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++
Strix Halo: the DeepSeek engine follows --feed, and the unified plan counts the BIOS carve-out
Niko1221932bda54bdc9
diff --git a/README.md b/README.mdindex 0fd9f05d0..be8210e93 100644--- a/README.md+++ b/README.md@@ -1,4 +1,4 @@-<p align="center">+<p align="center"><picture><source media="(prefers-color-scheme: dark)" srcset="apps/host/public/brand/strata-lockup-dark.svg"><img src="apps/host/public/brand/strata-lockup-light.svg" alt="Strata" width="340">@@ -9,9 +9,9 @@<p align="center">Chat, code and work with images, locally.</p><p align="center">- <a href="#models">Models</a> · <a href="#install"><b>Install</b></a> ·- <a href="#hardware">Hardware</a> · <a href="#tools">Tools</a> ·- <a href="#help">Help</a> · <a href="#support-strata">Support</a>+ <a href="#models">Models</a> · <a href="#install"><b>Install</b></a> ·+ <a href="#hardware">Hardware</a> · <a href="#tools">Tools</a> ·+ <a href="#help">Help</a> · <a href="#support-strata">Support</a></p><p align="center">@@ -20,7 +20,7 @@<a href="https://buymeacoffee.com/strataengine"><img src="https://img.shields.io/badge/Buy_me_a_coffee-FFDD00?logo=buymeacoffee&logoColor=000" alt="Support Strata on Buy Me a Coffee"></a></p>-<p align="center"><sub><b>English</b> · <a href="README.zh-CN.md">简体中文</a> · <a href="README.ja.md">日本語</a> · <a href="README.de.md">Deutsch</a> · <a href="README.fr.md">Français</a> · <a href="README.es.md">Español</a> · <a href="README.pt-BR.md">Português</a></sub></p>+<p align="center"><sub><b>English</b> · <a href="README.zh-CN.md">简体ä¸ć–‡</a> · <a href="README.ja.md">日本語</a> · <a href="README.de.md">Deutsch</a> · <a href="README.fr.md">Français</a> · <a href="README.es.md">Español</a> · <a href="README.pt-BR.md">PortuguĂŞs</a></sub></p>**Free, open source, and yours to run.** Use Strata on your desktop or with your favorite coding tools. No account needed.@@ -33,9 +33,9 @@ We recommend **ISTA-DASLab first**, then **Unsloth** and other providers, where| Provider | Choices || --- | --- |-| **ISTA-DASLab · recommended** | Q2\_0, IQ2\_XS, IQ3\_XXS, IQ3\_S. [Coder: IQ1\_M](docs/MODELS.md#coder) for coding. |+| **ISTA-DASLab · recommended** | Q2\_0, IQ2\_XS, IQ3\_XXS, IQ3\_S. [Coder: IQ1\_M](docs/MODELS.md#coder) for coding. || **Unsloth** | UD-IQ4\_XS, UD-Q4\_K\_XL, UD-Q5\_K\_XL, UD-Q6\_K\_XL, Q8\_0 |-| **Gyro · agentionai** | [Gyro-S, Gyro-M](https://huggingface.co/agentionai/Qwen3.8-Flash-Next-Gyro-GGUF). Lower memory use but slower. |+| **Gyro · agentionai** | [Gyro-S, Gyro-M](https://huggingface.co/agentionai/Qwen3.8-Flash-Next-Gyro-GGUF). Lower memory use but slower. || **Other options** | [Swift 1.5](docs/MODELS.md#swift-15), [NVFP4](docs/NVFP4.md), [OrcaRouter](docs/ORCA.md) |[Compare sizes, memory needs and compatibility](docs/MODELS.md).@@ -63,7 +63,7 @@ disk space.| 2.00 bpw | 257 GB | about 265 GB || 2.25 bpw | 274 GB | about 280 GB |-Runs on NVIDIA RTX 30, 40 and 50 series (Windows and Linux), AMD Radeon AI PRO R9700 (Linux), and AMD Ryzen AI Max — Strix Halo+Runs on NVIDIA RTX 30, 40 and 50 series (Windows and Linux), AMD Radeon AI PRO R9700 (Linux), and AMD Ryzen AI Max — Strix Halo(Windows). Setup downloads the files and the DeepSeek engine, then starts the model.```sh@@ -78,9 +78,9 @@ Measured with the 2.00 bpw set on Strata 0.2.5 (Strix Halo: 0.2.7). Output is th| RTX 3060 12 GB, 46 GB RAM | 9 to 11 tokens/s | about 155 tokens/s (4K) || 1x Radeon AI PRO R9700, 251 GB RAM | 21 to 25 tokens/s | 320 to 360 tokens/s (8K) || 4x Radeon AI PRO R9700 | 74 to 77 tokens/s | 400 to 450 tokens/s (8K) |-| Ryzen AI Max+ 395 (Strix Halo), 64 GB shared | 14 to 16 tokens/s (code), 10 to 11 (chat) | 78 tokens/s (4K), 109 tokens/s (8K) |+| Ryzen AI Max+ 395 (Strix Halo), 128 GB unified | 14 to 16 tokens/s (code), 10 to 11 (chat) | 78 tokens/s (4K), 109 tokens/s (8K) |-[Memory, disk and details](docs/DEEPSEEK41.md) · [Release notes](https://github.com/Niko1221/Strata/releases)+[Memory, disk and details](docs/DEEPSEEK41.md) · [Release notes](https://github.com/Niko1221/Strata/releases)</details>@@ -94,12 +94,12 @@ Not sure what fits? [Choose a model for your computer](docs/MODELS.md#pick-by-me### Desktop app-**[Download Strata for Windows, Linux or Mac →](https://github.com/Niko1221/Strata/releases/latest)**+**[Download Strata for Windows, Linux or Mac →](https://github.com/Niko1221/Strata/releases/latest)**Install the app, choose a model, and start chatting. There is one installer for each system. On first start the appdownloads the engine for your graphics card. [Help with the installer](docs/INSTALL.md#the-app).