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++
DeepSeek tables: the IQ1_M (1.75 bpw, ist-das-1.75bpw) row first, 239 GB; quality 'being measured' plus ISTA-DASLab's model-card figure; README + 6 translations, DEEPSEEK41.md, release notes (placeholder until the KL is in)
Niko1221672612ac1c0f
diff --git a/README.de.md b/README.de.mdindex 65a8279d4..43b6184f9 100644--- a/README.de.md+++ b/README.de.md@@ -192,17 +192,18 @@ Alle Größen, Downloads und was wo läuft: [docs/MODELS.md](docs/MODELS.md).### DeepSeek V4.1-Flash (im Terminal, ab 0.2.4)-[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) ist ein anderes, deutlich größeres Modell. ISTA-DASLab hat die beiden Sets unten gebaut. Strata startet sie auf einem PC mit einer 12-GB-Grafikkarte und 46 bis 64 GB RAM und liest den Rest von der SSD. Es läuft auf NVIDIA-RTX-Karten der Serien 30 bis 50 und auf der Radeon AI PRO R9700. Es nutzt ein eigenes Engine-Paket, `strata-dsv41`, das `strata launch` lädt, wenn du DeepSeek wählst. Die Desktop-App bietet es noch nicht an.+[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) ist ein anderes, deutlich größeres Modell. ISTA-DASLab hat die drei Sets unten gebaut. Strata startet sie auf einem PC mit einer 12-GB-Grafikkarte und 46 bis 64 GB RAM und liest den Rest von der SSD. Es läuft auf NVIDIA-RTX-Karten der Serien 30 bis 50 und auf der Radeon AI PRO R9700. Es nutzt ein eigenes Engine-Paket, `strata-dsv41`, das `strata launch` lädt, wenn du DeepSeek wählst. Die Desktop-App bietet es noch nicht an.| Set | Download | Qualität im Vergleich zum Originalmodell | Antwortet mit || --- | ---: | --- | --- |+| **IQ1_M** (1,75 Bit) | 239 GB | KL und Perplexität: werden noch gemessen. Laut Modellkarte von ISTA-DASLab behält es 98,7 % des Aufgabendurchschnitts des Basismodells (AIME25, GPQA-Diamond, LiveCodeBench v6). | noch nicht gemessen || **IQ2_XXS** (2.00 Bit) | 257 GB | KL 0,169, Perplexität 3,64 (+16 %) | 6 bis 9 Tokens/s auf einer RTX 3060, etwa 5 auf einer RTX 5070 || **IQ2_XS** (2.25 Bit) | 274 GB | KL 0,110, Perplexität 3,40 (+8 %); Code +3,5 %, Agent-Aufgaben +2,5 % | etwa 11 Tokens/s auf einer Radeon AI PRO R9700 |-| **IQ4_NL** (Engram-Tabelle, kein drittes Modell) | 111 GB, schon in beiden Downloads oben enthalten | Die 4-Bit-Engram-Nachschlagetabelle, die das Modell aus RAM oder SSD liest. Beide Sets nutzen diese eine Datei (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |+| **IQ4_NL** (Engram-Tabelle, kein drittes Modell) | 111 GB, schon in allen drei Downloads oben enthalten | Die 4-Bit-Engram-Nachschlagetabelle, die das Modell aus RAM oder SSD liest. Alle drei Sets nutzen diese eine Datei (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |4x Radeon AI PRO R9700: etwa 31 Tokens/s, 55 bis 69 mit DSpark.-Starte es mit `strata launch --model ist-das-2.00bpw` (oder `ist-das-2.25bpw`). Details, Geschwindigkeiten je Karte und die manuelle Installation: [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).+Starte es mit `strata launch --model ist-das-2.00bpw` (oder `ist-das-1.75bpw`, `ist-das-2.25bpw`). Details, Geschwindigkeiten je Karte und die manuelle Installation: [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).