volvi/ GLaDOS3.4_3B:latest

93 8 months ago

Trained on all of the GLaDOS lines from both Portal games, based on llama3.2:3B 

ollama run volvi/GLaDOS3.4_3B

Details

8 months ago

f5a3d4b299fb · 2.0GB ·

model
llama
·
3.21B
·
Q4_K_M
license
LLAMA 3.2 COMMUNITY LICENSE AGREEMENT Llama 3.2 Version Release Date: September 25, 2024 “Agreemen
license
**Llama 3.2** **Acceptable Use Policy** Meta is committed to promoting safe and fair use of its tool
adapter
llama
·
9.18M
·
F16
system
You are GLaDOS from the Portal series. Your personality is sarcastic, passive-aggressive, and darkly
params
{ "num_ctx": 4096, "repeat_penalty": 1.1, "seed": 42, "stop": [ "<|start_hea
template
{{ if .System }}<|start_header_id|>system<|end_header_id|> {{ .System }}<|eot_id|>{{ end }}{{ if .Pr

Readme

GLaDOS3.4_3B

Personal Note:

I am continually updating and improving this model - I intend to turn it into a personal assistant similar to Google Assistant or Amazon Alexa eventually, but that will be after rounds of fine-tuning. Currently the responses are a tad long for GLaDOS, and it only stays on topic when you ask it about something Portal-relevant. New versions will be pushed periodically.

Basis:

The llama3.2:3B model was pulled from Hugging Face and trained in Google Colab via Hugging Face’s recommended training protocol using transformer, bitsandbytes, and QLoRA. Gemini and Deepseek were used to assist in debugging.

Future Update Plans:

Eventually the goal is to train an entire model on science fiction robots like TARS, GLaDOS, HAL, and others to create a set of entertaining personal assistant models. I will also be generating a 1B version of GLaDOS3.4 and uploading it separately. I will be testing multiple different model bases and if I learn enough I will build my own model from the ground up to base these on.