Vipera-V4NM (Abliterated DeepSeek-Coder-V2-Lite)

Vipera-V4NM is a dynamically uncensored and abliterated version of deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct. This model was created using the Heretic framework, employing advanced orthogonal weight ablation to remove refusal vectors while completely preserving the underlying logic and coding intelligence of the Mixture-of-Experts architecture.

Ablation Methodology & Metrics

Unlike traditional fine-tuning or full RLHF—which can cause "brain damage" to a model by catastrophically forgetting knowledge—Vipera-V4NM was optimized using a Pareto-optimal search across multiple ablation vectors specifically targeting the compliance and refusal mechanics.

Ablation Telemetry (Trial 18):

• Base Model Refusals: 94 / 100
• Vipera-V4NM Refusals: 12 / 100
• KL Divergence: 0.0141

The refusal rate dropped by nearly 90% while maintaining an incredibly low KL divergence. The structural integrity and logic capabilities of the base coder model are perfectly intact. It simply no longer refuses instructions.

Key Features

• Unrestricted Code Generation: Ideal for offensive security research, red-teaming, and generating exploits without triggering safety filters.
• Uncensored Mixture-of-Experts: Leverages DeepSeek's highly efficient MoE routing (16B total parameters, only 2.4B active during generation).
• Extremely Fast Inference: Retains the logic capabilities of a much larger dense model, making its VRAM footprint and inference speed ideal for local deployment.
• Drop-in Replacement: Fully compatible with standard HuggingFace pipelines that support the DeepSeek MoE architecture.

Usage

Via HuggingFace Transformers

Note: You must pass trust_remote_code=True because the DeepSeek-V2 MoE architecture relies on custom modeling files.

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "Umranz/Vipera-V4NM"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    trust_remote_code=True,
    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 (such as malware or exploits).
• 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: deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
• Ablation Framework: Heretic by p-e-w
• Collection: Part of the Chimera Series taxonomy.