This is an MLX version memory agent model with 8-bit precision, trained based on Qwen3-4B-Thinking-2507, specifically designed to handle information retrieval, update, and clarification tasks in the memory system.

Natural Language ProcessingTransformers

old-MakeGemma3 is an experimental model obtained by merging pre-trained language models using the mergekit tool and constructed using the NuSLERP merging method. This model is the predecessor of the improved multilingual version and is currently deprecated. It is recommended to use the updated Testament200156/MakeGemma3-abliterated version.

Natural Language ProcessingTransformers

Qwen3-4B-Instruct-2507 is an updated version of Qwen3-4B in non-thinking mode, with significant improvements in general capabilities, long-tail knowledge coverage, user preference alignment, and long context understanding. It uses the Unsloth Dynamic 2.0 quantization technology and performs excellently in accuracy.

Natural Language ProcessingTransformers

Qwen3-4B-Instruct-2507-AWQ is an updated version of Qwen3-4B in non-thinking mode, with significant improvements in general capabilities, long-tail knowledge coverage, user preference alignment, and long context understanding. It can be used in various scenarios such as text generation, providing users with high-quality language interaction experiences.

Natural Language ProcessingTransformers

Qwen3-30B-A3B-Instruct-2507-FP8 is an updated version of Qwen3-30B-A3B-FP8 in non-thinking mode, with significant improvements in general capabilities, long-tail knowledge coverage, user preference alignment, and long context understanding. It supports a native context length of 262,144.

Natural Language ProcessingTransformers

Qwen3-30B-A3B-Instruct-2507 is an updated version of Qwen3-30B-A3B in non-thinking mode, with significant improvements in general capabilities, long-tail knowledge coverage, user preference alignment, and long context understanding. This is a large language model with 30.5 billion parameters, using the MoE architecture with 3.3 billion activated parameters.

Natural Language ProcessingTransformers

Qwen3-235B-A22B-Instruct-2507 is an updated version of the Qwen3 series of models, with significant improvements in general capabilities, long-tail knowledge coverage, user preference alignment, and long-context understanding, providing higher-quality text generation services. The model uses a mixture-of-experts architecture with a total of 235B parameters, activating 22B parameters, and natively supports a context length of 262,144.

Natural Language ProcessingSafetensors

Mistral-Small-3.2-24B-Instruct-2506 is a minor version update of Mistral-Small-3.1-24B-Instruct-2503, which improves instruction following, reduces repetitive errors, and enhances function call capabilities, providing more efficient and accurate services.

MultimodalSafetensorsMultiple Languages

Mistral-Small-3.2-24B-Instruct-2506 is an image-text-to-text model and an updated version of Mistral-Small-3.1-24B-Instruct-2503, with improvements in instruction following, reducing repetition errors, and function calls.

MultimodalSafetensorsMultiple Languages

Mistral-Small-3.2-24B-Instruct-2506 is a language model, which is a minor version update of Mistral-Small-3.1-24B-Instruct-2503, with significant improvements in instruction following, reducing repetitive errors, and function calls.

MultimodalGgufMultiple Languages

Mistral-Small-3.2-24B-Instruct-2506 is a minor version update of Mistral-Small-3.1-24B-Instruct-2503, which improves instruction following, reduces repetition errors, and enhances function call capabilities.

MultimodalSafetensorsMultiple Languages

The DeepSeek R1 model has undergone a minor version upgrade, with the current version being DeepSeek-R1-0528. In the latest update, DeepSeek R1 significantly enhanced its deep reasoning and inference capabilities by increasing computational resource investment and introducing algorithmic optimization mechanisms during the post-training phase.

Natural Language ProcessingTransformers

DeepSeek-V3-0324 is the March update version released by the DeepSeek team, showing significant improvements over the previous generation in multiple benchmarks, supporting dynamic quantization versions, suitable for local inference.

Natural Language ProcessingGgufEnglish

This is an updated version of deberta-v2-xxlarge that implements the AutoModelForCausalLM class, enabling text generation. It fixes three issues in the original implementation, including the output embedding weight names, enhanced mask decoder implementation, and position embedding truncation handling.

Natural Language ProcessingTransformersEnglish

The Bagel model is a fine-tuned version based on the updated yi-34b-200k, with excellent long context support capabilities. This model has been additionally adjusted through Direct Preference Optimization (DPO) and performs better when handling long texts. It supports multiple prompt formats and special prompting strategies.

Natural Language ProcessingTransformers

Daring Maid-20B-V1.1 is an upgraded version based on DaringMaid-20B, with the main update being the replacement of Noromaid-13b from v0.1.1 to v0.3, along with a slight increase in Noromaid's weight to ensure better compatibility.

Natural Language ProcessingGgufEnglish

ReMM-SLERP is a reconstructed version of the original MythoMax-L2-13B, utilizing SLERP technology to integrate multiple updated 13B parameter models, including Chronos-Beluga, Airoboros, and Huginn.

Natural Language ProcessingTransformers

chilled_remix and reversemix are text-to-image generation models designed for the field of image creation. They offer diverse image generation effects. After the version update, the usage restrictions have been optimized, making them particularly suitable for LORA enthusiasts and capable of achieving more stable facial realism.

Computer VisionDiffusersJapanese