diff --git a/docs/content/features/model-gallery.md b/docs/content/features/model-gallery.md index 56dde2118..b99eaf9a0 100644 --- a/docs/content/features/model-gallery.md +++ b/docs/content/features/model-gallery.md @@ -44,6 +44,24 @@ Both views store the view, search, filter, and selection in the URL. Installing from Explore does not move you away from the catalog; the entry updates in place when the operation finishes. +### Qwen3.8 Jumbo Fusion 27B + +The gallery includes DavidAU's Qwen3.8 Jumbo Fusion Q4_K_M and Q8_0 GGUF builds +for text chat through llama.cpp. Both use the embedded chat template and a +32,768-token default context. These entries do not include a vision projector +or enable MTP decoding. + +Install with automatic variant selection: + +```bash +local-ai models install qwen3.8-27b-jumbo-fusion-q4 +``` + +Use `--variant qwen3.8-27b-jumbo-fusion-q4` or +`--variant qwen3.8-27b-jumbo-fusion-q8` to select a specific build. +See the [publisher's model card](https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF) +for its custom reasoning and instruction modes. + ### MiniCPM-V 4.7 1B [MiniCPM-V 4.7 1B](https://huggingface.co/openbmb/MiniCPM-V-4.7-1B) diff --git a/gallery/index.yaml b/gallery/index.yaml index 981cc3563..050705f52 100644 --- a/gallery/index.yaml +++ b/gallery/index.yaml @@ -1087,6 +1087,66 @@ model: Hemmingway-1.Q8_0.gguf template: use_tokenizer_template: true +- url: github:mudler/LocalAI/gallery/qwen3.yaml@master + name: qwen3.8-27b-jumbo-fusion-q4 + description: | + DavidAU's Jumbo Fusion edition of Qwen3.8-27B merges community fine-tunes + and includes a custom template with reasoning and instruction modes. + This Q4_K_M GGUF build supports text chat with a 32,768-token default context. + license: apache-2.0 + urls: + - https://huggingface.co/Qwen/Qwen3.8-27B + - https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF + tags: + - llm + - gguf + - cpu + - gpu + - reasoning + files: + - filename: Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q4_K_M.gguf + sha256: 10110a2cb45cf827f8deb6c17946376691697080e4ea0e3ae1922a7dd06fc713 + uri: https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF/resolve/63559b2ec40db5d17c383afb7e8574e038a1dc82/Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q4_K_M.gguf + overrides: + context_size: 32768 + parameters: + min_p: 0.05 + model: Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q4_K_M.gguf + repeat_penalty: 1 + temperature: 0.7 + top_k: 64 + top_p: 0.95 + variants: + - model: qwen3.8-27b-jumbo-fusion-q8 +- url: github:mudler/LocalAI/gallery/qwen3.yaml@master + name: qwen3.8-27b-jumbo-fusion-q8 + description: | + DavidAU's Jumbo Fusion edition of Qwen3.8-27B merges community fine-tunes + and includes a custom template with reasoning and instruction modes. + This Q8_0 GGUF build supports text chat with a 32,768-token default context. + license: apache-2.0 + urls: + - https://huggingface.co/Qwen/Qwen3.8-27B + - https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF + tags: + - llm + - gguf + - cpu + - gpu + - reasoning + files: + - filename: Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q8_0.gguf + sha256: b18ae37279ebd34b92d2e15371cfe0a3aac490ccfc3f060ad9691c0aebfad6c1 + uri: https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF/resolve/63559b2ec40db5d17c383afb7e8574e038a1dc82/Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q8_0.gguf + overrides: + context_size: 32768 + parameters: + min_p: 0.05 + model: Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q8_0.gguf + repeat_penalty: 1 + temperature: 0.7 + top_k: 64 + top_p: 0.95 - url: github:mudler/LocalAI/gallery/qwen3.yaml@master name: