Browse Models
Launch a dedicated cloud GPU server running Laboratory OS to download and run MythoMax L2 using any compatible app or framework.
Must be used with a compatible app, notebook, or codebase. May run slowly, or not work at all, depending on local system resources, particularly GPU(s) and available VRAM.
Open WebUI is an open-source, self-hosted web interface with a polished, ChatGPT-like user experience for interacting with LLMs. Integrates seamlessly with local Ollama installation.
The most full-featured web interface for experimenting with open source Large Language Models. Featuring a wide range of configurable settings, inference engines, and plugins.
MythoMax L2 is a text generation model created by Gryphe through merging MythoLogic-L2 and Huginn using a tensor-based technique that applies individualized blending coefficients across 363 tensors. This approach integrates the strengths of both parent models to enhance coherency in dialogue and narrative generation, particularly for roleplaying and creative writing applications using Alpaca-style prompts.
MythoMax L2 is a generative AI language model specialized for roleplaying and creative writing applications. Developed as a synthesis of two foundational models, MythoLogic-L2 and Huginn, MythoMax L2 applies an advanced tensor-based merging process to integrate their respective capabilities. The model primarily aims to enhance coherency and creative expressiveness in dialogue and narrative generation.
At its core, MythoMax L2 is the result of merging the structures and learned parameters of MythoLogic-L2 and Huginn, two models known for their strengths in understanding and generation within conversational contexts. The integration leverages a "tensor type merge" technique. In this process, the creator applies unique, gradient-tuned ratios across a total of 363 tensors within the model. This approach assigns individualized blending coefficients to each tensor, resulting in nuanced behavioral adjustments and an overall increase in dialogic coherency.
Because each tensor in the network—spanning both the earliest (input) and latest (output) layers—can receive distinct proportions from its parental models, the merging remains highly granular. The technique thereby allows for targeted synthesis of desired traits: robust comprehension from MythoLogic-L2 and extended, immersive writing ability from Huginn. The source code and detailed configuration templates for MythoMax L2's production are available through the project's GitHub repository.
Unlike traditional models trained from scratch or through simple fine-tuning, MythoMax L2 inherits its foundational training from MythoLogic-L2 and Huginn. Both underlying models were developed using large-scale language pretraining, optimized for storytelling and dialog generation within various narrative settings. The distinguishing training process for MythoMax L2 lies in its block merge strategy, which fuses not only the learned weights but also the functional characteristics of different layers and modules.
The use of gradient-informed weights permits fine-grained adjustment of each tensor, facilitating a mesh of capabilities from the two source models. This sophisticated approach is intended to avoid the pitfalls of conventional model merging, such as abrupt behavioral shifts or loss of coherence during transition between tasks. Further technical details regarding the tensor merging process, including the mathematical formulations and practical implementation, can be reviewed in the official project documentation.
MythoMax L2 is principally oriented toward applications involving roleplay and creative storytelling. Its conversational abilities are designed to support complex character interactions, narrative generation, and immersive scenario construction. The model accepts prompts formatted in the Alpaca style, which encourages structured interactions between user and character. For example, an optimal prompt might specify a system or character card, followed by an explicit instruction for generating a reply in a simulated chat context, and conclude with a distinctly marked response section.
The model is effective for interactive writing environments and creative projects that demand both logical understanding and fluency in extended, stylistic prose. With its blend of input comprehension and generative versatility, MythoMax L2 is employed in text-based games, collaborative fiction platforms, and other applications necessitating dialogue continuity and narrative cohesion.
Recognizing the need for accessibility across diverse hardware environments, MythoMax L2 is available in several quantized formats. These include GGUF, GPTQ, and AWQ versions, which enable efficient deployment with minimal performance compromise. While the merge process itself is complex, the quantized variants facilitate practical use for both research and personal projects. The model is licensed under an "other" license category, details of which are outlined in the Hugging Face model card.
As a successor to previous projects in the MythoMax family, MythoMax L2 integrates improvements over earlier versions such as MythoMix. The expanded tensor intermingling at the front and end layers results in a more seamless mesoscale transition between the strengths of MythoLogic-L2 and Huginn. This leads to increased coherency in narrative output and heightened aptitude for immersive character-driven dialogue. While new large language models such as Llama 3 are expected to gradually supersede it, MythoMax L2 has served as a reference point within the genre of creative generative language systems.
The developers of MythoMax L2 note that rapid advances in foundational model architectures and training methodologies may eventually diminish its prominence. The release of newer models with greater parameter counts and more diverse training datasets, such as Llama 3, is anticipated to result in MythoMax L2's gradual phase-out. Despite this, the model's current architecture and accessible distribution formats allow it to retain utility in specific conversational and creative domains.