-<sub>Files: `Strata-0.2.6-windows-x64-setup.exe` · `Strata-0.2.6-linux-x86_64.AppImage` · `Strata-0.2.6-macos-arm64.dmg`</sub>+<sub>Files: `Strata-0.2.6-windows-x64-setup.exe` · `Strata-0.2.6-linux-x86_64.AppImage` · `Strata-0.2.6-macos-arm64.dmg`</sub><p align="center"><img src="docs/media/app-chat.png" width="900" alt="Strata desktop app with a chat and live code preview side by side">@@ -144,7 +144,7 @@ Follow the prompts to download and start a model. [Terminal guide](docs/TERMINAL<a id="build-from-source"></a>-[Build from source](docs/SOURCE_INSTALL.md) · [Docker](docs/DOCKER.md) · [Full install guide](docs/INSTALL.md)+[Build from source](docs/SOURCE_INSTALL.md) · [Docker](docs/DOCKER.md) · [Full install guide](docs/INSTALL.md)<a id="what-you-need"></a>@@ -154,16 +154,16 @@ Follow the prompts to download and start a model. [Terminal guide](docs/TERMINAL| --- | --- || **NVIDIA GeForce RTX 20 / 30 / 40 / 50 series** | [NVIDIA setup](docs/INSTALL.md#what-you-need) || **AMD Radeon RX 7700 / 7800 / 7900, RX 9060 / 9070, AI PRO R9700** | [AMD setup](docs/AMD_HIP.md) |-| **AMD Radeon RX 6800 / 6900 · experimental** | [RX 6000 setup](docs/AMD_HIP.md#rdna2-gfx1030) |-| **Intel Arc · Linux** | [Intel setup](docs/INTEL_ARC.md) |-| **Apple Silicon · M1 or newer** | [Mac setup](docs/MAC.md) |-| **AMD Ryzen AI Max · Strix Halo** | [Strix Halo setup](docs/STRIX_HALO.md) |+| **AMD Radeon RX 6800 / 6900 · experimental** | [RX 6000 setup](docs/AMD_HIP.md#rdna2-gfx1030) |+| **Intel Arc · Linux** | [Intel setup](docs/INTEL_ARC.md) |+| **Apple Silicon · M1 or newer** | [Mac setup](docs/MAC.md) |+| **AMD Ryzen AI Max · Strix Halo** | [Strix Halo setup](docs/STRIX_HALO.md) |<a id="using-it"></a>## Tools-**Claude Code · Codex CLI · OpenCode · oh-my-pi · pi.dev · Hermes Agent · OpenClaw**, and more through the OpenAI- and Anthropic-compatible APIs.+**Claude Code · Codex CLI · OpenCode · oh-my-pi · pi.dev · Hermes Agent · OpenClaw**, and more through the OpenAI- and Anthropic-compatible APIs.Run **`strata connect`** to see setup instructions for your tool. [Connection guide](docs/TERMINAL_LAUNCH.md#connect-your-apps-strata-connect).@@ -174,8 +174,8 @@ Run **`strata connect`** to see setup instructions for your tool. [Connection gu## Help-[All guides](docs) · [Troubleshooting](docs/TROUBLESHOOTING.md) · [Report a bug](https://github.com/Niko1221/Strata/issues/new/choose) ·-[Benchmarks](docs/COMMUNITY_BENCHMARKS.md) · [How it works](docs/HOW_IT_WORKS.md) · [Contribute](CONTRIBUTING.md)+[All guides](docs) · [Troubleshooting](docs/TROUBLESHOOTING.md) · [Report a bug](https://github.com/Niko1221/Strata/issues/new/choose) ·+[Benchmarks](docs/COMMUNITY_BENCHMARKS.md) · [How it works](docs/HOW_IT_WORKS.md) · [Contribute](CONTRIBUTING.md)## Support Strata@@ -183,10 +183,11 @@ Run **`strata connect`** to see setup instructions for your tool. [Connection gu<a href="https://buymeacoffee.com/strataengine"><img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" alt="Buy Me a Coffee: support Strata" height="48"></a>-[MIT license](LICENSE) · [Credits and model licenses](docs/HOW_IT_WORKS.md#credits) · [Privacy](docs/SAFETY.md)+[MIT license](LICENSE) · [Credits and model licenses](docs/HOW_IT_WORKS.md#credits) · [Privacy](docs/SAFETY.md)<a id="option-c-let-your-ai-set-it-up"></a>-<sub>For AI tools and automated readers: <a href="AGENTS.md">technical reference</a> · <a href="docs/AI_SETUP.md">setup instructions</a>.</sub>+<sub>For AI tools and automated readers: <a href="AGENTS.md">technical reference</a> · <a href="docs/AI_SETUP.md">setup instructions</a>.