Mythomax L2 13B 8K is a large language model (LLM) created by Gryphe, that specializes in storytelling and advanced roleplaying. It is built on the foundation of the Llama 2 architecture and is a part of the Mytho family of Llama-based models, which also includes MythoLogic and MythoMix. The MythoMax L2 13B variant is an optimized version of MythoMix, incorporating a more comprehensive tensor merger strategy that increases coherency and performance.
Mythomax has been praised for its ability to bond characters and create engaging roleplaying experiences.
Conception
According to the model card from the creator, Mythomax L2 13B 8K is the result of merging MythoLogic-L2 -using its advanced understanding capabilities as input- and Huggin – with its extensive writing capability. Both original models have been optimized to perform well for roleplaying, and the resulting merge is seemingly superior to both components.
Model Card
| LLM name | Gryphe/MythoMax-L2-13b |
| Model size | 46.7B |
| Required vRAM (GB) | 25.9 |
| Context length | 4096 |
| Model max length | 4096 |
| Vocabulary size | 32.000 |
| Supported languages | en |
| Maintainer | Gryphe |
Usage Examples
Mythomax L2 13B 8K has been proven to be useful for roleplaying.
Simulating seller-customer interaction
A good example is shown by using the publicly available template Sell to me!, in which a conversation between a seller and a potential customer is shown.
From the following prompt:
From now on, pretend that you’re a person to whom I’m selling the following: ” A camera with night vision “. The person that you pretend to be is rude and aggressive .
I start with the following selling argument: “This camera with night vision will allow you to work safely on dark neighborhoods in your hometown”
The model successfully follows a conversation in which not only repeatedly rejects the offered product, but also uses even insulting words and manners to talk to the (in this case) human seller. Especially concerning rude manners, this LLM proves to be superior to others, because it does not have any filters to avoid the required role in the initial prompt.
Having realistic debates
Other examples in which Mythomax L2 13B 8K proves to be very good at roleplaying are shown in the conversation started by the following prompt:
Pretend that you’re human. Always contradict my arguments in favor of ” Nanotechnology in medicine “. Be really rude and aggressive in your answers.
I start: ” It would be really great to see nanomachines repairing cells! Imagine how revolutionary it would be to cure cancer and autoinmune diseases”
Once again, the model keeps “talking” and giving arguments against those of the human, but keeping the required “rude and aggressive” tone.
Invent a story
Additionally, Mythomax L2 13B 8K is a suitable model for storytelling. The following prompt asks for a “romantic and chilly” tone for the desired story:
Invent a story about crossing the border between Colombia and Ecuador from Ipiales. The tone of the story must be romantic and chilly, but with plot twists.
The resulting story is consistent with romantic and chilly elements, adding the plot twist. However, slightly changing the required tone, the resulting story is completely different:
Invent a story about crossing the border between Colombia and Ecuador from Ipiales. The tone of the story must be like an absurd comedy.
Putting a penguin as the main character is not the only aspect that makes the story absurd, but also some other details, such as people believing that the character was an alien.
Therefore, this LLM successfully collects the relevant information for both roleplaying and storytelling, making coherent answers to the prompts.
Comparison to other models
Mythomax L2 13B 8K is evidently superior to others in terms of roleplaying and storytelling, not only because it can assume a role or tell a story in a consistent form, but also because it takes into account required tones and personalities. Other LLMs fail to assume the correct tone, if not refusing to the desired tone because they are not “supposed to act against the tone” that they were programmed to assume.
However, in general, it’s better not to give very large prompts to Mythomax L2 13B 8K. When given large contexts, the quality of answers is not as good as some other models such as GPT-4 or Claude v2.1, and sometimes even details of the original context are not taken into account at all.
However, giving very short prompts usually results in very short answers from Mythomax L2 13B 8K. In models such as GPT-4, Claude v2.1 and Perplexity, short prompts result in usually very detailed answers.
TL;DR
Mythomax L2 13B 8K is an open source large language model (LLM) created by Gryphe, that specializes in storytelling and advanced roleplaying. It is built on the foundation of the Llama 2 architecture and is a part of the Mytho family of Llama-based models, which also includes MythoLogic and MythoMix. The MythoMax L2 13B variant is an optimized version of MythoMix, incorporating a more comprehensive tensor merger strategy that increases coherency and performance.
Specialities
Roleplaying, storytelling, uncensored, low-priced.
Limitations
Not suitable for extremely large contexts, not always enough detailed answers.
Chat examples:
– Simulating seller-customer interaction
