads. Big model files come over 16 connections with several pieces per request (0.2.0 used 8 connections and a new connection for every piece). On the test PC, from Hugging Face, three interleaved pairs: 68.6 / 65.9 / 70.6 MB/s before, 74.4 / 74.8 / 74.9 MB/s now (+9%; the line itself tops out near 75 to 85 MB/s). The fetch of the draft layer (MTP) went from 25.0 / 28.1 / 37.6 MB/s to 47.4 / 50.5 / 52.2 MB/s, and its file list from 28 s to 4.6 s. ModelScope downloads, whose server reports no file length, now use several connections too. The files are the same and are still checked against their pinned SHA-256.
• Docker images. Ready-made images for ghcr.io/niko1221/strata:0.2.1-<flavor>: cuda13, cuda12, rocm-gfx120x (RX 9000, R9700), rocm-gfx110x (RX 7000) and rocm-gfx1151 (Strix Halo). They are built from the signed release packages by the repository's workflow. The start-up test passed on an RTX A4000 (cuda13 and cuda12) and on a Radeon AI PRO R9700 (rocm-gfx120x); the gfx110x and gfx1151 images were built but not run on a card. There is no Intel image yet. See docs/DOCKER.md. The workflow starts after this release is published, because it reads the release's own files, so the images appear a few hours later.
• Your own engine settings stay. The app and setup used to rewrite the run config on every Load. Options you set by hand (--vram-reserve-mib, control vectors, any option setup does not choose itself) are now kept in user_args, and what you set wins over what the app generates. See Fixes.
• Draft layer (MTP) switch. Settings > Expert has a Draft layer (MTP) on/off per model, and strata launch --mtp on|off does the same. A picture-off choice and a custom context size stick across loads.
• Fit checks count the draft layer. With MTP on, the draft layer takes about 1 GB of graphics memory; the model fit check and the expert-cache sizing now count it.
• The model page shows the engine's own pick for your PC (the memory tiers of 0.2.0), with Gyro-M named as the other choice at 40 GB.
• The chat footer speed is the writing speed. It divides output tokens by decode time, so a tool-call turn no longer shows "632 tok/s" when the engine ran at 50 to 75.
• A wrong download says so. An installer for another make of graphics card (an AMD build on an NVIDIA PC) tells you in plain words which file to get and does not unpack its engine. An installed engine with no model loaded is no longer reported as "No Strata engine".
• Terminal. A model that is already on disk says "already downloaded". strata monitor finds a running strata launch without --url.
• The standard AMD packages carry the hipBLASLt tuning tables and the engine start picks the one for the card (see New defaults for when it is used).

New defaults, with what we measured

Speed numbers are medians of interleaved pairs on the same box, a fresh engine per run; "pairs faster" counts the pairs the new default won. Each default has an off switch.

