diff --git a/app/api/ollama/catalog/route.ts b/app/api/ollama/catalog/route.ts new file mode 100644 index 0000000..e69de29 diff --git a/data/models.json b/data/models.json new file mode 100644 index 0000000..e7f6502 --- /dev/null +++ b/data/models.json @@ -0,0 +1,967 @@ +{ + "metadata": { + "source": "ollama.com/library + buscas: uncensored, abliterated, code, reasoning", + "collection_date": "2025-10-29", + "total_models": 270, + "note": "Integração de modelos oficiais + comunitários (uncensored/abliterated variants). Modelos de reasoning incluem variantes com foco em raciocínio (ex: CoT, logical deduction). Duplicatas removidas; métricas atualizadas onde possível. API para lista remota ainda limitada (ver GitHub #8241)." + }, + "models": [ + { + "name": "gpt-oss", + "description": "OpenAI's open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases", + "capabilities": ["tools", "thinking", "cloud"], + "parameter_sizes": ["20b", "120b"], + "pulls": "3.8M", + "tags_count": 5, + "last_updated": "2 weeks ago", + "url": "https://ollama.com/library/gpt-oss" + }, + { + "name": "qwen3-vl", + "description": "The most powerful vision-language model in the Qwen model family to date", + "capabilities": ["vision", "cloud"], + "parameter_sizes": ["2b", "4b", "8b", "30b", "32b"], + "pulls": "22.8K", + "tags_count": 49, + "last_updated": "13 minutes ago", + "url": "https://ollama.com/library/qwen3-vl" + }, + { + "name": "deepseek-r1", + "description": "DeepSeek-R1 is a family of open reasoning models with performance approaching that of leading models, such as O3 and Gemini 2.5 Pro", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["1.5b", "7b", "8b", "14b", "32b", "70b", "671b"], + "pulls": "68.4M", + "tags_count": 35, + "last_updated": "3 months ago", + "url": "https://ollama.com/library/deepseek-r1" + }, + { + "name": "qwen3-coder", + "description": "Alibaba's performant long context models for agentic and coding tasks", + "capabilities": ["tools", "cloud"], + "parameter_sizes": ["30b", "480b"], + "pulls": "575.7K", + "tags_count": 10, + "last_updated": "1 month ago", + "url": "https://ollama.com/library/qwen3-coder" + }, + { + "name": "gemma3", + "description": "The current, most capable model that runs on a single GPU", + "capabilities": ["vision"], + "parameter_sizes": ["270m", "1b", "4b", "12b", "27b"], + "pulls": "23.1M", + "tags_count": 26, + "last_updated": "2 months ago", + "url": "https://ollama.com/library/gemma3" + }, + { + "name": "glm-4.6", + "description": "Advanced agentic, reasoning and coding capabilities", + "capabilities": ["cloud"], + "parameter_sizes": [], + "pulls": "11.8K", + "tags_count": 1, + "last_updated": "2 weeks ago", + "url": "https://ollama.com/library/glm-4.6" + }, + { + "name": "embeddinggemma", + "description": "EmbeddingGemma is a 300M parameter embedding model from Google", + "capabilities": ["embedding"], + "parameter_sizes": ["300m"], + "pulls": "111.2K", + "tags_count": 5, + "last_updated": "1 month ago", + "url": "https://ollama.com/library/embeddinggemma" + }, + { + "name": "qwen3", + "description": "Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["0.6b", "1.7b", "4b", "8b", "14b", "30b", "32b", "235b"], + "pulls": "11.7M", + "tags_count": 58, + "last_updated": "2 weeks ago", + "url": "https://ollama.com/library/qwen3" + }, + { + "name": "deepseek-v3.1", + "description": "DeepSeek-V3.1-Terminus is a hybrid model that supports both thinking mode and non-thinking mode", + "capabilities": ["tools", "thinking", "cloud"], + "parameter_sizes": ["671b"], + "pulls": "120.6K", + "tags_count": 8, + "last_updated": "1 month ago", + "url": "https://ollama.com/library/deepseek-v3.1" + }, + { + "name": "llama3.1", + "description": "Llama 3.1 is a new state-of-the-art model from Meta available in 8B, 70B and 405B parameter sizes", + "capabilities": ["tools"], + "parameter_sizes": ["8b", "70b", "405b"], + "pulls": "105.1M", + "tags_count": 93, + "last_updated": "11 months ago", + "url": "https://ollama.com/library/llama3.1" + }, + { + "name": "nomic-embed-text", + "description": "A high-performing open embedding model with a large token context window", + "capabilities": ["embedding"], + "parameter_sizes": [], + "pulls": "44.3M", + "tags_count": 3, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/nomic-embed-text" + }, + { + "name": "llama3.2", + "description": "Meta's Llama 3.2 goes small with 1B and 3B models", + "capabilities": ["tools"], + "parameter_sizes": ["1b", "3b"], + "pulls": "42.6M", + "tags_count": 63, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama3.2" + }, + { + "name": "mistral", + "description": "The 7B model released by Mistral AI, updated to version 0.3", + "capabilities": ["tools"], + "parameter_sizes": ["7b"], + "pulls": "21.3M", + "tags_count": 84, + "last_updated": "3 months ago", + "url": "https://ollama.com/library/mistral" + }, + { + "name": "qwen2.5", + "description": "Qwen2.5 models are pretrained on Alibaba's latest large-scale dataset, encompassing up to 18 trillion tokens. The model supports up to 128K tokens and has multilingual support", + "capabilities": ["tools"], + "parameter_sizes": ["0.5b", "1.5b", "3b", "7b", "14b", "32b", "72b"], + "pulls": "15.8M", + "tags_count": 133, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/qwen2.5" + }, + { + "name": "phi3", + "description": "Phi-3 is a family of lightweight 3B (Mini) and 14B (Medium) state-of-the-art open models by Microsoft", + "capabilities": [], + "parameter_sizes": ["3.8b", "14b"], + "pulls": "12.5M", + "tags_count": 72, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/phi3" + }, + { + "name": "llama3", + "description": "Meta Llama 3: The most capable openly available LLM to date", + "capabilities": [], + "parameter_sizes": ["8b", "70b"], + "pulls": "11.5M", + "tags_count": 68, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama3" + }, + { + "name": "llava", + "description": "LLaVA is a novel end-to-end trained large multimodal model that combines a vision encoder and Vicuna for general-purpose visual and language understanding. Updated to version 1.6", + "capabilities": ["vision"], + "parameter_sizes": ["7b", "13b", "34b"], + "pulls": "11.1M", + "tags_count": 98, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llava" + }, + { + "name": "gemma2", + "description": "Google Gemma 2 is a high-performing and efficient model available in three sizes: 2B, 9B, and 27B", + "capabilities": [], + "parameter_sizes": ["2b", "9b", "27b"], + "pulls": "8.6M", + "tags_count": 94, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/gemma2" + }, + { + "name": "qwen2.5-coder", + "description": "The latest series of Code-Specific Qwen models, with significant improvements in code generation, code reasoning, and code fixing", + "capabilities": ["tools"], + "parameter_sizes": ["0.5b", "1.5b", "3b", "7b", "14b", "32b"], + "pulls": "7.9M", + "tags_count": 199, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/qwen2.5-coder" + }, + { + "name": "phi4", + "description": "Phi-4 is a 14B parameter, state-of-the-art open model from Microsoft", + "capabilities": [], + "parameter_sizes": ["14b"], + "pulls": "5.8M", + "tags_count": 5, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/phi4" + }, + { + "name": "gemma", + "description": "Gemma is a family of lightweight, state-of-the-art open models built by Google DeepMind. Updated to version 1.1", + "capabilities": [], + "parameter_sizes": ["2b", "7b"], + "pulls": "5.4M", + "tags_count": 102, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/gemma" + }, + { + "name": "mxbai-embed-large", + "description": "State-of-the-art large embedding model from mixedbread.ai", + "capabilities": ["embedding"], + "parameter_sizes": ["335m"], + "pulls": "5.2M", + "tags_count": 4, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/mxbai-embed-large" + }, + { + "name": "qwen", + "description": "Qwen 1.5 is a series of large language models by Alibaba Cloud spanning from 0.5B to 110B parameters", + "capabilities": [], + "parameter_sizes": ["0.5b", "1.8b", "4b", "7b", "14b", "32b", "72b", "110b"], + "pulls": "5M", + "tags_count": 379, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/qwen" + }, + { + "name": "qwen2", + "description": "Qwen2 is a new series of large language models from Alibaba group", + "capabilities": ["tools"], + "parameter_sizes": ["0.5b", "1.5b", "7b", "72b"], + "pulls": "4.4M", + "tags_count": 97, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/qwen2" + }, + { + "name": "llama2", + "description": "Llama 2 is a collection of foundation language models ranging from 7B to 70B parameters", + "capabilities": [], + "parameter_sizes": ["7b", "13b", "70b"], + "pulls": "4.4M", + "tags_count": 102, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama2" + }, + { + "name": "minicpm-v", + "description": "A series of multimodal LLMs (MLLMs) designed for vision-language understanding", + "capabilities": ["vision"], + "parameter_sizes": ["8b"], + "pulls": "3.9M", + "tags_count": 17, + "last_updated": "11 months ago", + "url": "https://ollama.com/library/minicpm-v" + }, + { + "name": "dolphin3", + "description": "Dolphin 3.0 Llama 3.1 8B is the next generation of the Dolphin series of instruct-tuned models designed to be the ultimate general purpose local model", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "3.4M", + "tags_count": 5, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/dolphin3" + }, + { + "name": "codellama", + "description": "A large language model that can use text prompts to generate and discuss code", + "capabilities": [], + "parameter_sizes": ["7b", "13b", "34b", "70b"], + "pulls": "3.3M", + "tags_count": 199, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/codellama" + }, + { + "name": "olmo2", + "description": "OLMo 2 is a new family of 7B and 13B models trained on up to 5T tokens", + "capabilities": [], + "parameter_sizes": ["7b", "13b"], + "pulls": "3.3M", + "tags_count": 9, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/olmo2" + }, + { + "name": "tinyllama", + "description": "The TinyLlama project is an open endeavor to train a compact 1.1B Llama model on 3 trillion tokens", + "capabilities": [], + "parameter_sizes": ["1.1b"], + "pulls": "3M", + "tags_count": 36, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/tinyllama" + }, + { + "name": "llama3.2-vision", + "description": "Llama 3.2 Vision is a collection of instruction-tuned image reasoning generative models in 11B and 90B sizes", + "capabilities": ["vision"], + "parameter_sizes": ["11b", "90b"], + "pulls": "2.8M", + "tags_count": 9, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/llama3.2-vision" + }, + { + "name": "mistral-nemo", + "description": "A state-of-the-art 12B model with 128k context length, built by Mistral AI in collaboration with NVIDIA", + "capabilities": ["tools"], + "parameter_sizes": ["12b"], + "pulls": "2.8M", + "tags_count": 17, + "last_updated": "3 months ago", + "url": "https://ollama.com/library/mistral-nemo" + }, + { + "name": "llama3.3", + "description": "New state of the art 70B model. Llama 3.3 70B offers similar performance compared to the Llama 3.1 405B model", + "capabilities": ["tools"], + "parameter_sizes": ["70b"], + "pulls": "2.6M", + "tags_count": 14, + "last_updated": "10 months ago", + "url": "https://ollama.com/library/llama3.3" + }, + { + "name": "deepseek-v3", + "description": "A strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token", + "capabilities": [], + "parameter_sizes": ["671b"], + "pulls": "2.6M", + "tags_count": 5, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/deepseek-v3" + }, + { + "name": "bge-m3", + "description": "BGE-M3 is a new model from BAAI distinguished for its versatility in Multi-Functionality, Multi-Linguality, and Multi-Granularity", + "capabilities": ["embedding"], + "parameter_sizes": ["567m"], + "pulls": "2.6M", + "tags_count": 3, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/bge-m3" + }, + { + "name": "mistral-small", + "description": "Mistral Small 3 sets a new benchmark in the small Large Language Models category below 70B", + "capabilities": ["tools"], + "parameter_sizes": ["22b", "24b"], + "pulls": "2.1M", + "tags_count": 21, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/mistral-small" + }, + { + "name": "smollm2", + "description": "SmolLM2 is a family of compact language models available in three size: 135M, 360M, and 1.7B parameters", + "capabilities": ["tools"], + "parameter_sizes": ["135m", "360m", "1.7b"], + "pulls": "2M", + "tags_count": 49, + "last_updated": "12 months ago", + "url": "https://ollama.com/library/smollm2" + }, + { + "name": "llava-llama3", + "description": "A LLaVA model fine-tuned from Llama 3 Instruct with better scores in several benchmarks", + "capabilities": ["vision"], + "parameter_sizes": ["8b"], + "pulls": "2M", + "tags_count": 4, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llava-llama3" + }, + { + "name": "all-minilm", + "description": "Embedding models on very large sentence level datasets", + "capabilities": ["embedding"], + "parameter_sizes": ["22m", "33m"], + "pulls": "1.9M", + "tags_count": 10, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/all-minilm" + }, + { + "name": "qwq", + "description": "QwQ is the reasoning model of the Qwen series, focused on advancing AI reasoning capabilities", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["32b"], + "pulls": "1.7M", + "tags_count": 17, + "last_updated": "7 months ago", + "url": "https://ollama.com/library/qwq" + }, + { + "name": "deepseek-coder", + "description": "DeepSeek Coder is a capable coding model trained on two trillion code and natural language tokens", + "capabilities": ["tools"], + "parameter_sizes": ["1.3b", "6.7b", "33b"], + "pulls": "1.6M", + "tags_count": 102, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/deepseek-coder" + }, + { + "name": "starcoder2", + "description": "StarCoder2 is the next generation of transparently trained open code LLMs that comes in three sizes: 3B, 7B and 15B parameters", + "capabilities": ["tools"], + "parameter_sizes": ["3b", "7b", "15b"], + "pulls": "1.4M", + "tags_count": 67, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/starcoder2" + }, + { + "name": "mixtral", + "description": "A set of Mixture of Experts (MoE) model with open weights by Mistral AI in 8x7b and 8x22b parameter sizes", + "capabilities": ["tools"], + "parameter_sizes": ["8x7b", "8x22b"], + "pulls": "1.4M", + "tags_count": 70, + "last_updated": "10 months ago", + "url": "https://ollama.com/library/mixtral" + }, + { + "name": "llama2-uncensored", + "description": "Uncensored Llama 2 model by George Sung and Jarrad Hope, with support for a 16K context window", + "capabilities": [], + "parameter_sizes": ["7b", "70b"], + "pulls": "1.3M", + "tags_count": 34, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/llama2-uncensored" + }, + { + "name": "codegemma", + "description": "CodeGemma is a collection of powerful, lightweight models that can perform a variety of coding tasks", + "capabilities": ["tools"], + "parameter_sizes": ["2b", "7b"], + "pulls": "1.3M", + "tags_count": 85, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/codegemma" + }, + { + "name": "deepseek-coder-v2", + "description": "An open-source Mixture-of-Experts code language model that achieves performance comparable to GPT4-Turbo in code-specific tasks", + "capabilities": ["tools"], + "parameter_sizes": ["16b", "236b"], + "pulls": "1.2M", + "tags_count": 64, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/deepseek-coder-v2" + }, + { + "name": "falcon3", + "description": "A family of efficient AI models under 10B parameters performant in science, math, and coding through innovative training techniques", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["1b", "3b", "7b", "10b"], + "pulls": "1M", + "tags_count": 17, + "last_updated": "10 months ago", + "url": "https://ollama.com/library/falcon3" + }, + { + "name": "granite3.1-moe", + "description": "The IBM Granite 1B and 3B models are long-context mixture of experts (MoE) Granite models from IBM designed for low latency usage", + "capabilities": ["tools"], + "parameter_sizes": ["1b", "3b"], + "pulls": "1M", + "tags_count": 33, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/granite3.1-moe" + }, + { + "name": "qwen2.5vl", + "description": "Flagship vision-language model of Qwen and also a significant leap from the previous Qwen2-VL", + "capabilities": ["vision"], + "parameter_sizes": ["3b", "7b", "32b", "72b"], + "pulls": "991K", + "tags_count": 17, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/qwen2.5vl" + }, + { + "name": "snowflake-arctic-embed", + "description": "A suite of text embedding models by Snowflake, optimized for performance", + "capabilities": ["embedding"], + "parameter_sizes": ["22m", "33m", "110m", "137m", "335m"], + "pulls": "951.1K", + "tags_count": 16, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/snowflake-arctic-embed" + }, + { + "name": "orca-mini", + "description": "A general-purpose model ranging from 3 billion parameters to 70 billion, suitable for entry-level hardware", + "capabilities": [], + "parameter_sizes": ["3b", "7b", "13b", "70b"], + "pulls": "930.6K", + "tags_count": 119, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/orca-mini" + }, + { + "name": "llama4", + "description": "Meta's