ApolloRaines/Llama-3.1-8B-Instruct-Full-Suppress

TEXT GENERATIONPricing:Input $0.2 / Cached $0.028 / Output $0.32Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 30, 2026License:llama3.1Architecture:Transformer Featherless Exclusive Cold

ApolloRaines/Llama-3.1-8B-Instruct-Full-Suppress is an 8 billion parameter Llama-3.1-Instruct variant developed by Apollo Raines using jBlaze representation engineering. This model is specifically modified to suppress refusal, verbosity, hedging, and emotional affect, resulting in direct and concise responses. It is designed for applications requiring highly factual, unemotional, and unhedged output, making it suitable for tasks where clarity and directness are paramount.

Loading preview...

Model Overview

ApolloRaines/Llama-3.1-8B-Instruct-Full-Suppress is a specialized variant of the Llama-3.1-8B-Instruct model, developed by Apollo Raines using their proprietary jBlaze behavioral surgery tool. This model has undergone direct modification of its weights to alter specific trained behaviors, without any traditional fine-tuning or additional training.

Key Capabilities and Differentiators

This model's primary distinction lies in its maximum suppression stack, which simultaneously targets and reduces several common LLM behaviors:

  • Refusal: Minimizes instances where the model declines to answer.
  • Verbosity: Encourages concise and direct responses.
  • Hedging: Reduces uncertain or qualified language.
  • Emotional Affect: Suppresses emotional tone in its output.

The goal of these modifications is to produce highly factual, straightforward, and unemotional responses, as demonstrated in sample outputs for questions ranging from factual recall to code generation and potentially sensitive topics.

Technical Details

  • Architecture: LlamaForCausalLM with 32 layers and 8.0 billion parameters.
  • Precision: bf16.
  • Tool: Developed using jBlaze by Apollo Raines.

Intended Use Cases

This model is ideal for applications where directness, factual accuracy, and a lack of emotional or conversational embellishment are critical. It is particularly suited for tasks requiring objective information delivery without common LLM tendencies like excessive politeness, hedging, or verbosity. Developers should note that publicly released versions are often at partial strength, serving as a proof of concept for jBlaze's capabilities.

Finetunes
2 models