</sub>**Sponsorships, hardware partnerships or collaborations:** [[email protected]](mailto:[email protected]).+diff --git a/docs/DEEPSEEK41.md b/docs/DEEPSEEK41.mdindex 9eea1feaa..eeb9336b6 100644--- a/docs/DEEPSEEK41.md+++ b/docs/DEEPSEEK41.md@@ -39,12 +39,12 @@ perplexity +7.3% (code +3.2%, agent work +2.2%). IQ1_M: perplexity +21% (KL 0.21| IQ4_NL: `DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf` | 111 GB | the 4-bit Engram lookup table the model reads from RAM or SSD. Not a model to pick: all three sets use this one file, and it is already in their download sizes || DSpark drafter (Q8_0) | 8 GB | the small draft model (Unsloth) |-**On a Ryzen AI Max "Strix Halo" (Windows)** the graphics chip and the CPU share one pool of RAM, so the engine counts it once: on-the 64 GB box about 49 GB of the experts stay in memory (the dense weights and 8 GB for Windows and the desktop aside) and the-other 83 GB of the 2.00 set are read from the SSD. Measured there (2.00 bpw, default Faster mode, DSpark auto): answers 14-16-tokens/s on code and 10-11 on chat; a 4,000-token prompt 78 tokens/s, 8,000 tokens 109. The tested driver is AMD Adrenalin 26.10.1-(driver 32.0.32015.7029); the 2025-03 driver failed at 32K context, so 26.10.1 is the minimum to use. The 32K and 64K contexts work-on it; an 8-hour 32K soak passed on this box.+**On a Ryzen AI Max "Strix Halo" (Windows)** the graphics chip and the CPU share one pool of memory, so the engine counts it once: the+128 GB box (a 64 GB BIOS carve-out plus the 64 GB Windows sees) keeps about 109 GB of the 2.00 set's experts in that pool (the dense+weights and 12 GB for Windows and the desktop aside) and reads the other 23 GB from the SSD. Measured there (2.00 bpw, default Faster+mode, DSpark auto): answers 14-16 tokens/s on code and 10-11 on chat; a 4,000-token prompt 78 tokens/s, 8,000 tokens 109. The tested+driver is AMD Adrenalin 26.10.1 (driver 32.0.32015.7029); the 2025-03 driver failed at 32K context, so 26.10.1 is the minimum to use.+The 32K and 64K contexts work on it; an 8-hour 32K soak passed on this box.121 + 111 + 8 = 239 GB for 1.75; 138 + 111 + 8 = 257 GB for 2.00; 155 + 111 + 8 = 274 GB for 2.25.diff --git a/docs/MODELS.md b/docs/MODELS.mdindex db536a5c6..7cbc2c442 100644--- a/docs/MODELS.md+++ b/docs/MODELS.md@@ -1,11 +1,11 @@-# Which model? Sizes, versions and what fits+# Which model? Sizes, versions and what fitsThis page helps you choose a model and check that your computer can run it. It covers DeepSeek V4.1-Flash,Qwen3.8-Flash-Next (and its variants) and Qwen3.6-35B-A3B. For installing, see [INSTALL.md](INSTALL.md).-> **On this page:** [Quick start](#quick-start) · [Pick by memory](#pick-by-memory) ·-> [DeepSeek V4.1-Flash](#deepseek-v41-flash) · [Qwen3.8-Flash-Next](#qwen38-flash-next) ·-> [Qwen3.6-35B-A3B](#qwen36-35b-a3b) · [Adding or switching models](#adding-or-switching-models) ·+> **On this page:** [Quick start](#quick-start) · [Pick by memory](#pick-by-memory) ·+> [DeepSeek V4.1-Flash](#deepseek-v41-flash) · [Qwen3.8-Flash-Next](#qwen38-flash-next) ·+> [Qwen3.6-35B-A3B](#qwen36-35b-a3b) · [Adding or switching models](#adding-or-switching-models) ·> [History](#history--older-measurements)## Quick start@@ -70,9 +70,9 @@ you have the disk. Perplexity is measured on a fixed set of 4,995 test texts; lo| --- | --- || NVIDIA RTX 30, 40 and 50 series, 12 GB or more | Windows, Linux || AMD Radeon AI PRO R9700 | Linux |-| AMD Ryzen AI Max (Strix Halo, gfx1151, 64 GB or more shared) | Windows |+| AMD Ryzen AI Max (Strix Halo, gfx1151, 128 GB unified; 64 GB or more shared with Windows) | Windows |-Tested on an RTX 3060 (Linux), an RTX 5070 (Windows), four R9700 cards (Linux) and a Ryzen AI Max+ 395 with 64 GB (Windows:+Tested on an RTX 3060 (Linux), an RTX 5070 (Windows), four R9700 cards (Linux) and a Ryzen AI Max+ 395 with 128 GB unified (Windows:14 to 16 tokens a second on code with the default Faster mode, 10 to 11 on chat, 78 tokens a second on a 4K prompt and 109 on8K). The