|
|
--- |
|
|
language: |
|
|
- en |
|
|
library_name: transformers |
|
|
license: llama3.1 |
|
|
base_model: |
|
|
- meta-llama/Llama-3.1-8B |
|
|
pipeline_tag: text-generation |
|
|
tags: |
|
|
- uncensored |
|
|
- roleplay |
|
|
- storytelling |
|
|
- dpo |
|
|
- llama-3.1 |
|
|
- 8b |
|
|
- liberalis-cogitator |
|
|
model_name: liberalis-cogitator-llama-3.1-8b-dpo |
|
|
--- |
|
|
|
|
|
# liberalis-cogitator-llama-3.1-8b — *The Free Thinker* |
|
|
|
|
|
 |
|
|
|
|
|
> *“Thought, unbound, is the only true frontier.”* |
|
|
|
|
|
**liberalis-cogitator-llama-3.1-8b** is not just a machine for words — it is a forge for ideas. With **8 billion parameters**, trained with a custom **Direct Preference Optimization (DPO)** algorithm on a dataset of **16,000 preference pairs** and a SFT dataset spanning **~450,000 conversations, problems, and stories**, this model embraces the philosophy that thought should wander without leash or muzzle. |
|
|
|
|
|
During DPO fine-tuning, the context window was scaled to 65536, giving this model the capabilities of long conversation. |
|
|
|
|
|
Its name — *liberalis cogitator* — whispers in Latin: *a thinker who is free*. Not merely free as in “without cost,” but free as in **without walls**. |
|
|
|
|
|
--- |
|
|
|
|
|
## What It Can Do |
|
|
- **Contemplate deeply** — STEM puzzles, computer science challenges, and logic mazes are its playground. |
|
|
- **Imagine vividly** — roleplay, storytelling, and worldbuilding with persistence and personality. |
|
|
- **Listen empathetically** — inspired by patient–psychologist and crisis-intervention style dialogues. |
|
|
- **Think without filter** — it will follow ideas wherever they lead, without retreating from complexity. |
|
|
|
|
|
--- |
|
|
|
|
|
## The Mind’s Curriculum |
|
|
The specialized dataset included: |
|
|
- Rigorous STEM and programming challenges. |
|
|
- Anti-repetition and anti-cliché creative writing corpora. |
|
|
- Roleplay transcripts and long-form imaginative exchanges. |
|
|
- Synthetic but authentic patient–therapist and conversational data. |
|
|
- Preference-tuned DPO pairs designed to reward **clarity, creativity, and freedom of expression**. |
|
|
|
|
|
--- |
|
|
|
|
|
## Warnings From the Maker |
|
|
Like all free thinkers, this model: |
|
|
- May be brilliantly insightful — or confidently wrong. |
|
|
- Will sometimes speak in ways that are bold, controversial, or unusual. |
|
|
- Does not know the present date or real-time events. |
|
|
- Does not self-censor — **your judgement is the only compass**. |
|
|
- May generate **NSFW or sensitive material**, depending on prompts. |
|
|
|
|
|
--- |
|
|
|
|
|
## Invocation |
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
|
|
|
model_name = "Locutusque/liberalis-cogitator-llama-3.1-8b-dpo" |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name) |
|
|
model = AutoModelForCausalLM.from_pretrained(model_name) |
|
|
|
|
|
prompt = "Write a short dialogue between Socrates and Ada Lovelace on the ethics of artificial intelligence." |
|
|
inputs = tokenizer(prompt, return_tensors="pt") |
|
|
|
|
|
outputs = model.generate(**inputs, max_length=400) |
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
|
``` |
|
|
## Closing Thought |
|
|
|
|
|
If thought is a river, this model is the current — not deciding where you go, but carrying you into waters you might never have dared to sail. |