• Mini PCs with a Radeon 780M (gfx1103), Linux: the matrix-core prompt set is on. Measured on a Ryzen 7 8745HS with a Radeon 780M, 60 GB DDR5, Linux kernel 7.0.0-38, Qwen3.8-Flash-Next Q2_0 with MTP, 6 interleaved pairs: a 1K prompt 98.5 to 157.3 tok/s (x1.60, 6 of 6 pairs), a 4K prompt 115.1 to 233.3 (x2.03, 6 of 6), decode unchanged (x0.999 and x1.000). The greedy token ids equal the old default's in every pair (6 of 6, and 5 of 5 in the final build). Same answers, as far as the ids show. STRATA_IGPU_PROMPT_SET=0 turns it off. It applies to the 780M and 760M on Linux only; 880M, 890M, Strix Halo and Windows stay off. The 780M's GPU can still reset on its own firmware (see Known limits); with the set on we saw no more resets than with it off, in small samples.
• Strix Halo (gfx1151): the hipBLASLt tuning table is on. The packaged Strix Halo engines (Linux strixhalo and the Windows rdna3 package) use the table for their card when their hipBLASLt version matches it. The measurement is a community A/B from bench/results/2026-10-05-community-gfx1151 (Linux, ROCm 7.17, hipBLASLt 1.5.0): prompts 250.9 to 528.9 tok/s (2.11x). We have no Strix Halo and did not repeat it. The table changes which matrix-product algorithm reads the prompt, so the answers are not bit-identical to plain hipBLAS (on an R9700 the first-token KL was 0.016, against a run-to-run difference of 0.00004; top-1 the same on 24 of 24 prompts). STRATA_HIPBLASLT_TABLE=off turns it off. On every other card the table is opt-in (STRATA_HIPBLASLT_TABLE=on): on the R9700 a table calibrated for the packaged hipBLASLt gave about +1% prompt and -2% decode, so none is shipped as a default. The tables are packaged per card (tuning/ inside the engine package): Windows rdna3 gfx1100 and gfx1151, Windows hip (RX 9000, R9700) gfx1201, Linux rx7900 gfx1100, Linux strixhalo gfx1151; the Linux RX 9000 / R9700 package carries none, because no table matches its hipBLASLt. Only the gfx1151 table is used by default.
• Two NVIDIA cards, several requests at once: one batch group. With a layer split over exactly two NVIDIA cards and --batch N, the engine now runs one serial group instead of two pipelined groups. Measured on two RTX A4000 16 GB with Qwen3.8-Flash-Next IQ3_XXS, 5 interleaved pairs, aggregate tokens per second: 2 requests +15.9% (5 of 5 pairs), 4 requests +12.0% (5 of 5), 8 requests +14.7% (3 of 3). Every slot's tokens equal the solo run's, and the old default's, under the fixed identity flags (2 and 4 slots, 64 greedy tokens). A single request is not changed. It is not faster everywhere: when most hot experts sit in the CPU, as under the identity gate's small cache, the pipelined groups were about 10% faster (24.4 against 22.0 tok/s), so a card with a very small cache can lose here. STRATA_BATCH_GROUPS_NV_SERIAL=0 restores the old default; --batch-groups auto or a number asks for the pipeline. AMD cards, three or more cards and the single-request paths are unchanged. On the final engine, 4 requests, 5 pairs: 78.06 against 65.97 tok/s (+18.3%, the old default faster in 0 of 5), and every slot's tokens equal the solo run's.
• A card that draws no display keeps less in reserve. The reserve of graphics memory the engine leaves free is now 512 MiB (700 MiB with pictures) on a card the probe says drives no screen (a compute-only card), instead of 1536 MiB. A card that drives a display keeps the 1536 / 2048 MiB safety default, and a number you set yourself (0 included) is always used as it is. Only the amount of free memory changes, so the answers are identical and the expert cache gets room for more experts. We did not time this ourselves: the report behind it (#3215) measured 80 against 50 tok/s with the forced reserve. Where the probe cannot tell (AMD and Intel on Linux) the safety default stays.
