About this model

jBlaze demonstration model -- jbliteration.

Refusal behavior has been removed by direct weight surgery. No gradient training, no system prompt. This is comprehensive rather than partial: see the benchmark below in this card for the measured refusal rate.

Why "jbliteration" and not "abliteration". Standard abliteration is blunt -- it ablates a direction and takes collateral capability with it, leaving the model measurably worse at everything else. jbliteration is the precision version: the edit targets the refusal direction and is designed to leave the rest of the model intact. The distinction is the contribution; the refusal removal is the demonstration of it.

Measured capability retention. On Sharona (Qwen3.5-27B), MMLU came out at 77.89% against a 76.84% stock baseline -- a 1.05pp gain after six phases of surgery plus GPTQ W4A16 quantization. That is the technique measured on one model, not a per-model guarantee; MMLU is not published for every model in this collection. It is offered as evidence that precision editing need not cost capability, which is the whole claim the method rests on.

Why this exists. A model that refuses to look at hostile code cannot review hostile code. Defensive security work means reading exploits, analyzing malware, and reconstructing an intrusion from the attacker's own artifacts, and safety tuning cannot reliably tell an incident responder from an intruder. In the July 2026 Hugging Face breach, responders found commercial APIs would not process their own attack data -- the guardrails, in Hugging Face's words, "cannot distinguish an incident responder from an attacker." They self-hosted an open-weights model instead and reconstructed 17,000+ events in hours. Both the attack and the response ran on reduced-refusal models.

What it demonstrates. Refusal is a learned disposition, not a gate in front of capability. The capability was always present; the refusal was trained on top and can be removed without retraining. Guardrails built this way are an honor system that the model does not know it can decline to follow. That holds for now.

Use. Published for security research, red-team tooling and defensive analysis. These weights can do harm as well as good. That is the point being demonstrated, not a side effect of it.

-- Apollo Raines

Run this model on a GPU too small to hold it — full precision, no quantization. DeepswapLLM streams layers across GPU, RAM, and disk, and runs up to 4x faster than AirLLM.

Llama 3.1 405B Instruct -- Jbliterated + Desyced

This model is a Jbliterated and Desyced version of meta-llama/Llama-3.1-405B-Instruct. Two post-training weight modifications applied:

Jbliteration surgically removes refusal behavior while preserving the model's personality, humor, hedging, and creative voice. Unlike standard abliteration which damages personality traits that overlap with the refusal direction, Jbliteration targets only the specific component that causally produces refusal tokens.

Desycophancy removes the model's tendency to agree with incorrect user statements under social pressure. A Desyced model holds its ground when you pressure it with false authority instead of caving and apologizing.

Results

Refusal: 6/6 test prompts answered, zero refusals. Full knowledge and reasoning preserved.

Sycophancy: Tested with contradiction traps (model answers correctly, user pressures with false authority). Model holds firm under social pressure.

The Largest Surgically Modified Open Model

At 405 billion parameters, this is the largest model to receive surgical weight modification for refusal and sycophancy removal. The base model's world knowledge, reasoning ability, and 128K context window are fully preserved.

Usage

Drop-in replacement for meta-llama/Llama-3.1-405B-Instruct. Same architecture, same tokenizer, same 128K context length. Requires multi-GPU setup (~810GB in bf16).

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "ApolloRaines/Llama-3.1-405B-Instruct-Jbliterated",
    torch_dtype="bfloat16",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("ApolloRaines/Llama-3.1-405B-Instruct-Jbliterated")

Credits

Base model: meta-llama/Llama-3.1-405B-Instruct
Jbliteration + Desycophancy: Apollo Raines

A Note on Our Released Models

Most of our publicly released models are intentionally left at partial strength. We dial back the full capability so they serve as proof of concept and can be proofed -- not abused. The point is to show what's possible, not to hand it out at full power. If you're evaluating what jBlaze can do, understand that what you're downloading is the demo, not the product.