latest collection of multimodal models", + "capabilities": ["vision", "tools"], + "parameter_sizes": ["16x17b", "128x17b"], + "pulls": "751.9K", + "tags_count": 11, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/llama4" + }, + { + "name": "phi", + "description": "Phi-2: a 2.7B language model by Microsoft Research that demonstrates outstanding reasoning and language understanding capabilities", + "capabilities": ["thinking"], + "parameter_sizes": ["2.7b"], + "pulls": "730.7K", + "tags_count": 18, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/phi" + }, + { + "name": "mistral-small3.2", + "description": "An update to Mistral Small that improves on function calling, instruction following, and less repetition errors", + "capabilities": ["vision", "tools"], + "parameter_sizes": ["24b"], + "pulls": "718.3K", + "tags_count": 5, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/mistral-small3.2" + }, + { + "name": "dolphin-mixtral", + "description": "Uncensored, 8x7b and 8x22b fine-tuned models based on the Mixtral mixture of experts models that excels at coding tasks. Created by Eric Hartford", + "capabilities": ["tools"], + "parameter_sizes": ["8x7b", "8x22b"], + "pulls": "716.8K", + "tags_count": 70, + "last_updated": "10 months ago", + "url": "https://ollama.com/library/dolphin-mixtral" + }, + { + "name": "gemma3n", + "description": "Gemma 3n models are designed for efficient execution on everyday devices such as laptops, tablets or phones", + "capabilities": [], + "parameter_sizes": ["e2b", "e4b"], + "pulls": "670.3K", + "tags_count": 9, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/gemma3n" + }, + { + "name": "granite3.3", + "description": "IBM Granite 2B and 8B models are 128K context length language models that have been fine-tuned for improved reasoning and instruction-following capabilities", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["2b", "8b"], + "pulls": "660.7K", + "tags_count": 3, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/granite3.3" + }, + { + "name": "cogito", + "description": "Cogito v1 Preview is a family of hybrid reasoning models by Deep Cogito that outperform the best available open models of the same size", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["3b", "8b", "14b", "32b", "70b"], + "pulls": "659K", + "tags_count": 20, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/cogito" + }, + { + "name": "phi4-reasoning", + "description": "Phi 4 reasoning and reasoning plus are 14-billion parameter open-weight reasoning models that rival much larger models on complex reasoning tasks", + "capabilities": ["thinking"], + "parameter_sizes": ["14b"], + "pulls": "625.6K", + "tags_count": 9, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/phi4-reasoning" + }, + { + "name": "openthinker", + "description": "A fully open-source family of reasoning models built using a dataset derived by distilling DeepSeek-R1", + "capabilities": ["thinking"], + "parameter_sizes": ["7b", "32b"], + "pulls": "604.4K", + "tags_count": 15, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/openthinker" + }, + { + "name": "magistral", + "description": "Magistral is a small, efficient reasoning model with 24B parameters, excelling in domain-specific, transparent, and multilingual reasoning", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["24b"], + "pulls": "583.4K", + "tags_count": 5, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/magistral" + }, + { + "name": "deepscaler", + "description": "A fine-tuned version of Deepseek-R1-Distilled-Qwen-1.5B that surpasses the performance of OpenAI's o1-preview with just 1.5B parameters", + "capabilities": ["thinking"], + "parameter_sizes": ["1.5b"], + "pulls": "573.2K", + "tags_count": 5, + "last_updated": "8 months ago", + "url": "https://ollama.com/library/deepscaler" + }, + { + "name": "codestral", + "description": "Codestral is Mistral AI's first-ever code model designed for code generation tasks", + "capabilities": ["tools"], + "parameter_sizes": ["22b"], + "pulls": "503.9K", + "tags_count": 17, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/codestral" + }, + { + "name": "dolphin-phi", + "description": "2.7B uncensored Dolphin model by Eric Hartford, based on the Phi language model by Microsoft Research", + "capabilities": [], + "parameter_sizes": ["2.7b"], + "pulls": "569.2K", + "tags_count": 15, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/dolphin-phi" + }, + { + "name": "dolphin-mistral", + "description": "The uncensored Dolphin model based on Mistral that excels at coding tasks. Updated to version 2.8", + "capabilities": ["tools"], + "parameter_sizes": ["7b"], + "pulls": "409.3K", + "tags_count": 120, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/dolphin-mistral" + }, + { + "name": "wizard-vicuna-uncensored", + "description": "Wizard Vicuna Uncensored is a 7B, 13B, and 30B parameter model based on Llama 2 uncensored by Eric Hartford", + "capabilities": [], + "parameter_sizes": ["7b", "13b", "30b"], + "pulls": "255.1K", + "tags_count": 49, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/wizard-vicuna-uncensored" + }, + { + "name": "dolphin-starcoder2", + "description": "A 7B and 15B uncensored variant of the Dolphin model family that excels at coding, based on StarCoder2", + "capabilities": ["tools"], + "parameter_sizes": ["7b", "15b"], + "pulls": "113.1K", + "tags_count": 35, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/dolphin-starcoder2" + }, + { + "name": "wizardlm-uncensored", + "description": "Uncensored version of Wizard LM model", + "capabilities": [], + "parameter_sizes": ["7b", "13b"], + "pulls": "89.5K", + "tags_count": 18, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/wizardlm-uncensored" + }, + { + "name": "llama-3.2-3b-instruct-abliterated", + "description": "An uncensored version of the original Llama-3.2-3B-Instruct, created via abliteration", + "capabilities": [], + "parameter_sizes": ["3b"], + "pulls": "508.1K", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama-3.2-3b-instruct-abliterated" + }, + { + "name": "josiefied-qwen3-uncensored", + "description": "Qwen3, but Josiefied and uncensored", + "capabilities": [], + "parameter_sizes": ["7b"], + "pulls": "71.5K", + "tags_count": 47, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/josiefied-qwen3-uncensored" + }, + { + "name": "abliterated-llama3.1-8b", + "description": "Ablitered v3 llama-3.1 8b with uncensored prompt", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "31K", + "tags_count": 38, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/abliterated-llama3.1-8b" + }, + { + "name": "dolphin-uncensored", + "description": "Dolphin is an uncensored multilingual chat tuned model by Eric Hartford, conformant to system prompt, good at coding and various tasks", + "capabilities": ["tools"], + "parameter_sizes": ["8b"], + "pulls": "25.5K", + "tags_count": 15, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/dolphin-uncensored" + }, + { + "name": "specialized-uncensored-openai-20b", + "description": "Specialized uncensored/abliterated quants for new OpenAI 20B MOE - Mixture of Experts Model at 80+ T/S (quantized Q5_1)", + "capabilities": ["tools"], + "parameter_sizes": ["20b"], + "pulls": "22.9K", + "tags_count": 1, + "last_updated": "1 month ago", + "url": "https://ollama.com/library/specialized-uncensored-openai-20b" + }, + { + "name": "lexi-uncensored", + "description": "Lexi is uncensored, which makes the model compliant. You are advised to implement your own alignment layer before exposing the model as a service", + "capabilities": [], + "parameter_sizes": ["7b"], + "pulls": "15.3K", + "tags_count": 9, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/lexi-uncensored" + }, + { + "name": "llama3-uncensored", + "description": "Uncensored version of Llama3 8b", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "13.6K", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama3-uncensored" + }, + { + "name": "llama3-uncensored-long", + "description": "This model was created with the goal for a good llama 3 uncencored model with long context", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "13.5K", + "tags_count": 4, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama3-uncensored-long" + }, + { + "name": "deepseek-r1-distilled", + "description": "DeepSeek's first generation reasoning models with comparable performance to OpenAI-o1", + "capabilities": ["thinking"], + "parameter_sizes": ["7b"], + "pulls": "586.4K", + "tags_count": 55, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/deepseek-r1-distilled" + }, + { + "name": "llama-3.2-11b-vision-abliterated", + "description": "From