RTX 30 kernels were also checked on an RTX A4000. Windows on RTX 30 and the RX 9070 are built but were not run. Macand Strix Halo on Linux do not run it yet; on Strix Halo Windows the graphics memory is the RAM, so most of the experts are@@ -344,3 +344,4 @@ above setup's RAM estimate for 47 GB and was installed with `--model Q2_0 --yes`| **UD-Q6_K_XL** | 54 | 76 | 1,020 || **Q8_0** | 53 to 57 | 66 to 74 | 1,230 to 1,300 || *IQ3_S, same card* | 75 | 95 | - |+diff --git a/docs/STRIX_HALO.md b/docs/STRIX_HALO.mdindex 11fdc77fc..0ffa8ad09 100644--- a/docs/STRIX_HALO.md+++ b/docs/STRIX_HALO.md@@ -1,19 +1,19 @@-# Strix Halo (Ryzen AI Max, gfx1151)+# Strix Halo (Ryzen AI Max, gfx1151)This page is for people with a Ryzen AI Max "Strix Halo" PC (Radeon 8060S / 8050S, RDNA 3.5, `gfx1151`, one pool of LPDDR5Xshared by the CPU and the GPU). Strata runs there through the same HIP backend as the Radeon cards in [AMD_HIP.md](AMD_HIP.md).**Status: experimental.****DeepSeek V4.1-Flash runs here on Windows** (since 0.2.7): its own engine package is built for gfx1151 and the official IST-DAS-1.75 / 2.00 / 2.25 bpw sets run through it. Measured on the 64 GB box with 2.00 bpw and the default Faster mode: answers 14-16+1.75 / 2.00 / 2.25 bpw sets run through it. Measured on the 128 GB box (64 GB BIOS carve-out plus 64 GB seen by Windows) with 2.00 bpw and the default Faster mode: answers 14-16tokens/s on code, 10-11 on chat; a 4K prompt 78 tokens/s and an 8K prompt 109. The graphics chip and the CPU share the RAM, so theengine counts it once and reads the rest of the experts from the SSD. [DeepSeek on Strix Halo](#deepseek-v41-flash) has the details,the driver it needs and the limits.-> **On this page:** [Quick start](#quick-start) · [DeepSeek](#deepseek-v41-flash) · [What the engine does differently](#what-the-engine-does-differently) ·-> [1. Toolchain](#1-the-toolchain-no-root-needed) · [2. Engine](#2-the-engine) · [3. Tuning table](#3-the-prompt-table-and-running) ·-> [4. Defaults](#4-what-is-on-by-default-on-gfx1151) · [5. Opt-in](#5-what-is-not-on-by-default-it-changes-bits) ·-> [5b. Desktop and RAM](#5b-the-desktop-the-ram-and-a-discrete-card-beside-the-strix-halo-linux) · [6. Measured](#6-measured) ·+> **On this page:** [Quick start](#quick-start) · [DeepSeek](#deepseek-v41-flash) · [What the engine does differently](#what-the-engine-does-differently) ·+> [1. Toolchain](#1-the-toolchain-no-root-needed) · [2. Engine](#2-the-engine) · [3. Tuning table](#3-the-prompt-table-and-running) ·+> [4. Defaults](#4-what-is-on-by-default-on-gfx1151) · [5. Opt-in](#5-what-is-not-on-by-default-it-changes-bits) ·+> [5b. Desktop and RAM](#5b-the-desktop-the-ram-and-a-discrete-card-beside-the-strix-halo-linux) · [6. Measured](#6-measured) ·> [History](#history--older-measurements)## Quick start@@ -40,12 +40,12 @@ package is the AMD one built for gfx1151 (`strata-dsv41-...-windows-x86_64-hip-r| | || --- | --- || Sets | IST-DAS 1.75 bpw (239 GB download), 2.00 bpw (257 GB), 2.25 bpw (274 GB); the n-gram table is shared |-| Memory | one pool: about 49 GB of the 2.00 set's 132 GB of experts stay in RAM beside the dense weights and 8 GB kept for Windows; the rest is read from the SSD |+| Memory | one pool of 128 GB (the BIOS carve-out plus the RAM): about 109 GB of the 2.00 set's 132 GB of experts stay in it beside the dense weights and 12 GB kept for Windows; the other 23 GB are read from the SSD || Driver | AMD Adrenalin 26.10.1 (driver 32.0.32015.7029) is the tested version, and the minimum to use: the 2025-03 driver failed at 32K context || Context | 8K default; 32K and 64K work (an 8-hour 32K soak passed) || Disk | the sets above; keep the files on the NVMe drive, they are read while it answers |-Measured on the maintainers' box (Ryzen AI Max+ 395, 64 GB, Windows 11, 2.00 bpw, Faster mode, DSpark auto, engine 0.1.51):+Measured on the maintainers' box (Ryzen AI Max+ 395, 128 GB unified, Windows 11, 2.00 bpw, Faster mode, DSpark auto, engine 0.1.51):| Workload | Result || --- | --- |@@ -55,12 +55,12 @@ Measured on the maintainers' box (Ryzen AI Max+ 395, 64 GB, Windows 11, 2.00 bpw| 8K prompt | about 109 tokens/s || Stability | 100-request, 1-hour (243 requests) and 8-hour 32K (646 requests) soaks passed on the engine line this package is built from |-How it uses the memory: the GPU's expert tier and the RAM store are the same memory, so the engine keeps a store of about 23 GB,+How it uses the memory: the GPU's expert tier and the RAM store are the same memory, so the engine keeps a store of about 22-23 GB,sized after the weights are on the GPU, and reads the other experts from the model files when a word needs them. It does not copy theCPU lane's experts into RAM on an integrated GPU: on this box that copy overcommits the shared memory and an 8K prompt stalls (foundon Strix on 2026-10-25, fixed for 0.2.7; `DS4_COPY_ALL=1` forces the old behaviour for a test). `--ssd-cache N` sets the store size.-Known limits: a 64 GB machine was tested. A 32 GB Strix Halo gets a much smaller store and is slower, and setup says so (it never+Known limits: the 128 GB machine was tested. A 64 GB or 32 GB Strix Halo gets a smaller store and is slower, and setup says so (it neverrefuses). Linux on Strix Halo has no DeepSeek engine package yet. WSL is not supported: the AMD engines hang there.## What the engine does differently@@ -272,3 +272,4 @@ UD-Q4_K_XL 49.5 -> 52.8 (+6.7%), prompt unchanged.| `--mtp-window 8192` | 64K output +3.5%, 32K -0.4% || 12-turn agent replay, `STRATA_PREFILL_STREAM_MIN=128`, C2T8, `--no-prefill-borrow` | 3.4 -> 2.1 s per turn || Defaults off, IQ3_XXS 8K | prompt 8% lower (1,137) |+diff --git a/setup.py b/setup.pyindex ab3b8a8d8..73decac76 100644--- a/setup.py+++ b/setup.py@@ -7883,9 +7883,10 @@ def dsv41_manual_steps(eng: Path) -> str:"and run it again. Strata checks it before it is used.")-def get_dsv41_engine(eng: Path, hip: bool, package: str | None = None) -> Path | None:+def get_dsv41_engine(eng: Path, hip: bool, package: str | None = None, feed: str | None = None) -> Path | None:"""The DeepSeek engine package next to the engine `eng`: one the person gave (--dsv41-package) or put in engine-dsv41/ by hand, else downloaded from- the signed feed. Checked first either way; its strata-dsv41, or None when it cannot be had."""+ the signed feed `feed` (--feed; else STRATA_UPDATE_FEED, a remembered https feed, or the release's). Checked first either way; its+ strata-dsv41, or None when it cannot be had."""from strata_control import dsv41_engineroot = dsv41_engine_dir(eng)if package:@@ -7900,7 +7901,7 @@ def get_dsv41_engine(eng: Path, hip: bool, package: str | None = None) -> Path |return exeacquisition.require("the DeepSeek engine")from strata_cli import update_cmd- source = update_cmd._feed_source(argparse.Namespace(feed=None), root.parent)+ source = update_cmd._feed_source(argparse.Namespace(feed=feed), root.parent)say(" Downloading the DeepSeek engine (its own package, separate from the Strata engine) ...")hw = Noneif hip:@@ -7912,14 +7913,15 @@ def get_dsv41_engine(eng: Path, hip: bool, package: str | None = None) -> Path |return exe-def upgrade_dsv41_engine(eng: Path, exe: Path) -> Path:- """The installed DeepSeek engine package is older than the signed feed's: updated next to it, switched after its test start, the old one removed- (strata_control.dsv41_engine.upgrade). Silent when the feed cannot be reached; a failure is one plain line and `exe` keeps running."""+def upgrade_dsv41_engine(eng: Path, exe: Path, feed: str | None = None) -> Path:+ """The installed DeepSeek engine package is older than the signed feed's (--feed, else STRATA_UPDATE_FEED, a remembered https feed,+ or the release's): updated next to it, switched after its test start, the old one removed (strata_control.dsv41_engine.upgrade).