---
license: gpl
license_name: sleepycat
license_link: LICENSE
---
Project-VEGA: Q-Mjollnir-3B-iQ5
> [!TIP]
> This Model in not censored and limit broken, Tested on a closed system please use and transfer with caution. This model is more function based We are working on the Llama version for Gaming and effective on device assistance in the real world
# IntelligentEstate/Mjollnir-3B-iQ5
A low rescource frontier GitC(Guy in the Chair) model with multi-in-turn-function calling.(This model is a GPT4.5 style model) From Project VEGA. This little model is meant to be the perfect low key tactical assistant. Batman Has Alfred, Iron man has Jarvis and you, you beautiful son of a bitch well you have Mjollnir. Even if it's just for a walkthrough of doom or to take down the cabal of local bullies This little model has your back. Should work on Android and Apple, PC or Mac.
This is an application of a small Effective non-reasoning tool use model with Multiple in turn Tool User for all devices. This little baby doesn't just have reasoning and tool use but multiple tool use and exceptional insight for its little size. It brings an awesome amount of functionality and usefulness to edge setups and is specifically built as a functional mobile assistant. For its size it leads the frontier.
This model beats OpenAI models in Function calling in an overall score on Berkeley Function-Calling Leaderboard (BFCL-v3). It was converted using a custom importance matrix specialized for tool use in the real world. Vector smoothing and S-AGI gives this model better overall responce and assistant capabilities than any other in it's class.
# Model Details
The Mjollnir model, fine-tuned from the Hammer 2.1 and Qwen 2.5 coder series, inherit Many advantages and are enhanced as follows:
- Multi-Step Function Calling: The assistant can perform multiple internal function calls to handle a single user request, actively planning and gathering information to fulfill complex tasks.
- Multi-Turn Function Calling: Enables continuous and context-aware interactions over multiple exchanges, with each turn potentially containing multiple steps, for a more natural conversation experience.
- Enhanced Irrelevant Information Inspection:: Better at identifying when provided functions are irrelevant to a user query, by providing a non-function call response.
Further enhanced through a task specific Importance matrix to ensure real world application and relevance where applicable.
- Custom Alignment in Real World Tasks: Intricate exploration human conscienceness, Gaming walk through effectiveness, Red Teaming situations and on device interaction and function calling and vastly more situation applications for Business and School.
- Simplified Artificial General Intelligence: Given the unique role of caretake for one who's world is evil and out to get him/her it is meant to be protective of the user and refers to him/her as the young Lord of War other aspects of human drivers added for a unique driving behavior(Not tested with many tools but has effective tool calling so use with care.)
## Evaluation
The evaluation results of Base Hammer 2.1 3B model on the Berkeley Function-Calling Leaderboard (BFCL-v3) are presented in the following table:
Built from Hammer's 2.1 Model. It showcases highly stable performance over all, suggesting the robustness of Hammer 2.1 series. In contrast, The others in the frontier show only specialized usefulness and overfitting for poor levels of effectiveness in the field. While this runs on 1 cheap samsung some of the models it competes with and beats would require 100 of the same system to run and there are no people who can compete for the tasks in the field.