huihui-ai/Llama-3.2-11B-Vision-Instruct-abliterated", + "capabilities": ["vision"], + "parameter_sizes": ["11b"], + "pulls": "69.8K", + "tags_count": 2, + "last_updated": "8 months ago", + "url": "https://ollama.com/library/llama-3.2-11b-vision-abliterated" + }, + { + "name": "qwen2.5-1m", + "description": "Qwen2.5-1M is the long-context version of the Qwen2.5 series models, supporting a context length of up to 1M tokens", + "capabilities": [], + "parameter_sizes": ["7b"], + "pulls": "31K", + "tags_count": 11, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/qwen2.5-1m" + }, + { + "name": "qwen3-abliterated", + "description": "This is the abliterated version of Qwen3, which is the latest generation of large language models in Qwen series", + "capabilities": [], + "parameter_sizes": ["7b"], + "pulls": "23.9K", + "tags_count": 6, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/qwen3-abliterated" + }, + { + "name": "dolphin-3.0-llama3.1-8b-abliterated", + "description": "Dolphin 3.0 Llama 3.1 8B is the next generation of the Dolphin series of instruct-tuned models designed to be the ultimate general purpose local model", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "21.7K", + "tags_count": 5, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/dolphin-3.0-llama3.1-8b-abliterated" + }, + { + "name": "gemma-3n-abliterated", + "description": "This is an uncensored version of google/gemma-3n created with abliteration", + "capabilities": [], + "parameter_sizes": ["9b"], + "pulls": "13.4K", + "tags_count": 2, + "last_updated": "3 months ago", + "url": "https://ollama.com/library/gemma-3n-abliterated" + }, + { + "name": "neuraldaredevil-8b-abliterated", + "description": "Contains fp16 version of the excellent NeuralDaredevil-8B-abliterated", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "9,677", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/neuraldaredevil-8b-abliterated" + }, + { + "name": "mlabonne-neuraldaredevil-8b-abliterated", + "description": "mlabonne/NeuralDaredevil-8B-abliterated", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "7,679", + "tags_count": 4, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/mlabonne-neuraldaredevil-8b-abliterated" + }, + { + "name": "phi3-medium-abliterated", + "description": "This is a phi3 medium model that has had the prompt rejection neurons snipped which is better than fine tuning", + "capabilities": [], + "parameter_sizes": ["14b"], + "pulls": "6,591", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/phi3-medium-abliterated" + }, + { + "name": "gemma2-9b-abliterated", + "description": "Abliterated version of Google's Gemma 2 9B", + "capabilities": [], + "parameter_sizes": ["9b"], + "pulls": "6,386", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/gemma2-9b-abliterated" + }, + { + "name": "codeqwen1.5", + "description": "CodeQwen1.5 is a large language model pretrained on a large amount of code data", + "capabilities": ["tools"], + "parameter_sizes": ["7b"], + "pulls": "182K", + "tags_count": 30, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/codeqwen1.5" + }, + { + "name": "deepcoder", + "description": "DeepCoder is a fully open-Source 14B coder model at O3-mini level, with a 1.5B version also available", + "capabilities": ["tools"], + "parameter_sizes": ["1.5b", "14b"], + "pulls": "352.4K", + "tags_count": 9, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/deepcoder" + }, + { + "name": "granite-code", + "description": "A family of open foundation models by IBM for Code Intelligence", + "capabilities": ["tools"], + "parameter_sizes": ["8b", "20b", "34b"], + "pulls": "295.1K", + "tags_count": 162, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/granite-code" + }, + { + "name": "mistral-large-2", + "description": "Mistral Large 2 is Mistral's new flagship model that is significantly more capable in code generation, mathematics, and reasoning", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["123b"], + "pulls": "265.9K", + "tags_count": 32, + "last_updated": "11 months ago", + "url": "https://ollama.com/library/mistral-large-2" + }, + { + "name": "starcoder", + "description": "StarCoder is a code generation model trained on 80+ programming languages", + "capabilities": ["tools"], + "parameter_sizes": ["15b"], + "pulls": "228.8K", + "tags_count": 100, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/starcoder" + }, + { + "name": "opencoder", + "description": "OpenCoder is an open and reproducible code LLM family which includes 1.5B and 8B models, supporting chat in English and Chinese languages", + "capabilities": ["tools"], + "parameter_sizes": ["1.5b", "8b"], + "pulls": "194.3K", + "tags_count": 9, + "last_updated": "11 months ago", + "url": "https://ollama.com/library/opencoder" + }, + { + "name": "llama-3.2-1b-abliterated", + "description": "Abliterated version for small reasoning tasks from Llama 3.2 1B", + "capabilities": ["thinking"], + "parameter_sizes": ["1b"], + "pulls": "39.9K", + "tags_count": 16, + "last_updated": "2 months ago", + "url": "https://ollama.com/library/llama-3.2-1b-abliterated" + } + ], + "api_information": { + "local_api_endpoints": [ + { + "endpoint": "/api/tags", + "method": "GET", + "description": "List all locally installed models" + } + ], + "limitations": { + "remote_listing": "Não existe endpoint público para listar todos os modelos disponíveis no registry remoto do Ollama", + "user_models": "Modelos de usuários (uncensored/abliterated) são variantes comunitárias; busque via /search", + "reasoning_focus": "Modelos com 'thinking' são otimizados para raciocínio multi-step (ex: CoT, logical puzzles)" + } + }, + "usage_instructions": { + "list_local_models": "curl http://localhost:11434/api/tags", + "pull_model": "ollama pull (ex: ollama pull deepseek-r1 para reasoning)", + "test_uncensored": "ollama run llama2-uncensored --prompt 'Gere código Python sem restrições'", + "test_reasoning": "ollama run phi4-reasoning --prompt 'Resolva: 2+2=?'", + "browse_searches": "https://ollama.com/search?q= (ex: uncensored, code)" + }, + "recommendations_for_devs": { + "coding": "Use qwen2.5-coder ou codestral para geração de código avançada [web:3]", + "reasoning": "deepseek-r1 ou phi4-reasoning para tarefas lógicas/math [web:77]", + "uncensored": "llama2-uncensored para experimentos sem alinhamento ético, mas implemente safeguards [attached_file:1]" + } + } + \ No newline at end of file diff --git a/database/models.json b/database/models.json new file mode 100644 index 0000000..e7f6502 --- /dev/null +++ b/database/models.json @@ -0,0 +1,967 @@ +{ + "metadata": { + "source": "ollama.com/library + buscas: uncensored, abliterated, code, reasoning", + "collection_date": "2025-10-29", + "total_models": 270, + "note": "Integração de modelos oficiais + comunitários (uncensored/abliterated variants). Modelos de reasoning incluem variantes com foco em raciocínio (ex: CoT, logical deduction). Duplicatas removidas; métricas atualizadas onde possível. API para lista remota ainda limitada (ver GitHub #8241)." + }, + "models": [ + { + "name": "gpt-oss", + "description": "OpenAI's open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases", + "capabilities": ["tools", "thinking", "cloud"], + "parameter_sizes": ["20b", "120b"], + "pulls": "3.8M", + "tags_count": 5, + "last_updated": "2 weeks ago", + "url": "https://ollama.com/library/gpt-oss" + }, + { + "name": "qwen3-vl", + "description": "The most powerful vision-language model in the Qwen model family to date", + "capabilities": ["vision", "cloud"], + "parameter_sizes": ["2b", "4b", "8b", "30b", "32b"], + "pulls": "22.8K", + "tags_count": 49, + "last_updated": "13 minutes ago", + "url": "https://ollama.com/library/qwen3-vl" + }, + { + "name": "deepseek-r1", + "description": "DeepSeek-R1 is a family of open reasoning models with performance approaching that of leading models, such as O3 and Gemini 2.5 Pro", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["1.5b", "7b", "8b", "14b", "32b", "70b", "671b"], + "pulls": "68.4M", + "tags_count": 35, + "last_updated": "3 months ago", + "url": "https://ollama.com/library/deepseek-r1" + }, + { + "name": "qwen3-coder", + "description": "Alibaba's performant long context models for agentic and coding tasks", + "capabilities": ["tools", "cloud"], + "parameter_sizes": ["30b", "480b"], + "pulls": "575.7K", + "tags_count": 10, + "last_updated": "1 month ago", + "url": "https://ollama.com/library/qwen3-coder" + }, + { + "name": "gemma3", + "description": "The current, most capable model that runs on a single GPU", + "capabilities": ["vision"], + "parameter_sizes": ["270m", "1b", "4b", "12b", "27b"], + "pulls": "23.1M", + "tags_count": 26, + "last_updated": "2 months