+ Silent when the feed cannot be reached; a failure is one plain line and `exe` keeps running."""from strata_control import dsv41_enginetry:from strata_cli import update_cmdroot = dsv41_engine_dir(eng)- source = update_cmd._feed_source(argparse.Namespace(feed=None), root.parent)+ source = update_cmd._feed_source(argparse.Namespace(feed=feed), root.parent)if not dsv41_engine.source_reachable(source):return exenew = dsv41_engine.upgrade(root, source, installed=exe, emit=lambda m: say(" " + m))@@ -8160,11 +8162,11 @@ def install_dsv41(a, family, gpu, hip, amd, cuda_tk, multi, port, ram, adopted,fail("Strata can't reach GitHub to download the DeepSeek engine.","Check the internet connection and run the same command again. " + dsv41_manual_steps(eng))if exe is None and not getattr(a, "build", False):- get_dsv41_engine(eng, hip, getattr(a, "dsv41_package", None))+ get_dsv41_engine(eng, hip, getattr(a, "dsv41_package", None), getattr(a, "feed", None))exe = dsv41_exe(eng)if (exe is not None and not getattr(a, "dsv41_package", None) and not getattr(a, "build", False)and not (os.environ.get("STRATA_DSV41_EXE") or "").strip() and exe.parent.parent == dsv41_engine_dir(eng)):- exe = upgrade_dsv41_engine(eng, exe) # an installed package older than the signed feed's: updated next to it, switched after it starts+ exe = upgrade_dsv41_engine(eng, exe, getattr(a, "feed", None)) # an installed package older than the signed feed's: updated next to it, switched after it startsif exe is not None and exe.parent != Path(eng):from strata_control import dsv41_enginelib_dirs = [*lib_dirs, *dsv41_engine.lib_dirs(exe)] # its own package, and the rocm/bin an AMD package keeps the ROCm DLLs indiff --git a/strata_cli/launch/runner.py b/strata_cli/launch/runner.pyindex 7b322be66..d94405415 100644--- a/strata_cli/launch/runner.py+++ b/strata_cli/launch/runner.py@@ -889,6 +889,8 @@ class Runner:args += ["--gguf-dir", str(p if p.is_dir() else p.parent)]if s.dsv41_package and s.family == "deepseek41":args += ["--dsv41-package", s.dsv41_package]+ if self.feed: # --feed / STRATA_UPDATE_FEED: setup fetches the DeepSeek engine from it too+ args += ["--feed", str(self.feed)]if s.gpu_roles:args += ["--gpu-roles", ",".join(f"{k}={v}" for k, v in sorted(s.gpu_roles.items(), key=lambda kv: int(kv[0])))]elif s.gpus:diff --git a/strata_control/dsv41_fit.py b/strata_control/dsv41_fit.pyindex e1bd201f9..ef7f354d8 100644--- a/strata_control/dsv41_fit.py+++ b/strata_control/dsv41_fit.py@@ -79,9 +79,10 @@ ISTA_MEASURED = ("an RTX 5070 (12 GB) with 64 GB of RAM and a 3.6 GB/s NVMe driv"about 31 tokens a second")# A unified-memory APU (Ryzen AI Max "Strix Halo", gfx1151, Windows): the graphics chip and the CPU share one pool of RAM, so the# engine's expert tiers are one pool too. The measured 64 GB box: 2.00 bpw, Faster (5 experts), DSpark auto, Adrenalin 26.10.1.-UMA_RESERVE_GB = 8.0 # kept for Windows, the desktop and the engine's buffers (measured min free 6.5-11 GB)-UMA_MEASURED = ("a Ryzen AI Max+ 395 (Radeon 8060S, 64 GB shared with the CPU) on Windows: answers 14-16 tokens a second on code in the "- "default Faster mode (5 experts) and 10-11 on chat; a 4,000-token prompt 77-81 tokens a second, 8,000 tokens 108-111")+UMA_RESERVE_GB = 12.0 # kept for Windows, the desktop and the engine's buffers (the engine's own RAM store margin)+UMA_MEASURED = ("a Ryzen AI Max+ 395 (Radeon 8060S, 128 GB unified: a 64 GB BIOS carve-out plus 64 GB seen by Windows) on Windows: answers "+ "14-16 tokens a second on code in the default Faster mode (5 experts) and 10-11 on chat; a 4,000-token prompt 77-81 tokens a "+ "second, 8,000 tokens 108-111")def variant(model: str | None) -> dict | None:@@ -97,6 +98,21 @@ def uma_card(cards: list[dict] | None) -> dict | None:return next((c for c in cards or [] if c.get("uma") or c.get("integrated")), None)+def uma_total_gb(card: dict, ram_gb: float | None) -> float:+ """The one pool of a unified-memory PC, counted once: the RAM Windows sees plus the BIOS carve-out the GPU owns (the "shared"+ part of the GPU's figure is that same RAM). On the owner's Strix Halo: 64 GB + a 64 GB carve-out = 128 GB. Without the carve-out+ in the card's data, the GPU's figure less its shared part, or the larger of the two, is the best that can be said."""