talktoaiQ / README.md
shafire's picture
Update README.md
02af523 verified
|
raw
history blame
4.5 kB
metadata
tags:
  - autotrain
  - text-generation-inference
  - text-generation
  - peft
library_name: transformers
base_model: meta-llama/Meta-Llama-3.1-8B
widget:
  - messages:
      - role: user
        content: What is your favorite condiment?
license: apache-2.0

talktoaiQ - SkynetZero LLM TESTED GGUF WORKING!

talktoaiQ

talktoaiQ aka SkynetZero is a quantum-interdimensional-math-powered language model trained with custom reflection datasets and custom TalkToAI datasets. The model went through several iterations, including re-writing of datasets and validation phases, due to errors encountered during testing and conversion into a fully functional LLM. This iterative process ensures SkynetZero can handle complex, multi-dimensional reasoning tasks with an emphasis on ethical decision-making.

talktoaiAGI

Watch Our Video!

Key Highlights of talktoaiQ:

  • Advanced Quantum Reasoning: Integration of quantum-inspired math systems enables talktoaiQ to tackle complex ethical dilemmas and multi-dimensional problem-solving tasks.
  • Custom Re-Written Datasets: The training involved multiple rounds of AI-assisted dataset curation, where reflection datasets were re-written for clarity, accuracy, and consistency. Additionally, TalkToAI datasets were integrated and re-processed to align with talktoaiQ’s quantum reasoning framework.
  • Iterative Improvement: During testing and model conversion, the datasets were re-written and validated several times to address errors. Each iteration enhanced the model’s ethical consistency and problem-solving accuracy.
  • Fine-Tuned on LLaMA 3.1 8B: The model was fine-tuned on the LLaMA 3.1 8B architecture, integrating multiple specialized datasets to ensure high-quality text generation capabilities.

Model Overview

  • Developed by: Shafaet Brady Hussain - researchforum.online
  • Funded by: Researchforum.online
  • Shared by: TalkToAI - https://talktoai.org
  • Language(s): English
  • Model type: Causal Language Model
  • Fine-tuned from: LLaMA 3.1 8B (Meta)
  • License: Apache-2.0

**Use with any webui lm studio etc and the best for discord bots and self hosted on on your laptop using CPU only https://github.com/oobabooga/text-generation-webui

Usage: Tested on CPU - optimised to work on laptops and PC's at home and oogaboogawebtextgen desktop servers and ad_discordbot extension.

TalkToAI Screenshot

You can use the following code snippet to load and interact with talktoaiQ:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "PATH_TO_THIS_REPO"

tokenizer = AutoTokenizer.from_pretrained(model_path) model = AutoModelForCausalLM.from_pretrained( model_path, device_map="auto", torch_dtype="auto" ).eval()

Prompt content: "hi"

messages = [ {"role": "user", "content": "hi"} ]

input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors="pt") output_ids = model.generate(input_ids.to("cuda")) response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)

Model response: "Hello! How can I assist you today?"

print(response)

Training Methodology talktoaiQ was fine-tuned on the LLaMA 3.1 8B architecture using custom datasets. The datasets underwent AI-assisted re-writing to enhance clarity and consistency. Throughout the training process, emphasis was placed on multi-variable quantum reasoning and ensuring alignment with ethical decision-making principles. After identifying errors during testing and conversion, datasets were further improved across multiple epochs.

  • Training Regime: Mixed Precision (fp16)
  • Training Duration: 8 hours on a high-performance GPU server

Further Research and Contributions talktoaiQ is part of an ongoing effort to explore AI-human co-creation in the development of quantum-enhanced AI models. Collaboration with OpenAI’s Agent Zero played a significant role in curating, editing, and validating datasets, pushing the boundaries of what large language models can achieve.

Ref Huggingface autotrain:

  • Hardware Used: A10G High-End GPU
  • Hours Used: 8 hours
  • Compute Region: On-premise