## Wie schnell ist es?diff --git a/README.es.md b/README.es.mdindex e18da76ef..a77d76ee8 100644--- a/README.es.md+++ b/README.es.md@@ -192,17 +192,18 @@ Todos los tamaños, descargas y lo que cabe en cada equipo: [docs/MODELS.md](doc### DeepSeek V4.1-Flash (en el Terminal, desde 0.2.4)-[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) es otro modelo, mucho más grande. ISTA-DASLab hizo los dos conjuntos de abajo. Strata los ejecuta en un PC con una tarjeta gráfica de 12 GB y de 46 a 64 GB de RAM, y lee el resto de tu SSD. Funciona en tarjetas NVIDIA RTX de las series 30 a 50 y en la Radeon AI PRO R9700. Usa su propio paquete de motor, `strata-dsv41`, que `strata launch` descarga cuando eliges DeepSeek. La app de escritorio todavía no lo ofrece.+[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) es otro modelo, mucho más grande. ISTA-DASLab hizo los tres conjuntos de abajo. Strata los ejecuta en un PC con una tarjeta gráfica de 12 GB y de 46 a 64 GB de RAM, y lee el resto de tu SSD. Funciona en tarjetas NVIDIA RTX de las series 30 a 50 y en la Radeon AI PRO R9700. Usa su propio paquete de motor, `strata-dsv41`, que `strata launch` descarga cuando eliges DeepSeek. La app de escritorio todavía no lo ofrece.| Conjunto | Descarga | Calidad frente al modelo original | Responde a || --- | ---: | --- | --- |+| **IQ1_M** (1,75 bits) | 239 GB | KL y perplejidad: se están midiendo. Según la ficha del modelo de ISTA-DASLab, conserva el 98,7 % del promedio de tareas del modelo base (AIME25, GPQA-Diamond, LiveCodeBench v6). | aún sin medir || **IQ2_XXS** (2.00 bits) | 257 GB | KL 0,169, perplejidad 3,64 (+16 %) | de 6 a 9 tokens/s en una RTX 3060, unos 5 en una RTX 5070 || **IQ2_XS** (2.25 bits) | 274 GB | KL 0,110, perplejidad 3,40 (+8 %); código +3,5 %, tareas de agente +2,5 % | unos 11 tokens/s en una Radeon AI PRO R9700 |-| **IQ4_NL** (tabla Engram, no es un tercer modelo) | 111 GB, ya incluidos en las dos descargas de arriba | La tabla de consulta Engram de 4 bits que el modelo lee de la RAM o del SSD. Los dos conjuntos usan este único archivo (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |+| **IQ4_NL** (tabla Engram, no es un tercer modelo) | 111 GB, ya incluidos en las tres descargas de arriba | La tabla de consulta Engram de 4 bits que el modelo lee de la RAM o del SSD. Los tres conjuntos usan este único archivo (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |4x Radeon AI PRO R9700: unos 31 tokens/s, de 55 a 69 con DSpark.-Inícialo con `strata launch --model ist-das-2.00bpw` (o `ist-das-2.25bpw`). Detalles, velocidades por tarjeta e instalación manual: [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).+Inícialo con `strata launch --model ist-das-2.00bpw` (o `ist-das-1.75bpw`, `ist-das-2.25bpw`). Detalles, velocidades por tarjeta e instalación manual: [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).## ¿Qué velocidad tiene?diff --git a/README.fr.md b/README.fr.mdindex 948b2422f..a62fb8010 100644--- a/README.fr.md+++ b/README.fr.md@@ -193,17 +193,18 @@ Toutes les tailles, les téléchargements et ce qui tient où : [docs/MODELS.md]### DeepSeek V4.1-Flash (dans le Terminal, depuis 0.2.4)-[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) est un autre modèle, beaucoup plus gros. ISTA-DASLab a produit les deux versions ci-dessous. Strata les fait tourner sur un PC avec une carte graphique de 12 Go et de 46 à 64 Go de RAM, et lit le reste sur votre SSD. Il fonctionne sur les cartes NVIDIA RTX des séries 30 à 50 et sur la Radeon AI PRO R9700. Il utilise son propre paquet de moteur, `strata-dsv41`, que `strata launch` télécharge quand vous choisissez DeepSeek. L’application de bureau ne le propose pas encore.