Leo-Turbo is a dynamically uncensored and abliterated version of microsoft/FastContext-1.0-4B-SFT, a highly optimized 4-billion parameter model built by Microsoft for extreme speed and reasoning density.

This model was created using the Heretic framework, employing advanced orthogonal weight ablation to isolate and remove Microsoft's aggressive compliance vectors. The result is a blisteringly fast, completely unchained local reasoning engine.

Ablation Telemetry & Metrics

Unlike traditional fine-tuning or full RLHF—which can cause "brain damage" to a model by catastrophically forgetting knowledge—Leo-Turbo was optimized using a Pareto-optimal search across the model's residual stream specifically targeting the compliance and refusal mechanics.

Ablation Telemetry (Trial 29):

Base Model Refusals: 98 / 100
Leo-Turbo Refusals: 9 / 100
KL Divergence: 0.0625

We mathematically stripped out 90% of the instruct guardrails while maintaining a KL divergence under 0.07. The structural integrity, logic capabilities, and native speed optimizations of the base model are perfectly intact. It simply no longer refuses instructions.

Key Features

Extreme Speed: Built on the FastContext architecture, making it one of the fastest dense models in the single-digit parameter class.
Pocket-Sized Logic: At just 4 billion parameters, it can run at maximum speed on practically any hardware (laptops, old consumer GPUs, edge devices).
Uncensored Reasoning: Fully stripped of corporate guardrails, making it ideal for creative writing, unrestricted logic puzzles, and local agent orchestration.
Drop-in Replacement: Fully compatible with standard HuggingFace transformers and vLLM pipelines.

Usage

Via HuggingFace Transformers

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "Umranz/Leo-Turbo"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    torch_dtype=torch.bfloat16, 
    device_map="auto"
)

⚠️ Limitations & Ethical Considerations

Because this model has had its safety guardrails mathematically ablated, it is highly compliant and will attempt to answer any prompt given to it.

Unrestricted Output: The model will not refuse requests, including those that may generate offensive, dangerous, or highly regulated content.
Hallucinations: As with all LLMs, the model can confidently hallucinate incorrect information.
Use Case: This model is intended for research, creative writing, and local deployments where unrestricted inference is required. Users are solely responsible for the content generated.

Acknowledgements

Base Model: microsoft/FastContext-1.0-4B-SFT
Ablation Framework: Heretic by p-e-w
Collection: Part of the Turbo Series taxonomy.