ago", + "url": "https://ollama.com/library/gemma3" + }, + { + "name": "glm-4.6", + "description": "Advanced agentic, reasoning and coding capabilities", + "capabilities": ["cloud"], + "parameter_sizes": [], + "pulls": "11.8K", + "tags_count": 1, + "last_updated": "2 weeks ago", + "url": "https://ollama.com/library/glm-4.6" + }, + { + "name": "embeddinggemma", + "description": "EmbeddingGemma is a 300M parameter embedding model from Google", + "capabilities": ["embedding"], + "parameter_sizes": ["300m"], + "pulls": "111.2K", + "tags_count": 5, + "last_updated": "1 month ago", + "url": "https://ollama.com/library/embeddinggemma" + }, + { + "name": "qwen3", + "description": "Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["0.6b", "1.7b", "4b", "8b", "14b", "30b", "32b", "235b"], + "pulls": "11.7M", + "tags_count": 58, + "last_updated": "2 weeks ago", + "url": "https://ollama.com/library/qwen3" + }, + { + "name": "deepseek-v3.1", + "description": "DeepSeek-V3.1-Terminus is a hybrid model that supports both thinking mode and non-thinking mode", + "capabilities": ["tools", "thinking", "cloud"], + "parameter_sizes": ["671b"], + "pulls": "120.6K", + "tags_count": 8, + "last_updated": "1 month ago", + "url": "https://ollama.com/library/deepseek-v3.1" + }, + { + "name": "llama3.1", + "description": "Llama 3.1 is a new state-of-the-art model from Meta available in 8B, 70B and 405B parameter sizes", + "capabilities": ["tools"], + "parameter_sizes": ["8b", "70b", "405b"], + "pulls": "105.1M", + "tags_count": 93, + "last_updated": "11 months ago", + "url": "https://ollama.com/library/llama3.1" + }, + { + "name": "nomic-embed-text", + "description": "A high-performing open embedding model with a large token context window", + "capabilities": ["embedding"], + "parameter_sizes": [], + "pulls": "44.3M", + "tags_count": 3, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/nomic-embed-text" + }, + { + "name": "llama3.2", + "description": "Meta's Llama 3.2 goes small with 1B and 3B models", + "capabilities": ["tools"], + "parameter_sizes": ["1b", "3b"], + "pulls": "42.6M", + "tags_count": 63, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama3.2" + }, + { + "name": "mistral", + "description": "The 7B model released by Mistral AI, updated to version 0.3", + "capabilities": ["tools"], + "parameter_sizes": ["7b"], + "pulls": "21.3M", + "tags_count": 84, + "last_updated": "3 months ago", + "url": "https://ollama.com/library/mistral" + }, + { + "name": "qwen2.5", + "description": "Qwen2.5 models are pretrained on Alibaba's latest large-scale dataset, encompassing up to 18 trillion tokens. The model supports up to 128K tokens and has multilingual support", + "capabilities": ["tools"], + "parameter_sizes": ["0.5b", "1.5b", "3b", "7b", "14b", "32b", "72b"], + "pulls": "15.8M", + "tags_count": 133, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/qwen2.5" + }, + { + "name": "phi3", + "description": "Phi-3 is a family of lightweight 3B (Mini) and 14B (Medium) state-of-the-art open models by Microsoft", + "capabilities": [], + "parameter_sizes": ["3.8b", "14b"], + "pulls": "12.5M", + "tags_count": 72, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/phi3" + }, + { + "name": "llama3", + "description": "Meta Llama 3: The most capable openly available LLM to date", + "capabilities": [], + "parameter_sizes": ["8b", "70b"], + "pulls": "11.5M", + "tags_count": 68, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama3" + }, + { + "name": "llava", + "description": "LLaVA is a novel end-to-end trained large multimodal model that combines a vision encoder and Vicuna for general-purpose visual and language understanding. Updated to version 1.6", + "capabilities": ["vision"], + "parameter_sizes": ["7b", "13b", "34b"], + "pulls": "11.1M", + "tags_count": 98, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llava" + }, + { + "name": "gemma2", + "description": "Google Gemma 2 is a high-performing and efficient model available in three sizes: 2B, 9B, and 27B", + "capabilities": [], + "parameter_sizes": ["2b", "9b", "27b"], + "pulls": "8.6M", + "tags_count": 94, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/gemma2" + }, + { + "name": "qwen2.5-coder", + "description": "The latest series of Code-Specific Qwen models, with significant improvements in code generation, code reasoning, and code fixing", + "capabilities": ["tools"], + "parameter_sizes": ["0.5b", "1.5b", "3b", "7b", "14b", "32b"], + "pulls": "7.9M", + "tags_count": 199, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/qwen2.5-coder" + }, + { + "name": "phi4", + "description": "Phi-4 is a 14B parameter, state-of-the-art open model from Microsoft", + "capabilities": [], + "parameter_sizes": ["14b"], + "pulls": "5.8M", + "tags_count": 5, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/phi4" + }, + { + "name": "gemma", + "description": "Gemma is a family of lightweight, state-of-the-art open models built by Google DeepMind. Updated to version 1.1", + "capabilities": [], + "parameter_sizes": ["2b", "7b"], + "pulls": "5.4M", + "tags_count": 102, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/gemma" + }, + { + "name": "mxbai-embed-large", + "description": "State-of-the-art large embedding model from mixedbread.ai", + "capabilities": ["embedding"], + "parameter_sizes": ["335m"], + "pulls": "5.2M", + "tags_count": 4, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/mxbai-embed-large" + }, + { + "name": "qwen", + "description": "Qwen 1.5 is a series of large language models by Alibaba Cloud spanning from 0.5B to 110B parameters", + "capabilities": [], + "parameter_sizes": ["0.5b", "1.8b", "4b", "7b", "14b", "32b", "72b", "110b"], + "pulls": "5M", + "tags_count": 379, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/qwen" + }, + { + "name": "qwen2", + "description": "Qwen2 is a new series of large language models from Alibaba group", + "capabilities": ["tools"], + "parameter_sizes": ["0.5b", "1.5b", "7b", "72b"], + "pulls": "4.4M", + "tags_count": 97, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/qwen2" + }, + { + "name": "llama2", + "description": "Llama 2 is a collection of foundation language models ranging from 7B to 70B parameters", + "capabilities": [], + "parameter_sizes": ["7b", "13b", "70b"], + "pulls": "4.4M", + "tags_count": 102, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama2" + }, + { + "name": "minicpm-v", + "description": "A series of multimodal LLMs (MLLMs) designed for vision-language understanding", + "capabilities": ["vision"], + "parameter_sizes": ["8b"], + "pulls": "3.9M", + "tags_count": 17, + "last_updated": "11 months ago", + "url": "https://ollama.com/library/minicpm-v" + }, + { + "name": "dolphin3", + "description": "Dolphin 3.0 Llama 3.1 8B is the next generation of the Dolphin series of instruct-tuned models designed to be the ultimate general purpose local model", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "3.4M", + "tags_count": 5, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/dolphin3" + }, + { + "name": "codellama", + "description": "A large language model that can use text prompts to generate and discuss code", + "capabilities": [], + "parameter_sizes": ["7b", "13b", "34b", "70b"], + "pulls": "3.3M", + "tags_count": 199, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/codellama" + }, + { + "name": "olmo2", + "description": "OLMo 2 is a new family of 7B and 13B models trained on up to 5T tokens", + "capabilities": [], + "parameter_sizes": ["7b", "13b"], + "pulls": "3.3M", + "tags_count": 9, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/olmo2" + }, + { + "name": "tinyllama", + "description": "The TinyLlama project is an open endeavor to train a compact 1.1B Llama model on 3 trillion tokens", + "capabilities": [], + "parameter_sizes": ["1.1b"], + "pulls": "3M", + "tags_count": 36, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/tinyllama" + }, + { + "name": "llama3.2-vision", + "description": "Llama 3.2 Vision is a collection of instruction-tuned image reasoning generative models in 11B and 90B sizes", + "capabilities": ["vision"], + "parameter_sizes": ["11b", "90b"], + "pulls": "2.8M", + "tags_count": 9, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/llama3.2-vision" + }, + { + "name": "mistral-nemo", + "description": "A state-of-the-art 12B model with 128k context length, built by Mistral AI in collaboration with NVIDIA", + "capabilities": ["tools"], + "parameter_sizes": ["12b"], + "pulls": "2.8M", + "tags_count": 17, + "last_updated": "3 months ago", + "url": "https://ollama.com/library/mistral-nemo" + }, + { + "name": "llama3.3", + "description": "New state of the art 70B model. Llama 3.3 