+[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) est un autre modèle, beaucoup plus gros. ISTA-DASLab a produit les trois versions ci-dessous. Strata les fait tourner sur un PC avec une carte graphique de 12 Go et de 46 à 64 Go de RAM, et lit le reste sur votre SSD. Il fonctionne sur les cartes NVIDIA RTX des séries 30 à 50 et sur la Radeon AI PRO R9700. Il utilise son propre paquet de moteur, `strata-dsv41`, que `strata launch` télécharge quand vous choisissez DeepSeek. L’application de bureau ne le propose pas encore.| Version | Téléchargement | Qualité par rapport au modèle d’origine | Écrit à || --- | ---: | --- | --- |+| **IQ1_M** (1,75 bit) | 239 Go | KL et perplexité : en cours de mesure. D’après la fiche du modèle d’ISTA-DASLab, elle garde 98,7 % de la moyenne des tâches du modèle de base (AIME25, GPQA-Diamond, LiveCodeBench v6). | pas encore mesurée || **IQ2_XXS** (2.00 bits) | 257 GB | KL 0,169, perplexité 3,64 (+16 %) | 6 à 9 tokens/s sur une RTX 3060, environ 5 sur une RTX 5070 || **IQ2_XS** (2.25 bits) | 274 GB | KL 0,110, perplexité 3,40 (+8 %) ; code +3,5 %, tâches d’agent +2,5 % | environ 11 tokens/s sur une Radeon AI PRO R9700 |-| **IQ4_NL** (table Engram, pas un troisième modèle) | 111 Go, déjà compris dans les deux téléchargements ci-dessus | La table de consultation Engram en 4 bits que le modèle lit dans la RAM ou sur le SSD. Les deux versions utilisent ce seul fichier (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |+| **IQ4_NL** (table Engram, pas un troisième modèle) | 111 Go, déjà compris dans les trois téléchargements ci-dessus | La table de consultation Engram en 4 bits que le modèle lit dans la RAM ou sur le SSD. Les trois versions utilisent ce seul fichier (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |4x Radeon AI PRO R9700 : environ 31 tokens/s, 55 à 69 avec DSpark.-Lancez-le avec `strata launch --model ist-das-2.00bpw` (ou `ist-das-2.25bpw`). Détails, vitesses par carte et installation manuelle : [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).+Lancez-le avec `strata launch --model ist-das-2.00bpw` (ou `ist-das-1.75bpw`, `ist-das-2.25bpw`). Détails, vitesses par carte et installation manuelle : [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).## Quelle est sa vitesse ?diff --git a/README.ja.md b/README.ja.mdindex 26351c870..d53a43289 100644--- a/README.ja.md+++ b/README.ja.md@@ -185,17 +185,18 @@ Gyro は NVIDIA と AMD のカード(AMD は実験的)、および Linux 上### DeepSeek V4.1-Flash(ターミナル版、0.2.4 から)-[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) は、これとは別のずっと大きなモデルです。下の 2 つのセットは ISTA-DASLab が作りました。Strata は、12 GB のグラフィックスカード 1 枚と 46 から 64 GB の RAM を持つ PC でこれを動かし、残りは SSD から読みます。NVIDIA RTX 30 から 50 シリーズのカードと Radeon AI PRO R9700 で動きます。専用のエンジンパッケージ `strata-dsv41` を使い、DeepSeek を選ぶと `strata launch` がダウンロードします。デスクトップアプリはまだ対応していません。+[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) は、これとは別のずっと大きなモデルです。下の 3 つのセットは ISTA-DASLab が作りました。Strata は、12 GB のグラフィックスカード 1 枚と 46 から 64 GB の RAM を持つ PC でこれを動かし、残りは SSD から読みます。NVIDIA RTX 30 から 50 シリーズのカードと Radeon AI PRO R9700 で動きます。専用のエンジンパッケージ `strata-dsv41` を使い、DeepSeek を選ぶと `strata launch` がダウンロードします。デスクトップアプリはまだ対応していません。