qwen3.8-27b-turbo-brilliance-50x-q4 description: | diff --git a/gallery/models/index.yaml b/gallery/models/index.yaml index da124ab3a..b62c1868e 100644 --- a/gallery/models/index.yaml +++ b/gallery/models/index.yaml @@ -1,9 +1,6 @@ -# Model library: the models of the model gallery in the library entry shape. The -# gallery, gallery/index.yaml, stays the source of truth. This file is rendered -# with scripts/models/cmd/modelcatalog; render it again after the gallery -# changes and never edit it by hand. CI fails when it is stale. A build from -# master reads it as the default model source; a tagged build reads the gallery -# until the release cutover. +# Model library: the source of truth for models in the gallery. +# Edit this file, then generate gallery/index.yaml with: +# go run ./scripts/models/cmd/modelcatalog --library gallery/models/index.yaml --out gallery/index.yaml - kind: model id: adavar-3b-q4 name: adavar-3b-q4 @@ -1288,6 +1285,82 @@ model: Hemmingway-1.Q8_0.gguf template: use_tokenizer_template: true +- kind: model + id: qwen3.8-27b-jumbo-fusion-q4 + name: qwen3.8-27b-jumbo-fusion-q4 + version: 1.0.0 + description: | + DavidAU's Jumbo Fusion edition of Qwen3.8-27B merges community fine-tunes + and includes a custom template with reasoning and instruction modes. + This Q4_K_M GGUF build supports text chat with a 32,768-token default context. + license: apache-2.0 + urls: + - https://huggingface.co/Qwen/Qwen3.8-27B + - https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF + tags: + - llm + - gguf + - cpu + - gpu + - reasoning + channel: gallery + policy: + requireSignature: false + schemaVersion: 1 + model: + url: github:mudler/LocalAI/gallery/qwen3.yaml@master + files: + - filename: Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q4_K_M.gguf + sha256: 10110a2cb45cf827f8deb6c17946376691697080e4ea0e3ae1922a7dd06fc713 + uri: https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF/resolve/63559b2ec40db5d17c383afb7e8574e038a1dc82/Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q4_K_M.gguf + overrides: + context_size: 32768 + parameters: + min_p: 0.05 + model: Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q4_K_M.gguf + repeat_penalty: 1 + temperature: 0.7 + top_k: 64 + top_p: 0.95 + variants: + - model: qwen3.8-27b-jumbo-fusion-q8 +- kind: model + id: qwen3.8-27b-jumbo-fusion-q8 + name: qwen3.8-27b-jumbo-fusion-q8 + version: 1.0.0 + description: | + DavidAU's Jumbo Fusion edition of Qwen3.8-27B merges community fine-tunes + and includes a custom template with reasoning and instruction modes. + This Q8_0 GGUF build supports text chat with a 32,768-token default context. + license: apache-2.0 + urls: + - https://huggingface.co/Qwen/Qwen3.8-27B + - https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF + tags: + - llm + - gguf + - cpu + - gpu + - reasoning + channel: gallery + policy: + requireSignature: false + schemaVersion: 1 + model: + url: github:mudler/LocalAI/gallery/qwen3.yaml@master + files: + - filename: Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q8_0.gguf + sha256: b18ae37279ebd34b92d2e15371cfe0a3aac490ccfc3f060ad9691c0aebfad6c1 + uri: https://huggingface.co/DavidAU/Qwen3.8-27B-Jumbo-Fusion-IQ-747-900-Turbo-Brilliance-50X-Reasoning-Instruct-NEO-MAX-MTP-GGUF/resolve/63559b2ec40db5d17c383afb7e8574e038a1dc82/Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q8_0.gguf + overrides: + context_size: 32768 + parameters: + min_p: 0.05 + model: Qwen3.8-27B-Jumbo-FIQ-747-900-T-Brill-50X-NEO-MTP-Q8_0.gguf + repeat_penalty: 1 + temperature: 0.7 + top_k: 64 + top_p: 0.95 - kind: model id: qwen3.8-27b-turbo-brilliance-50x-q4 name: qwen3.8-27b-turbo-brilliance-50x-q4