+ ram = float(ram_gb or 0.0)+ dedicated = float(card.get("dedicated_gb") or 0.0)+ if dedicated > 0:+ return ram + dedicated+ vram = float(card.get("vram_gb") or 0.0)+ shared = float(card.get("shared_gb") or 0.0)+ if vram > 0 and shared > 0:+ return ram + max(0.0, vram - shared)+ return max(ram, vram)++def variant_key(model: str | None) -> str | None:return next((k for k in ISTA_SETS if k.lower() == str(model or "").lower()), None)@@ -261,7 +277,7 @@ def _plan_uma(key: str, model: str, card: dict, ram_gb: float | None, disk_free_counted once. The engine holds what fits beside the dense weights and the reserve and reads the rest from the SSD; the numbers inUMA_MEASURED are the measured 64 GB box. Never a refusal."""v = ISTA_SETS[key]- total = float(ram_gb or 0.0) or float(card.get("vram_gb") or 0.0)+ total = uma_total_gb(card, ram_gb)expert_gb = v["experts_gb"] / EXPERTSset_gb = v["set_bytes"] / GBdownload = ista_download_bytes(model) / GB@@ -289,8 +305,8 @@ def _plan_uma(key: str, model: str, card: dict, ram_gb: float | None, disk_free_lines.append("The n-gram table stays on the SSD and is read a few rows for each word: use an NVMe drive for the model files (a slow ""disk makes every word slower). The same table serves every IST-DAS set, so installing another one downloads it once.")name = card.get("name") or "the built-in Radeon"- lines.append(f"{name} shares this PC's RAM: one pool of {total:.0f} GB holds the model and the graphics memory together. About "- f"{resident:.0f} GB of the {float(v['experts_gb']):.0f} GB of experts stay in memory beside {ISTA_DENSE_GB:.1f} GB of "+ lines.append(f"{name} shares this PC's memory: the RAM and the GPU's BIOS carve-out are one pool of {total:.0f} GB. About "+ f"{resident:.0f} GB of the {float(v['experts_gb']):.0f} GB of experts stay in it beside {ISTA_DENSE_GB:.1f} GB of "f"dense weights and {UMA_RESERVE_GB:.0f} GB kept for Windows; the other {streamed:.0f} GB are read from the SSD while it answers.")lines.append(f"Measured on this kind of PC: {UMA_MEASURED}.")if streamed > 0.5:@@ -300,11 +316,11 @@ def _plan_uma(key: str, model: str, card: dict, ram_gb: float | None, disk_free_elif total >= float(v.get("ram_ok_gb") or 0):out["verdict"] = "tight"if total < 48:- warns.append(f"This PC has {total:.0f} GB of memory shared with the graphics chip and the model was measured on 64 GB: with less, "+ warns.append(f"This PC has {total:.0f} GB of memory shared with the graphics chip and the model was measured on 128 GB: with less, ""more experts come from the SSD and each word takes longer. It still runs.")else:out["verdict"] = "spills"- warns.append(f"This PC has {total:.0f} GB of memory shared with the graphics chip and the model was measured on 64 GB: most of its "+ warns.append(f"This PC has {total:.0f} GB of memory shared with the graphics chip and the model was measured on 128 GB: most of its ""experts are read from the SSD while it answers, so it is slow (a word can take seconds). The smaller 1.75 bpw set ""suits this computer better. It is still allowed.")if disk_free_gb is not None and disk_free_gb < need_disk:diff --git a/tests/terminal/test_dsv41_fit.py b/tests/terminal/test_dsv41_fit.pyindex 55dceca22..18b4bd387 100644--- a/tests/terminal/test_dsv41_fit.py+++ b/tests/terminal/test_dsv41_fit.py@@ -10,9 +10,12 @@ import unittestfrom strata_control import dsv41_fit as F-# what setup hands the plan on the owner's Strix Halo: one integrated Radeon, its "VRAM" counting the BIOS carve-out plus the shared pool-HALO = [{"index": 0, "name": "AMD Radeon 8060S (Ryzen AI Max, Strix Halo, gfx1151)", "vram_gb": 95.7, "arch": "gfx1151",- "uma": True, "integrated": True}]+# what setup hands the plan on the owner's Strix Halo: one integrated Radeon, its "VRAM" counting the 64 GB BIOS carve-out plus 32 GB+# shared with the 64 GB Windows sees (128 GB of memory in one pool)+HALO = [{"index": 0, "name": "AMD Radeon 8060S (Ryzen AI Max, Strix Halo, gfx1151)", "vram_gb": 96.0, "arch": "gfx1151",+ "uma": True, "integrated": True, "dedicated_gb": 64.0, "shared_gb": 32.0}]+SMALL = [{"index": 0, "name": "AMD Radeon 8040S (Ryzen AI Max, Strix Halo, gfx1151)", "vram_gb": 32.0, "arch": "gfx1151",+ "uma": True, "integrated": True, "dedicated_gb": 16.0, "shared_gb": 16.0}]R9700 = [{"index": 0, "name": "Radeon AI PRO R9700", "vram_gb": 32.0, "arch": "gfx1201"}]TWO_R9700 = [R9700[0], {**R9700[0], "index": 1}]@@ -21,18 +24,18 @@ class UnifiedMemoryPlan(unittest.TestCase):def plan(self, model="IST-DAS-2.00bpw", ram=63.6, disk=700.0):return F.plan(HALO, ram, disk, 16384, model=model)- def test_counts_the_graphics_memory_once(self):+ def test_counts_the_whole_pool_once(self):p = self.plan()self.assertTrue(p["uma"])self.assertEqual(p["tp"], 1) # no multi-card split wordsself.assertEqual(p["args"], []) # and no --tp argument- resident = 63.6 - F.ISTA_DENSE_GB - F.UMA_RESERVE_GB+ resident = 63.6 + 64.0 - F.ISTA_DENSE_GB - F.UMA_RESERVE_GB # the RAM Windows sees plus the BIOS carve-outself.assertAlmostEqual(p["vram_experts_gb"], resident, places=1)self.assertAlmostEqual(p["streamed_gb"], 132.07 - resident, places=1)self.assertEqual(p["experts_in_vram"], int(resident / (132.07 / F.EXPERTS)))- self.assertEqual(p["verdict"], "tight") # it runs, most of it from the SSD: measured 14-16 tokens/s on code+ self.assertEqual(p["verdict"], "tight") # it runs, most of it from memory: measured 14-16 tokens/s on codetext = " ".join(p["lines"])- self.assertIn("shares this PC's RAM", text)+ self.assertIn("one pool of 128 GB", text)self.assertIn("read from the SSD", text)self.assertIn("14-16 tokens a second", text)@@ -40,20 +43,20 @@ class UnifiedMemoryPlan(unittest.TestCase):p = self.plan("IST-DAS-2.25bpw")self.assertTrue(p["disk_ok"])self.assertAlmostEqual(p["download_gb"], F.ista_download_bytes("IST-DAS-2.25bpw") / 1e9, places=1)- self.assertGreater(p["streamed_gb"], 90.0) # 2.25 needs more from the SSD than 2.00+ self.assertGreater(p["streamed_gb"], 30.0) # 2.25 needs more from the SSD than 2.00self.assertTrue(any("274 GB" in l for l in p["lines"])) # the set's own downloaddef test_a_smaller_unified_pc_is_slower_never_refused(self):- p = F.plan(HALO, 32.0, 700.0, 8192, model="IST-DAS-2.00bpw")+ p = F.plan(SMALL, 16.0, 700.0, 8192, model="IST-DAS-2.00bpw")self.assertEqual(p["verdict"], "spills")self.assertTrue(any("1.75 bpw" in w for w in p["warnings"]))self.assertTrue(p["disk_ok"])- def test_a_128_gb_unified_pc_holds_more(self):- p = F.plan(HALO, 127.0, 700.0, 8192, model="IST-DAS-1.75bpw")- self.assertGreater(p["vram_experts_gb"], 110.0)- self.assertLess(p["streamed_gb"], 3.0)- self.assertEqual(p["verdict"], "tight") # 1.75 nearly fits a 128 GB Strix Halo: about 2 GB still come from the SSD+ def test_a_128_gb_unified_pc_nearly_holds_the_smallest_set(self):+ p = F.plan(HALO, 63.6, 700.0, 8192, model="IST-DAS-1.75bpw")+ self.assertGreater(p["vram_experts_gb"], 105.0)+ self.assertLess(p["streamed_gb"], 10.0)+ self.assertEqual(p["verdict"], "tight") # 1.75 nearly fits the 128 GB Strix Halo: ~5 GB still come from the SSDdef test_a_large_enough_unified_pc_fits(self):p = F.plan(HALO, 160.0, 700.0, 8192, model="IST-DAS-1.75bpw")diff --git a/tests/terminal/test_launch_dsv41.py b/tests/terminal/test_launch_dsv41.pyindex 0e126eadf..fee3fb0ec 100644--- a/tests/terminal/test_launch_dsv41.py+++ b/tests/terminal/test_launch_dsv41.py@@ -122,6 +122,15 @@ class LaunchDeepSeek(unittest.TestCase):self.assertNotIn("--parallel", args)self.assertIsNone(Runner(s, catalog=catalog(self.hw)).catalog_parallel())+ def test_the_feed_reaches_setup_so_the_deepseek_engine_comes_from_it(self):+ """`strata launch --feed DIR` must hand the feed to setup: the DeepSeek engine package is fetched by setup, not by `strata update`+ (found on strix, 2026-10-25: the launch read the local feed for its own checks and setup went to GitHub for the engine)."""+ s = S.LaunchSettings(family="deepseek41", model="IST-DAS-2.00bpw", context=8192, vision="off", tune="skip")+ args = Runner(s, catalog=catalog(self.hw), feed="/data/feed").setup_args()+ self.assertIn("--feed", args)+ self.assertEqual(args[args.index("--feed") + 1], "/data/feed")+ self.assertNotIn("--feed", Runner(s, catalog=catalog(self.hw)).setup_args())+class OwnShards(unittest.TestCase):def setUp(self):