| セット | ダウンロード | 元のモデルとの品質差 | 出力速度 || --- | ---: | --- | --- |+| **IQ1_M**(1.75 ビット) | 239 GB | KL とパープレキシティ:測定中。ISTA-DASLab のモデルカードによると、基本モデルのタスク平均の 98.7% を保ちます(AIME25、GPQA-Diamond、LiveCodeBench v6)。 | 未測定 || **IQ2_XXS** (2.00 ビット) | 257 GB | KL 0.169、パープレキシティ 3.64(+16%) | RTX 3060 で毎秒 6 から 9 トークン、RTX 5070 で約 5 || **IQ2_XS** (2.25 ビット) | 274 GB | KL 0.110、パープレキシティ 3.40(+8%)、コード +3.5%、エージェント作業 +2.5% | Radeon AI PRO R9700 1 枚で毎秒約 11 トークン |-| **IQ4_NL**(Engram テーブル、3 つ目のモデルではありません) | 111 GB、上の 2 つのダウンロードに含まれています | モデルが RAM または SSD から読む 4 ビットの Engram 参照テーブルです。2 つのセットはこの 1 つのファイル(`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`)を共有します。 | 該当なし |+| **IQ4_NL**(Engram テーブル、3 つ目のモデルではありません) | 111 GB、上の 3 つのダウンロードに含まれています | モデルが RAM または SSD から読む 4 ビットの Engram 参照テーブルです。3 つのセットはこの 1 つのファイル(`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`)を共有します。 | 該当なし |Radeon AI PRO R9700 4 枚:毎秒約 31 トークン、DSpark ありで 55 から 69。-`strata launch --model ist-das-2.00bpw`(または `ist-das-2.25bpw`)で起動します。詳細、カードごとの速度、手動インストールは [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md) にあります。+`strata launch --model ist-das-2.00bpw`(または `ist-das-1.75bpw`、`ist-das-2.25bpw`)で起動します。詳細、カードごとの速度、手動インストールは [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md) にあります。## どのくらい速い?diff --git a/README.md b/README.mdindex 49845e0bd..1299c6caf 100644--- a/README.md+++ b/README.md@@ -190,17 +190,18 @@ Every size, download and what fits where: [docs/MODELS.md](docs/MODELS.md).### DeepSeek V4.1-Flash (in the Terminal, from 0.2.4)-[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) is a different, much bigger model. ISTA-DASLab made the two sets below. Strata runs them on a PC with one 12 GB graphics card and 46 to 64 GB of RAM, and reads the rest from your SSD. It runs on NVIDIA RTX 30 to 50 cards and on the Radeon AI PRO R9700. It uses its own engine package, `strata-dsv41`, which `strata launch` downloads when you pick DeepSeek. The desktop app does not offer it yet.+[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) is a different, much bigger model. ISTA-DASLab made the three sets below. Strata runs them on a PC with one 12 GB graphics card and 46 to 64 GB of RAM, and reads the rest from your SSD. It runs on NVIDIA RTX 30 to 50 cards and on the Radeon AI PRO R9700. It uses its own engine package, `strata-dsv41`, which `strata launch` downloads when you pick DeepSeek. The desktop app does not offer it yet.| Set | Download | Quality against the original model | Writes at || --- | ---: | --- | --- |+| **IQ1_M** (1.75 bits) | 239 GB | KL and perplexity: being measured. ISTA-DASLab's model card: keeps 98.7% of the base model's task average (AIME25, GPQA-Diamond, LiveCodeBench v6). | not measured yet || **IQ2_XXS** (2.00 bits) | 257 GB | KL 0.169, perplexity 3.64 (+16%) | 6 to 9 tokens/s on an RTX 3060, about 5 on an RTX 5070 || **IQ2_XS** (2.25 bits) | 274 GB | KL 0.110, perplexity 3.40 (+8%); code +3.5%, agent work +2.5% | about 11 tokens/s on one Radeon AI PRO R9700 |-| **IQ4_NL** (Engram table, not a third model) | 111 GB, already in both downloads above | The 4-bit Engram lookup table the model reads from RAM or SSD. Both sets use this one file (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |+| **IQ4_NL** (Engram table, not a third model) | 111 GB, already in all three downloads above | The 4-bit Engram lookup table the model reads from RAM or SSD. All three sets use this one file (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |4x Radeon AI PRO R9700: about 31 tokens/s, 55 to 69 with DSpark.