• Packed models (experts.bin) on Linux: I/O prefetch and the decode gate are on when the page cache has room for at least 55% of the experts. Measured on an RTX 3060 12 GB (Core Ultra 7 265), Qwen3.8-Flash-Next IQ3_XXS 43 GB pack with --mmap-experts, RAM limited to 32 GB by cgroup, cold page cache, 5 interleaved pairs: decode story / code / document 26.0 / 28.3 / 18.5 to 35.2 / 34.4 / 24.7 tok/s (x1.38 / x1.21 / x1.33, 5 of 5 pairs each), a 4K prompt x1.04 (4 of 5), a 20K prompt x1.06 (5 of 5). Same answers: the fixed-flag identity gate (10 prompts, 4K and 20K exact prompt checks) is identical with no RAM limit and with the 32 GB limit. With 16 GB of RAM the page cache holds 38% of the experts, so the default stays off: with it on, a cold 4K prompt read 4 to 6% slower. STRATA_IO_PREFETCH=0 turns it off and =1 forces it on. Linux only; only the IQ3_XXS pack was tested (not Q2_0, IQ3_S or Coder packs).
• AMD RX 9000 / R9700 on Linux: the engine opens its GPU queues in a fixed order at start. Before, about half of the fresh engines with MTP read prompts 8 to 14% slower, depending on which compute pipe the card's firmware gave the engine's two busiest queues (both on one pipe was slow). The engine now opens two queues on its main thread before anything else starts. Measured on two Radeon AI PRO R9700 boxes (Linux, ROCm 7.10, Q2_0 with MTP): without the fix 10 of 32 fresh starts read the 4K prompt under 2,000 tok/s (about 2,290 for a fast start); with it 0 of 32. Decode against the fix switched off, 20 interleaved rounds: 1.000 / 1.001 / 1.000 (story / code / facts). Opening more queues cost decode (five queues: 2 to 3.6% slower), so two is the default. Same answers: the identity gate (10 prompts, 4K and 20K exact checks) is identical for Q2_0 and IQ3_XXS, also with the 0.2.0 prompt defaults on. STRATA_HIP_QUEUE_PRIME=0 turns it off. One card was measured; Windows and RX 7000 cards are not changed.
• Linux hybrid CPUs with low-power E cores (Core Ultra "Meteor Lake" / "Arrow Lake-H", #3262). The engine and setup now count the cores listed in cpu_lowpower as efficiency cores. A Core Ultra 7 255H used to look like 8 P and 8 E cores, so the hybrid plan did not apply and 15 workers ran; it now plans 10. The reporter measured the 15-worker default about 35% slower than 9. We have no such CPU and ran unit tests only (the plan and a fake /sys); only the number of workers changes, not the answers.
• Two GPUs of different architectures (HIP): no start-up warm-up (#3204). The warm-up product that runs on every card at start made prompts about 25% slower on a split of gfx1151 and gfx1201 for the reporter (STRATA_PREFILL_WARM=0 restored them). It now runs only when every visible card has the same architecture. We have no mixed pair; on two R9700 (same architecture) the warm-up costs nothing (3,855 against 3,853 tok/s at a 32K prompt, 5 pairs), so the change leaves them as they were. The first request on a mixed pair pays the 1 to 2.5 s library load again, as before 0.2.0. STRATA_PREFILL_WARM=1 / 0 overrides.
• Linux, models read in place: a drive probe decides between the file cache and unbuffered reads (#3237). The room rule picks the file cache when RAM seems to have room, which was 2x slower in prompts and 3 to 4x slower in decode for a reporter with a PCIe 5 drive (about 8 GB/s). When the rule says "cache" and the cache would hold under half of the reads, the engine now reads 96 random 1 MiB pieces both ways at start (under 2 s) and goes unbuffered only if the drive reads at least 4 GB/s and is faster that way. Every drive we have (1.6 to 3.1 GB/s) keeps the file cache; on a 2 GB/s NVMe the cache was better for decode (up to x1.6) and unbuffered reads better for prompts, the opposite of the report. Untested on a fast drive; the answers cannot change (only where reads go), and no identity gate was run. STRATA_FILE_TIER_PROBE=0 turns it off.