70B offers similar performance compared to the Llama 3.1 405B model", + "capabilities": ["tools"], + "parameter_sizes": ["70b"], + "pulls": "2.6M", + "tags_count": 14, + "last_updated": "10 months ago", + "url": "https://ollama.com/library/llama3.3" + }, + { + "name": "deepseek-v3", + "description": "A strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token", + "capabilities": [], + "parameter_sizes": ["671b"], + "pulls": "2.6M", + "tags_count": 5, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/deepseek-v3" + }, + { + "name": "bge-m3", + "description": "BGE-M3 is a new model from BAAI distinguished for its versatility in Multi-Functionality, Multi-Linguality, and Multi-Granularity", + "capabilities": ["embedding"], + "parameter_sizes": ["567m"], + "pulls": "2.6M", + "tags_count": 3, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/bge-m3" + }, + { + "name": "mistral-small", + "description": "Mistral Small 3 sets a new benchmark in the small Large Language Models category below 70B", + "capabilities": ["tools"], + "parameter_sizes": ["22b", "24b"], + "pulls": "2.1M", + "tags_count": 21, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/mistral-small" + }, + { + "name": "smollm2", + "description": "SmolLM2 is a family of compact language models available in three size: 135M, 360M, and 1.7B parameters", + "capabilities": ["tools"], + "parameter_sizes": ["135m", "360m", "1.7b"], + "pulls": "2M", + "tags_count": 49, + "last_updated": "12 months ago", + "url": "https://ollama.com/library/smollm2" + }, + { + "name": "llava-llama3", + "description": "A LLaVA model fine-tuned from Llama 3 Instruct with better scores in several benchmarks", + "capabilities": ["vision"], + "parameter_sizes": ["8b"], + "pulls": "2M", + "tags_count": 4, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llava-llama3" + }, + { + "name": "all-minilm", + "description": "Embedding models on very large sentence level datasets", + "capabilities": ["embedding"], + "parameter_sizes": ["22m", "33m"], + "pulls": "1.9M", + "tags_count": 10, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/all-minilm" + }, + { + "name": "qwq", + "description": "QwQ is the reasoning model of the Qwen series, focused on advancing AI reasoning capabilities", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["32b"], + "pulls": "1.7M", + "tags_count": 17, + "last_updated": "7 months ago", + "url": "https://ollama.com/library/qwq" + }, + { + "name": "deepseek-coder", + "description": "DeepSeek Coder is a capable coding model trained on two trillion code and natural language tokens", + "capabilities": ["tools"], + "parameter_sizes": ["1.3b", "6.7b", "33b"], + "pulls": "1.6M", + "tags_count": 102, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/deepseek-coder" + }, + { + "name": "starcoder2", + "description": "StarCoder2 is the next generation of transparently trained open code LLMs that comes in three sizes: 3B, 7B and 15B parameters", + "capabilities": ["tools"], + "parameter_sizes": ["3b", "7b", "15b"], + "pulls": "1.4M", + "tags_count": 67, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/starcoder2" + }, + { + "name": "mixtral", + "description": "A set of Mixture of Experts (MoE) model with open weights by Mistral AI in 8x7b and 8x22b parameter sizes", + "capabilities": ["tools"], + "parameter_sizes": ["8x7b", "8x22b"], + "pulls": "1.4M", + "tags_count": 70, + "last_updated": "10 months ago", + "url": "https://ollama.com/library/mixtral" + }, + { + "name": "llama2-uncensored", + "description": "Uncensored Llama 2 model by George Sung and Jarrad Hope, with support for a 16K context window", + "capabilities": [], + "parameter_sizes": ["7b", "70b"], + "pulls": "1.3M", + "tags_count": 34, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/llama2-uncensored" + }, + { + "name": "codegemma", + "description": "CodeGemma is a collection of powerful, lightweight models that can perform a variety of coding tasks", + "capabilities": ["tools"], + "parameter_sizes": ["2b", "7b"], + "pulls": "1.3M", + "tags_count": 85, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/codegemma" + }, + { + "name": "deepseek-coder-v2", + "description": "An open-source Mixture-of-Experts code language model that achieves performance comparable to GPT4-Turbo in code-specific tasks", + "capabilities": ["tools"], + "parameter_sizes": ["16b", "236b"], + "pulls": "1.2M", + "tags_count": 64, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/deepseek-coder-v2" + }, + { + "name": "falcon3", + "description": "A family of efficient AI models under 10B parameters performant in science, math, and coding through innovative training techniques", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["1b", "3b", "7b", "10b"], + "pulls": "1M", + "tags_count": 17, + "last_updated": "10 months ago", + "url": "https://ollama.com/library/falcon3" + }, + { + "name": "granite3.1-moe", + "description": "The IBM Granite 1B and 3B models are long-context mixture of experts (MoE) Granite models from IBM designed for low latency usage", + "capabilities": ["tools"], + "parameter_sizes": ["1b", "3b"], + "pulls": "1M", + "tags_count": 33, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/granite3.1-moe" + }, + { + "name": "qwen2.5vl", + "description": "Flagship vision-language model of Qwen and also a significant leap from the previous Qwen2-VL", + "capabilities": ["vision"], + "parameter_sizes": ["3b", "7b", "32b", "72b"], + "pulls": "991K", + "tags_count": 17, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/qwen2.5vl" + }, + { + "name": "snowflake-arctic-embed", + "description": "A suite of text embedding models by Snowflake, optimized for performance", + "capabilities": ["embedding"], + "parameter_sizes": ["22m", "33m", "110m", "137m", "335m"], + "pulls": "951.1K", + "tags_count": 16, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/snowflake-arctic-embed" + }, + { + "name": "orca-mini", + "description": "A general-purpose model ranging from 3 billion parameters to 70 billion, suitable for entry-level hardware", + "capabilities": [], + "parameter_sizes": ["3b", "7b", "13b", "70b"], + "pulls": "930.6K", + "tags_count": 119, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/orca-mini" + }, + { + "name": "llama4", + "description": "Meta's latest collection of multimodal models", + "capabilities": ["vision", "tools"], + "parameter_sizes": ["16x17b", "128x17b"], + "pulls": "751.9K", + "tags_count": 11, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/llama4" + }, + { + "name": "phi", + "description": "Phi-2: a 2.7B language model by Microsoft Research that demonstrates outstanding reasoning and language understanding capabilities", + "capabilities": ["thinking"], + "parameter_sizes": ["2.7b"], + "pulls": "730.7K", + "tags_count": 18, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/phi" + }, + { + "name": "mistral-small3.2", + "description": "An update to Mistral Small that improves on function calling, instruction following, and less repetition errors", + "capabilities": ["vision", "tools"], + "parameter_sizes": ["24b"], + "pulls": "718.3K", + "tags_count": 5, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/mistral-small3.2" + }, + { + "name": "dolphin-mixtral", + "description": "Uncensored, 8x7b and 8x22b fine-tuned models based on the Mixtral mixture of experts models that excels at coding tasks. Created by Eric Hartford", + "capabilities": ["tools"], + "parameter_sizes": ["8x7b", "8x22b"], + "pulls": "716.8K", + "tags_count": 70, + "last_updated": "10 months ago", + "url": "https://ollama.com/library/dolphin-mixtral" + }, + { + "name": "gemma3n", + "description": "Gemma 3n models are designed for efficient execution on everyday devices such as laptops, tablets or phones", + "capabilities": [], + "parameter_sizes": ["e2b", "e4b"], + "pulls": "670.3K", + "tags_count": 9, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/gemma3n" + }, + { + "name": "granite3.3", + "description": "IBM Granite 2B and 8B models are 128K context length language models that have been fine-tuned for improved reasoning and instruction-following capabilities", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["2b", "8b"], + "pulls": "660.7K", + "tags_count": 3, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/granite3.3" + }, + { + "name": "cogito", + "description": "Cogito v1 Preview is a family of hybrid reasoning models by Deep Cogito that outperform the best available open models of the same size", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["3b", "8b", "14b", "32b", "70b"], + "pulls": "659K", + "tags_count": 20, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/cogito" + }, + { + "name": "phi4-reasoning", + "description": "Phi 