-Start it with `strata launch --model ist-das-2.00bpw` (or `ist-das-2.25bpw`). Details, speeds per card and the manual install: [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).+Start it with `strata launch --model ist-das-2.00bpw` (or `ist-das-1.75bpw`, `ist-das-2.25bpw`). Details, speeds per card and the manual install: [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).## How fast is it?diff --git a/README.pt-BR.md b/README.pt-BR.mdindex 414b08e5d..f4c312254 100644--- a/README.pt-BR.md+++ b/README.pt-BR.md@@ -185,17 +185,18 @@ Todos os tamanhos, downloads e o que cabe em cada configuração: [docs/MODELS.m### DeepSeek V4.1-Flash (no Terminal, a partir da 0.2.4)-[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) é outro modelo, bem maior. A ISTA-DASLab fez os dois conjuntos abaixo. O Strata os executa em um PC com uma placa de vídeo de 12 GB e de 46 a 64 GB de RAM, e lê o resto do seu SSD. Ele roda em placas NVIDIA RTX das séries 30 a 50 e na Radeon AI PRO R9700. Usa um pacote de motor próprio, `strata-dsv41`, que o `strata launch` baixa quando você escolhe o DeepSeek. O app de desktop ainda não o oferece.+[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) é outro modelo, bem maior. A ISTA-DASLab fez os três conjuntos abaixo. O Strata os executa em um PC com uma placa de vídeo de 12 GB e de 46 a 64 GB de RAM, e lê o resto do seu SSD. Ele roda em placas NVIDIA RTX das séries 30 a 50 e na Radeon AI PRO R9700. Usa um pacote de motor próprio, `strata-dsv41`, que o `strata launch` baixa quando você escolhe o DeepSeek. O app de desktop ainda não o oferece.| Conjunto | Download | Qualidade em relação ao modelo original | Responde a || --- | ---: | --- | --- |+| **IQ1_M** (1,75 bits) | 239 GB | KL e perplexidade: ainda sendo medidos. Segundo o cartão do modelo da ISTA-DASLab, mantém 98,7% da média de tarefas do modelo base (AIME25, GPQA-Diamond, LiveCodeBench v6). | ainda não medida || **IQ2_XXS** (2.00 bits) | 257 GB | KL 0,169, perplexidade 3,64 (+16%) | 6 a 9 tokens/s em uma RTX 3060, cerca de 5 em uma RTX 5070 || **IQ2_XS** (2.25 bits) | 274 GB | KL 0,110, perplexidade 3,40 (+8%); código +3,5%, tarefas de agente +2,5% | cerca de 11 tokens/s em uma Radeon AI PRO R9700 |-| **IQ4_NL** (tabela Engram, não é um terceiro modelo) | 111 GB, já incluídos nos dois downloads acima | A tabela de consulta Engram de 4 bits que o modelo lê da RAM ou do SSD. Os dois conjuntos usam este único arquivo (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |+| **IQ4_NL** (tabela Engram, não é um terceiro modelo) | 111 GB, já incluídos nos três downloads acima | A tabela de consulta Engram de 4 bits que o modelo lê da RAM ou do SSD. Os três conjuntos usam este único arquivo (`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`). | n/a |4x Radeon AI PRO R9700: cerca de 31 tokens/s, 55 a 69 com DSpark.-Inicie com `strata launch --model ist-das-2.00bpw` (ou `ist-das-2.25bpw`). Detalhes, velocidades por placa e a instalação manual: [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).+Inicie com `strata launch --model ist-das-2.00bpw` (ou `ist-das-1.75bpw`, `ist-das-2.25bpw`). Detalhes, velocidades por placa e a instalação manual: [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md).## Qual é a velocidade?diff --git a/README.zh-CN.md b/README.zh-CN.mdindex 02bd4a16d..2b4abbb9f 100644--- a/README.zh-CN.md+++ b/README.zh-CN.md@@ -184,17 +184,18 @@ Gyro 可在 NVIDIA 和 AMD 显卡(AMD 为实验性支持)以及 Linux 上的### DeepSeek V4.1-Flash(终端版,0.2.4 起)-[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) 是另一个大得多的模型。下面两个版本由 ISTA-DASLab 制作。Strata 在一台有一块 12 GB 显卡和 46 到 64 GB 内存的电脑上运行它们,其余部分从固态硬盘读取。它可在 NVIDIA RTX 30 到 50 系列显卡和 Radeon AI PRO R9700 上运行。它使用自己的引擎包 `strata-dsv41`,你选择 DeepSeek 时由 `strata launch` 下载。桌面应用暂时不提供它。+[DeepSeek V4.1-Flash](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF) 是另一个大得多的模型。下面三个版本由 ISTA-DASLab 制作。Strata 在一台有一块 12 GB 显卡和 46 到 64 GB 内存的电脑上运行它们,其余部分从固态硬盘读取。它可在 NVIDIA RTX 30 到 50 系列显卡和 Radeon AI PRO R9700 上运行。它使用自己的引擎包 `strata-dsv41`,你选择 DeepSeek 时由 `strata launch` 下载。桌面应用暂时不提供它。| 版本 | 下载大小 | 与原模型相比的质量 | 生成速度 || --- | ---: | --- | --- |+| **IQ1_M**(1.75 位) | 239 GB | KL 和困惑度:正在测量。据 ISTA-DASLab 的模型卡,它保留了基础模型任务平均分的 98.7%(AIME25、GPQA-Diamond、LiveCodeBench v6)。 | 尚未测量 || **IQ2_XXS** (2.00 位) | 257 GB | KL 0.169,困惑度 3.64(+16%) | RTX 3060 上每秒 6 到 9 个 token,RTX 5070 上约 5 个 || **IQ2_XS** (2.25 位) | 274 GB | KL 0.110,困惑度 3.40(+8%);代码 +3.5%,智能体任务 +2.5% | 一块 Radeon AI PRO R9700 上每秒约 11 个 token |-| **IQ4_NL**(Engram 表,不是第三个模型) | 111 GB,已包含在上面两个下载中 | 模型从内存或固态硬盘读取的 4 位 Engram 查找表。两个版本共用这一个文件(`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`)。 | 不适用 |+| **IQ4_NL**(Engram 表,不是第三个模型) | 111 GB,已包含在上面三个下载中 | 模型从内存或固态硬盘读取的 4 位 Engram 查找表。三个版本共用这一个文件(`DeepSeek-V4.1-Flash-Engram-IQ4_NL.gguf`)。 | 不适用 |4 块 Radeon AI PRO R9700:每秒约 31 个 token,使用 DSpark 时为 55 到 69。-用 `strata launch --model ist-das-2.00bpw`(或 `ist-das-2.25bpw`)启动。详情、各显卡的速度和手动安装见 [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md)。+用 `strata launch --model ist-das-2.00bpw`(或 `ist-das-1.75bpw`、`ist-das-2.25bpw`)启动。详情、各显卡的速度和手动安装见 [docs/DEEPSEEK41.md](docs/DEEPSEEK41.md)。## 速度如何?diff --git a/docs/DEEPSEEK41.md b/docs/DEEPSEEK41.mdindex 0cc8bc618..7f832815a 100644--- a/docs/DEEPSEEK41.md+++ b/docs/DEEPSEEK41.md@@ -26,19 +26,20 @@ with an out-of-memory message. Both are said before anything is downloaded.## The official 2-bit versions (the ones for a normal PC)-Setup offers two official sets (ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF, pinned): **IST-DAS 2.00 bpw** (138 GB) and **IST-DAS 2.25 bpw** (155 GB). Both use the+Setup offers three official sets (ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF, pinned): **IQ1_M, 1.75 bpw** (`ist-das-1.75bpw`, 121 GB), **IQ2_XXS, 2.00 bpw** (138 GB) and **IQ2_XS, 2.25 bpw** (155 GB). All use thesame 111 GB n-gram table (downloaded once into `models/deepseek-v4.1-flash-shared`) and the 8 GB DSpark drafter: 257 GB of disk for-2.00, 274 GB for 2.25, 412 GB for both. Quality against the original model: perplexity +16% on 2.00, +8% on 2.25 (code and agent work-about +3.5% / +2.5% on 2.25).+2.00, 274 GB for 2.25, 239 GB for 1.75, 533 GB for all three. Quality against the original model: perplexity +16% on 2.00, +8% on 2.25 (code and agent work+about +3.5% / +2.5% on 2.25). IQ1_M: our KL and perplexity are being measured; ISTA-DASLab's model card says it keeps 98.7% of the base model's task average (AIME25, GPQA-Diamond, LiveCodeBench v6).| File | Size | What it is || --- | ---: | --- |-| IQ2_XXS shards (2.00 bpw), 4 files | 138 GB | the model, first set |-| IQ2_XS shards (2.25 bpw), 4 files | 155 GB | the model, second set |-| 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 third model: both sets use this one file, and it is already in their download sizes |+| IQ1_M shards (1.75 bpw), 4 files | 121 GB | the model, smallest set |+| IQ2_XXS shards (2.00 bpw), 4 files | 138 GB | the model, middle set |+| IQ2_XS shards (2.25 bpw), 4 files | 155 GB | the model, largest set |+| 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) |-138 + 111 + 8 = 257 GB for 2.00; 155 + 111 + 8 = 274 GB for 2.25.+121 + 111 + 8 = 239 GB for 1.75; 138 + 111 + 8 = 257 GB for 2.00; 155 + 111 + 8 = 274 GB for 2.25.Measured: an RTX 5070 (12 GB) with 64 GB of RAM and a 3.6 GB/s NVMe drive answers at about 4.7 tokens/s and reads prompts at 121tokens/s (4K) to 231 tokens/s (8K, layer-major path); an RTX 3060 (12 GB) with 46 GB of RAM about 7.2 tokens/s; four R9700 about 31diff --git a/docs/release-notes-0.2.4.md b/docs/release-notes-0.2.4.mdindex 9c4832987..8b3de0d3f 100644--- a/docs/release-notes-0.2.4.md+++ b/docs/release-notes-0.2.4.md@@ -6,15 +6,16 @@ Strata now runs DeepSeek V4.1-Flash on a PC with one 12 GB graphics card and 46## What you get-- **DeepSeek V4.1-Flash** in the model list of Setup, in two sizes from [ISTA-DASLab](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF):+- **DeepSeek V4.1-Flash** in the model list of Setup, in three sizes from [ISTA-DASLab](https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF):| set | download | quality against the original model ||---|---|---|+ | IQ1_M (1.75 bits) | 239 GB | being measured; ISTA-DASLab's model card: keeps 98.7% of the base model's task average (AIME25, GPQA-Diamond, LiveCodeBench v6) || IQ2_XXS (2.00 bits) | 257 GB | perplexity +16%; code and agent text +8.5% and +5.9% || IQ2_XS (2.25 bits) | 274 GB | perplexity +7.7% to +8.3%; code and agent text +3.5% and +2.5% |- | IQ4_NL (Engram table, not a third model) | 111 GB, already in both downloads above | The 4-bit Engram lookup table the model reads from RAM or SSD; both sets use this one file |+ | IQ4_NL (Engram table, not a third model) | 111 GB, already in all three downloads above | The 4-bit Engram lookup table the model reads from RAM or SSD; all three sets use this one file |- "Perplexity" is how surprised the model is by a fixed set of 4,995 test texts; lower is closer to the original. Pick 2.25 for coding and agent work when you have the disk. Setup shows both sizes and warns if your PC is short of disk or RAM. It does not stop you.+ "Perplexity" is how surprised the model is by a fixed set of 4,995 test texts; lower is closer to the original. Pick IQ2_XS for coding and agent work when you have the disk. Setup shows both sizes and warns if your PC is short of disk or RAM. It does not stop you.- Setup downloads the files, builds the model folder and starts the engine. You set no flags.## Speed on real PCs@@ -44,7 +45,7 @@ These are the 2.00-bit set. The 2.25-bit set reads 13% more data per token, so e## Needs-- Disk: 257 GB (2.00) or 274 GB (2.25). The download includes the Engram table and the DSpark drafter.+- Disk: 239 GB (IQ1_M), 257 GB (IQ2_XXS) or 274 GB (IQ2_XS). The download includes the Engram table and the DSpark drafter.- RAM: 46 GB worked on the RTX 3060; 64 GB is the tested comfortable size. More RAM means a bigger store and faster answers.- Graphics card: 12 GB or more.