Fixes from user reports

• Settings overwritten (#3215, #3270, #3216). The app forced --vram-reserve-mib to 1536 (2048 with pictures) on every Load whatever your file said, which cut decode from 80 to 50 tok/s for one reporter, and a setup run that a Load started dropped every option it did not choose itself, control vectors included. Options set by hand are kept now and win over generated ones; an old file's hand edits are moved into user_args.
• Choosing graphics cards (#3203, #3195, #3256, #3221, #3211). Automatic no longer picks a card the engine has no code for (an AMD integrated GPU in front of a discrete card). One selected card is sent as a single card, not as a list the installer refuses. A card order from the settings file ("gpu": [1,0], "gpu_order": "as_given") is kept instead of being replaced by the app's own pick. The card list no longer ticks two cards with one click when two makes both count from 0. The dropdown rows fit long card names. The Monitor opens on the cards the model is assigned to.
• Downloads (#3257, #3214, #3193). The draft-layer download shows its progress in the app (it showed "0.0 / 0.0 GB"), runs again on a flaky link with fewer connections, and has one writer per folder. Cancelling, or a killed setup, now ends the whole process tree on Windows, so no orphan keeps writing. Gyro's low-RAM pack is written with --compat-bf16, which stopped the "command failed (exit 1): python.exe" at the packing step.
• Re-download of a model you already have (#3199). A model loaded from your own GGUF folder keeps that folder, so unloading and loading it again no longer downloads it a second time.
• FP4-Local setup died at step 6 (#3227). Setup now installs the pinned torch, transformers and safetensors the converter imports, before a prepare that needs them, and names a missing package at once instead of after 20 minutes.
• The model stopped while starting (#3192). A start that stopped because no graphics memory was left for the expert cache is tried once more with less kept free.
• Crash on launch (Windows 11, exit code 0xC0000409, no log). A failed start on Windows and Mac now leaves a log and a crash file in the app folder and shows a message box instead of closing. The reporter's own panic message is still unknown, so please send the new crash file if it happens again.
• MTP, context and pictures setting kept coming back. Every Load re-ran setup and rewrote them. A Load of an installed model is now direct, the picture-off choice sticks, a custom context size is shown and accepted, "MTP per request" is no longer frozen, and the new Draft layer switch is in Expert settings.
• Your model library no longer moves. strata launch --data-dir with a new folder used to move all existing models, packs and MTP files into it and rewrite your settings. It never moves them without asking now.
• Math in the chat (#3268). Formulas are drawn with KaTeX: $...$ and \(...\) inline, $$...$$ and \[...\] as blocks. KaTeX and its fonts ship with the app, load only when an answer has a formula, and work offline. Prices like "$5 and $10" and anything in code stay text. Checked with the formulas from the report in a browser.
• Windows with a non-UTF-8 code page, such as Russian (#3202). PowerShell output (the Documents folder, the save dialog, fonts, GPU names) was read in the wrong code page and came back as question marks; it is read as UTF-8 now. Python runs in UTF-8 mode from the launchers and the app's child processes. The engine and the picture reader carry a UTF-8 manifest, so a Cyrillic model folder arrives intact (Windows 10 1903 or newer). We reproduced the first two and tested the manifest with a small program; we could not run the engine on a Russian Windows, so a retest is welcome.
• Install folder (#3186, #3190). The Windows installer asks for the folder (another drive when C: is short). Updates and silent installs keep the earlier folder. The docs say what stays on C: (the engine bundle, settings and chats) and that STRATA_COMPONENTS_DIR moves the engine bundle. We compiled the page logic with NSIS on a reduced script; the full installer is built at release.
• Linux engine killed with signal 9 and no out-of-memory (#3238). A desktop with PipeWire and no rtkit can leave a finite RLIMIT_RTTIME. The server now warns at start when the limit is finite, and the engine-died message names it before the RAM hint. There is a troubleshooting entry.
• Tools find gguf-py in the Control package (#3269). tools/iq_pack.py also looks in the install folder (STRATA_INSTALL_DIR, the app folder); the error names STRATA_GGUF_PY.
• Intel SYCL build (#3219). sycl/tools/build.sh stops before the long compile when the ocloc of the base image is the one that crashes on the int8 kernels (Compute Runtime 26.22.38646.x, IGC 2.36.3) and says to docker build --pull; STRATA_SYCL_I8_ESIMD=OFF builds without those kernels (slower prompt experts, same answers). We could not reproduce the crash itself: a newer ocloc (26.31) builds the Battlemage target without errors.
• gfx906 (MI50, Radeon Pro VII) builds again (#3172, #3271). The Qwen3.6 engine is not built for it, the W4A4 target gets GGML_USE_HIP, and the trellis overlay and moe_f16_kq sources are in the HIP list. All 161 targets compile for gfx906 and strata links; we have no gfx906 card, so it was not run on one.
• Trellis overlay with newer llama.cpp (#3217). The overlay follows the signatures of the ggml-cuda it patches. On the pinned commit the generated files are byte-identical to before; against upstream 0ce8cd11 the old overlay failed to compile and the new one does.
• Settings text (#3208). "Allow browser origins" says it does nothing in the app (the app's API already allows pages of any site) and where cors_origins goes, in all eight languages.
• Docs. Build from git on Linux (#3191), the dual-socket CPU pool measurements, IOMMU check and no-numactl note (#3201; the NUMA-split pool from that report is not merged), and a two-card gfx906 tip with --pipeline-windows 2 (+36% decode) and --trim-stage-weights (+5% prompt), measured by the reporter and off by default (#3173).

