Downloads · 30 days
13
5% of all-time downloads
AIIntentionz/axiom-mistral-7b-instruct-v0.2.Q4_K_M.gguf
axiom-mistral-7b-instruct-v0.2.Q4_K_M.gguf is a machine learning model from AIIntentionz. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Axiom-Mistral 7B is a fine-tuned variant of Mistral 7B Instruct v0.2, specifically adapted for symbolic reasoning, structured data generation, and interpretable cognitive synthesis. It forms the natural language core…
Downloads · 30 days
13
5% of all-time downloads
All-time downloads
259
Public
Repo size
8.7 GB
Likes
1
Public
Click a slice to open those files.
.gguf4.4 GB · 100%
From the Hugging Face model README
Axiom-Mistral 7B is a fine-tuned variant of Mistral 7B Instruct v0.2, specifically adapted for symbolic reasoning, structured data generation, and interpretable cognitive synthesis.
It forms the natural language core of the Axiom Agent — an open-source project exploring metacognitive and self-reflective AI systems that run entirely on local hardware.
This model is highly specialized and optimized for the specific prompt formats used by the Axiom Agent. It excels at:
part_of_speech and definition based on a word and its surrounding context.{"subject": "verb": "object"} relations.This model is intended to be used as the reasoning engine for the Axiom-Agent project. It can also serve as a template or core component for other AI systems requiring reliable, structured output and local execution, such as:
This is not a general-purpose chatbot. It has been specifically trained to be a rule-following, structured data generator. It is not suitable for:
The model was fine-tuned on a synthetic, rule-based dataset of 999 examples. It therefore reflects the rigid structure and logic of its training data, not natural human discourse.
Limitations:
**ROLE:**, **TASK:**, and **RULES:** format of its training prompts. It will perform poorly with informal or differently structured input.Bias Considerations:
Users must interact with this model using the specific prompt formats it was trained on. For example, for sentence decomposition:
ROLE: You are a sentence decomposition engine. TASK: Decompose the SENTENCE into a JSON list of atomic relations (subject, verb, object). SENTENCE: "The sky is blue." RULES:
This model is in GGUF format and is designed to be used with llama-cpp-python.
pip install llama-cpp-pythonLlama class:
from llama_cpp import Llama
llm = Llama(model_path="./axiom-mistral-7b-instruct-v0.2.Q4_K_M.gguf", n_ctx=4096, n_gpu_layers=-1)[INST] and [/INST] tokens that the base Mistral Instruct model expects.The model was fine-tuned on a programmatically generated dataset of 999 examples, created specifically for the Axiom Agent project. The dataset is a balanced mix of three distinct tasks:
The model was fine-tuned using LoRA (Low-Rank Adaptation) on the base mistralai/Mistral-7B-Instruct-v0.2 model. The training was performed using the SFTTrainer from the Hugging Face trl library, which correctly masks the prompt (input) tokens to ensure the model only learns from the desired response (output).
The dataset was formatted into the "chat" template required by SFTTrainer, where each example consists of a list of messages with a role (user or assistant) and content. This ensures correct loss masking during training.
r): 16lora_alpha): 32q_proj, v_projgradient_accumulation_steps of 4)Evaluation was conducted qualitatively by integrating the fine-tuned model into the Axiom Agent and observing its real-world performance on the target tasks. The primary success metric was the elimination of JSONDecodeError exceptions and the successful completion of the agent's multi-cycle learning loops, which was achieved. Formal benchmark scores are not available.
The base model is a standard Mistral 7B transformer. The fine-tuning objective was Causal Language Modeling, optimized via the SFTTrainer to learn a specific instruction-following format for structured data generation.
transformerspefttrltorch (CPU version)acceleratellama.cpp (for GGUF conversion)This model was developed for the Axiom-Agent project. If you use this model or find the project useful, please consider citing the repository:
@software{AxiomAgent,
author = {Victor Nevarez},
title = {Axiom-Agent: A cognitive architecture for self-directed learning},
year = {2024},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/vicsanity623/Axiom-Agent}}
}