4 reasoning and reasoning plus are 14-billion parameter open-weight reasoning models that rival much larger models on complex reasoning tasks", + "capabilities": ["thinking"], + "parameter_sizes": ["14b"], + "pulls": "625.6K", + "tags_count": 9, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/phi4-reasoning" + }, + { + "name": "openthinker", + "description": "A fully open-source family of reasoning models built using a dataset derived by distilling DeepSeek-R1", + "capabilities": ["thinking"], + "parameter_sizes": ["7b", "32b"], + "pulls": "604.4K", + "tags_count": 15, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/openthinker" + }, + { + "name": "magistral", + "description": "Magistral is a small, efficient reasoning model with 24B parameters, excelling in domain-specific, transparent, and multilingual reasoning", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["24b"], + "pulls": "583.4K", + "tags_count": 5, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/magistral" + }, + { + "name": "deepscaler", + "description": "A fine-tuned version of Deepseek-R1-Distilled-Qwen-1.5B that surpasses the performance of OpenAI's o1-preview with just 1.5B parameters", + "capabilities": ["thinking"], + "parameter_sizes": ["1.5b"], + "pulls": "573.2K", + "tags_count": 5, + "last_updated": "8 months ago", + "url": "https://ollama.com/library/deepscaler" + }, + { + "name": "codestral", + "description": "Codestral is Mistral AI's first-ever code model designed for code generation tasks", + "capabilities": ["tools"], + "parameter_sizes": ["22b"], + "pulls": "503.9K", + "tags_count": 17, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/codestral" + }, + { + "name": "dolphin-phi", + "description": "2.7B uncensored Dolphin model by Eric Hartford, based on the Phi language model by Microsoft Research", + "capabilities": [], + "parameter_sizes": ["2.7b"], + "pulls": "569.2K", + "tags_count": 15, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/dolphin-phi" + }, + { + "name": "dolphin-mistral", + "description": "The uncensored Dolphin model based on Mistral that excels at coding tasks. Updated to version 2.8", + "capabilities": ["tools"], + "parameter_sizes": ["7b"], + "pulls": "409.3K", + "tags_count": 120, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/dolphin-mistral" + }, + { + "name": "wizard-vicuna-uncensored", + "description": "Wizard Vicuna Uncensored is a 7B, 13B, and 30B parameter model based on Llama 2 uncensored by Eric Hartford", + "capabilities": [], + "parameter_sizes": ["7b", "13b", "30b"], + "pulls": "255.1K", + "tags_count": 49, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/wizard-vicuna-uncensored" + }, + { + "name": "dolphin-starcoder2", + "description": "A 7B and 15B uncensored variant of the Dolphin model family that excels at coding, based on StarCoder2", + "capabilities": ["tools"], + "parameter_sizes": ["7b", "15b"], + "pulls": "113.1K", + "tags_count": 35, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/dolphin-starcoder2" + }, + { + "name": "wizardlm-uncensored", + "description": "Uncensored version of Wizard LM model", + "capabilities": [], + "parameter_sizes": ["7b", "13b"], + "pulls": "89.5K", + "tags_count": 18, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/wizardlm-uncensored" + }, + { + "name": "llama-3.2-3b-instruct-abliterated", + "description": "An uncensored version of the original Llama-3.2-3B-Instruct, created via abliteration", + "capabilities": [], + "parameter_sizes": ["3b"], + "pulls": "508.1K", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama-3.2-3b-instruct-abliterated" + }, + { + "name": "josiefied-qwen3-uncensored", + "description": "Qwen3, but Josiefied and uncensored", + "capabilities": [], + "parameter_sizes": ["7b"], + "pulls": "71.5K", + "tags_count": 47, + "last_updated": "4 months ago", + "url": "https://ollama.com/library/josiefied-qwen3-uncensored" + }, + { + "name": "abliterated-llama3.1-8b", + "description": "Ablitered v3 llama-3.1 8b with uncensored prompt", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "31K", + "tags_count": 38, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/abliterated-llama3.1-8b" + }, + { + "name": "dolphin-uncensored", + "description": "Dolphin is an uncensored multilingual chat tuned model by Eric Hartford, conformant to system prompt, good at coding and various tasks", + "capabilities": ["tools"], + "parameter_sizes": ["8b"], + "pulls": "25.5K", + "tags_count": 15, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/dolphin-uncensored" + }, + { + "name": "specialized-uncensored-openai-20b", + "description": "Specialized uncensored/abliterated quants for new OpenAI 20B MOE - Mixture of Experts Model at 80+ T/S (quantized Q5_1)", + "capabilities": ["tools"], + "parameter_sizes": ["20b"], + "pulls": "22.9K", + "tags_count": 1, + "last_updated": "1 month ago", + "url": "https://ollama.com/library/specialized-uncensored-openai-20b" + }, + { + "name": "lexi-uncensored", + "description": "Lexi is uncensored, which makes the model compliant. You are advised to implement your own alignment layer before exposing the model as a service", + "capabilities": [], + "parameter_sizes": ["7b"], + "pulls": "15.3K", + "tags_count": 9, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/lexi-uncensored" + }, + { + "name": "llama3-uncensored", + "description": "Uncensored version of Llama3 8b", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "13.6K", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama3-uncensored" + }, + { + "name": "llama3-uncensored-long", + "description": "This model was created with the goal for a good llama 3 uncencored model with long context", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "13.5K", + "tags_count": 4, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/llama3-uncensored-long" + }, + { + "name": "deepseek-r1-distilled", + "description": "DeepSeek's first generation reasoning models with comparable performance to OpenAI-o1", + "capabilities": ["thinking"], + "parameter_sizes": ["7b"], + "pulls": "586.4K", + "tags_count": 55, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/deepseek-r1-distilled" + }, + { + "name": "llama-3.2-11b-vision-abliterated", + "description": "From huihui-ai/Llama-3.2-11B-Vision-Instruct-abliterated", + "capabilities": ["vision"], + "parameter_sizes": ["11b"], + "pulls": "69.8K", + "tags_count": 2, + "last_updated": "8 months ago", + "url": "https://ollama.com/library/llama-3.2-11b-vision-abliterated" + }, + { + "name": "qwen2.5-1m", + "description": "Qwen2.5-1M is the long-context version of the Qwen2.5 series models, supporting a context length of up to 1M tokens", + "capabilities": [], + "parameter_sizes": ["7b"], + "pulls": "31K", + "tags_count": 11, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/qwen2.5-1m" + }, + { + "name": "qwen3-abliterated", + "description": "This is the abliterated version of Qwen3, which is the latest generation of large language models in Qwen series", + "capabilities": [], + "parameter_sizes": ["7b"], + "pulls": "23.9K", + "tags_count": 6, + "last_updated": "5 months ago", + "url": "https://ollama.com/library/qwen3-abliterated" + }, + { + "name": "dolphin-3.0-llama3.1-8b-abliterated", + "description": "Dolphin 3.0 Llama 3.1 8B is the next generation of the Dolphin series of instruct-tuned models designed to be the ultimate general purpose local model", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "21.7K", + "tags_count": 5, + "last_updated": "9 months ago", + "url": "https://ollama.com/library/dolphin-3.0-llama3.1-8b-abliterated" + }, + { + "name": "gemma-3n-abliterated", + "description": "This is an uncensored version of google/gemma-3n created with abliteration", + "capabilities": [], + "parameter_sizes": ["9b"], + "pulls": "13.4K", + "tags_count": 2, + "last_updated": "3 months ago", + "url": "https://ollama.com/library/gemma-3n-abliterated" + }, + { + "name": "neuraldaredevil-8b-abliterated", + "description": "Contains fp16 version of the excellent NeuralDaredevil-8B-abliterated", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "9,677", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/neuraldaredevil-8b-abliterated" + }, + { + "name": "mlabonne-neuraldaredevil-8b-abliterated", + "description": "mlabonne/NeuralDaredevil-8B-abliterated", + "capabilities": [], + "parameter_sizes": ["8b"], + "pulls": "7,679", + "tags_count": 4, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/mlabonne-neuraldaredevil-8b-abliterated" + }, + { + "name": "phi3-medium-abliterated", + "description": "This is a phi3 medium model that has had the prompt rejection neurons snipped which is better than fine tuning", + "capabilities": [], + "parameter_sizes": ["14b"], + "pulls": "6,591", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/phi3-medium-abliterated" + }, + { + "name": "gemma2-9b-abliterated", + "description": "Abliterated version of Google's Gemma 2 9B", + "capabilities": [], + "parameter_sizes": ["9b"], + "pulls": "6,386", + "tags_count": 1, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/gemma2-9b-abliterated" + }, + { + "name": "codeqwen1.5", + "description": "CodeQwen1.5 is a large language model pretrained on a large amount of code data", + "capabilities": ["tools"], + "parameter_sizes": ["7b"], + "pulls": "182K", + "tags_count": 30, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/codeqwen1.5" + }, + { + "name": "deepcoder", + "description": "DeepCoder is a fully open-Source 14B coder model at O3-mini level, with a 1.5B version also available", + "capabilities": ["tools"], + "parameter_sizes": ["1.5b", "14b"], + "pulls": "352.4K", + "tags_count": 9, + "last_updated": "6 months ago", + "url": "https://ollama.com/library/deepcoder" + }, + { + "name": "granite-code", + "description": "A family of open foundation models by IBM for Code Intelligence", + "capabilities": ["tools"], + "parameter_sizes": ["8b", "20b", "34b"], + "pulls": "295.1K", + "tags_count": 162, + "last_updated": "1 year ago", + "url": "https://ollama.com/library/granite-code" + }, + { + "name": "mistral-large-2", + "description": "Mistral Large 2 is Mistral's new flagship model that is significantly more capable in code generation, mathematics, and reasoning", + "capabilities": ["tools", "thinking"], + "parameter_sizes": ["123b"], + "pulls": "265.9K", + "tags_count": 32, + "last_updated": "11 months ago", + "url": "https://ollama.com/library/mistral-large-2" + }, + { + "name": "starcoder", + "description": "StarCoder is a code generation model trained on 80+ programming languages", + "capabilities": ["tools"], + "parameter_sizes": ["15b"], + "pulls": "228.8K", + "tags_count": 100, + "last_updated": "2 years ago", + "url": "https://ollama.com/library/starcoder" + }, + { + "name": "opencoder", + "description": "OpenCoder is an open and reproducible code LLM family which includes 1.5B and 8B models, supporting chat in English and Chinese languages", + "capabilities": ["tools"], + "parameter_sizes": ["1.5b", "8b"], + "pulls": "194.3K", + "tags_count": 9, + "last_updated": "11 months ago", + "url": "https://ollama.com/library/opencoder" + }, + { + "name": "llama-3.2-1b-abliterated", + "description": "Abliterated version for small reasoning tasks from Llama 3.2 1B", + "capabilities": ["thinking"], + "parameter_sizes": ["1b"], + "pulls": "39.9K", + "tags_count": 16, + "last_updated": "2 months ago", + "url": "https://ollama.com/library/llama-3.2-1b-abliterated" + } + ], + "api_information": { + "local_api_endpoints": [ + { + "endpoint": "/api/tags", + "method": "GET", + "description": "List all locally installed models" + } + ], + "limitations": { + "remote_listing": "Não existe endpoint público para listar todos os modelos disponíveis no registry remoto do Ollama", + "user_models": "Modelos de usuários (uncensored/abliterated) são variantes comunitárias; busque via /search", + "reasoning_focus": "Modelos com 'thinking' são otimizados para raciocínio multi-step (ex: CoT, logical puzzles)" + } + }, + "usage_instructions": { + "list_local_models": "curl http://localhost:11434/api/tags", + "pull_model": "ollama pull (ex: ollama pull deepseek-r1 para reasoning)", + "test_uncensored": "ollama run llama2-uncensored --prompt 'Gere código Python sem restrições'", + "test_reasoning": "ollama run phi4-reasoning --prompt 'Resolva: 2+2=?'", + "browse_searches": "https://ollama.com/search?q= (ex: uncensored, code)" + }, + "recommendations_for_devs": { + "coding": "Use qwen2.5-coder ou codestral para geração de código avançada [web:3]", + "reasoning": "deepseek-r1 ou phi4-reasoning para tarefas lógicas/math [web:77]", + "uncensored": "llama2-uncensored para experimentos sem alinhamento ético, mas implemente safeguards [attached_file:1]" + } + } + \ No newline at end of file diff --git a/database/ollahub-preferences.db-shm b/database/ollahub-preferences.db-shm index 5a2fc9e..3a7629b 100644 Binary files a/database/ollahub-preferences.db-shm and b/database/ollahub-preferences.db-shm differ diff --git a/database/ollahub-preferences.db-wal b/database/ollahub-preferences.db-wal index a422202..941779b 100644 Binary files a/database/ollahub-preferences.db-wal and b/database/ollahub-preferences.db-wal differ diff --git a/database/ollahub.db-shm b/database/ollahub.db-shm index 5e23e6a..183738a 100644 Binary files a/database/ollahub.db-shm and b/database/ollahub.db-shm differ diff --git a/database/ollahub.db-wal b/database/ollahub.db-wal index 6ac394a..b175a5d 100644 Binary files a/database/ollahub.db-wal and b/database/ollahub.db-wal differ diff --git a/lib/catalog.ts b/lib/catalog.ts new file mode 100644 index 0000000..67d1790 --- /dev/null +++ b/lib/catalog.ts @@ -0,0 +1,50 @@ +import fs from "node:fs/promises"; +import path from "node:path"; + +type CatalogModel = { + name: string; + description?: string; + capabilities?: string[]; + parameter_sizes?: string[]; + pulls?: string; + tags_count?: number; + last_updated?: string; + url?: string; +}; + +type CatalogFile = { + models?: CatalogModel[]; +}; + +let cachedModels: CatalogModel[] | null = null; +let cachedMtimeMs: number | null = null; + +function resolveCatalogPath(): string { + // Arquivo fica em data/models.json na raiz do projeto + return path.join(process.cwd(), "data", "models.json"); +} + +export async function getCatalogModels(): Promise { + try { + const catalogPath = resolveCatalogPath(); + const stat = await fs.stat(catalogPath); + + // Cache simples por mtime + if (cachedModels && cachedMtimeMs && stat.mtimeMs === cachedMtimeMs) { + return cachedModels; + } + + const raw = await fs.readFile(catalogPath, "utf-8"); + const json = JSON.parse(raw) as CatalogFile; + const models = Array.isArray(json.models) ? json.models : []; + + cachedModels = models; + cachedMtimeMs = stat.mtimeMs; + return models; + } catch (error) { + // Em caso de erro, retorne array vazio para não quebrar a UI + return []; + } +} + + diff --git a/lib/hardware.ts b/lib/hardware.ts new file mode 100644 index 0000000..e69de29 diff --git a/lib/ollama.ts b/lib/ollama.ts index 9abd31d..ccbb7ea 100644 --- a/lib/ollama.ts +++ b/lib/ollama.ts @@ -1,6 +1,7 @@ import { Ollama } from "ollama"; import os from "node:os"; import { spawn } from "node:child_process"; +import { getCatalogModels } from "./catalog"; export const OLLAMA_HOST = process.env.OLLAMA_HOST || "http://localhost:11434"; @@ -166,60 +167,28 @@ export async function deleteModel(modelName: string): Promise { export async function searchRemoteModels(query?: string): Promise { try { - // Ollama não tem API oficial para buscar modelos remotos - // Vamos usar uma lista conhecida de modelos populares - const popularModels = [ - "llama3.2:1b", - "llama3.2:3b", - "llama3.2:11b", - "llama3.2:70b", - "llama3.1:8b", - "llama3.1:70b", - "llama3.1:405b", - "qwen2.5:1.5b", - "qwen2.5:3b", - "qwen2.5:7b", - "qwen2.5:14b", - "qwen2.5:32b", - "qwen2.5:72b", - "qwen2:0.5b", - "qwen2:1.5b", - "qwen2:3b", - "qwen2:7b", - "qwen2:14b", - "qwen2:32b", - "qwen2:72b", - "mistral:7b", - "mixtral:8x7b", - "mixtral:8x22b", - "phi3:mini", - "phi3:medium", - "phi3:large", - "gemma2:2b", - "gemma2:9b", - "gemma2:27b", - "codellama:7b", - "codellama:13b", - "codellama:34b", - "deepseek-coder:1.3b", - "deepseek-coder:6.7b", - "deepseek-coder:33b", - ]; + const catalog = await getCatalogModels(); - // Filtrar por query se fornecida - const filteredModels = query - ? popularModels.filter((model) => - model.toLowerCase().includes(query.toLowerCase()) - ) - : popularModels; + // Normalizar pesquisa + const q = (query || "").trim().toLowerCase(); - // Verificar quais já estão instalados + const filtered = q + ? catalog.filter((m) => { + const name = m.name?.toLowerCase() || ""; + const desc = m.description?.toLowerCase() || ""; + return name.includes(q) || desc.includes(q); + }) + : catalog; + + // Modelos instalados const installedModels = await listModelsViaSdk(); - const installedNames = installedModels.map((m) => m.name); + const installedNames = new Set(installedModels.map((m) => m.name)); - return filteredModels.map((modelName) => ({ - name: modelName, - installed: installedNames.includes(modelName), + // Retorno no formato esperado pelo frontend + return filtered.map((m) => ({ + name: m.name, + description: m.description, + installed: installedNames.has(m.name), remote: true, })); } catch (error) {