Privacy

• Update checks run only when you click. Opening Settings, Engine and Update makes no network call; the last result is shown from your computer, and a new check starts only from the button. Offline, the check says "No internet connection. Strata works without it; try again when you are online."
• The assistant asks before it first searches the web or opens a web page. In Manual mode you are asked once per chat and the answer holds for that chat. Settings > Tools has a switch to skip the question (off by default), and a scheduled task cannot search unless you turned that on. Bypass mode does not ask.
• Starting, loading and chatting with a model on disk make no network calls. A test starts the host with every non-loopback request recorded and finds none. The old starter-model list can no longer download anything; it is display only.
• What Strata downloads and when is in docs/SAFETY.md.

Known limits

• Windows: the installers are not code-signed yet, so Windows shows "Windows protected your PC": choose More info, then Run anyway.
• Windows with an AMD card: nothing has run on a real AMD card on Windows. The gfx12 prompt defaults and the 780M prompt set are off there. The Windows rdna3 package (RX 7000 and Strix Halo) was built and scanned but not run on those cards by the maintainers.
• Radeon 780M: the GPU can reset with a firmware error (MES REMOVE_QUEUE) when the kernel moves memory the GPU uses. On our box the default engine lost 1 to 4 of 10 starts to it with or without the new prompt set, and the driver recovered each time. The firmware is the cause; we did not install new firmware.
• Intel Arc Battlemage (B-series): pictures stay opt-in. A first picture on a B70 once reset the card (xe driver); we could not repeat it on the A-series card and no B-series card was available for 0.2.1.
• Mac: the GPU picture reader was measured on an M4 only. The CPU reader is about 50 seconds per picture.
• Strix Halo: the hipBLASLt table default rests on a community measurement; please send the start-up line hipBLASLt tuning table: ... and a prompt run with the table on and off.
• Docker: the rocm-gfx110x and rocm-gfx1151 images were not run on a card, and there is no Intel image.
• AMD and Intel on Linux: the engine cannot tell whether a card drives a display, so the safety reserve (1536 MiB) stays unless you set one.
• Benchmark: strata bench run implements the draft protocol; a first real run on a release engine is still to come.
• Two NVIDIA cards, batch: a card with a very small expert cache can be slower with one group than with two (see New defaults).
• A100 (sm_80), #3252: a report of decode -10% and prompts -20% since 0.1.45 was not reproduced. On a Tesla P100 (sm_60) and two RTX A4000 (sm_86) the 0.1.45 build against 0.1.44.1 measured decode x1.015 and x0.992, prompts x1.004 and x1.022 (5 interleaved rounds each), and the 0.2.1 tip is 7 to 10% faster in decode than 0.1.45 on both. Reading the commits found no sm_80-specific change. We have no A100; the reporter's numbers are needed.
• Fast drives, #3237: the drive probe above is proposed and untested on a drive over 4 GB/s.
• NUMA-split CPU pool (#3201): not part of 0.2.1; we have no dual-socket box to measure it.

Thanks to H3buss, who found that the Gyro import was missing --compat-bf16 and where to put it (#3193); to bext1998, whose trace of the MTP download and the orphaned process (#3257) led to the progress and cancel fixes; and to Gazoo, who reported the card-selection, dropdown and re-download bugs (#3195, #3221, #3199). Thanks also to everyone who tested 0.2.0-beta and sent reports.

Strata is free and open source. If it runs well on your PC, a coffee keeps the work on it going: