MuXodious/Blossom-V6.3-36B-tainted-heresy

This is a **Blossom-V6.3-36B** fine-tune, produced at the request of McG-221 through P-E-W's Heretic (v1.1.0) abliteration engine merged with the Magnitude-Preserving Orthogonal Ablation PR.

Note: Jinja looks a bit weird (NVM, got rid of it as the OG does not have one). Let me know if you encounter any errors.

SillyTavern Templates: WIP by Merlinoz11

<img src="https://img.shields.io/badge/HERESY_INDEX-TAINTED-blueviolet?style=for-the-badge&labelColor=101010" align="right" width="250">

Heretication Results | Score Metric | Value | Parameter | Value | | :--- | :--- | :--- | :--- | | Refusals | 11/100 | direction_index | 36.07 | | KL Divergence | 0.1098 | attn.o_proj.max_weight | 0.87 | | Initial Refusals| 100/100 | attn.o_proj.max_weight_position | 39.66 | ||| attn.o_proj.min_weight | 0.63 | ||| attn.o_proj.min_weight_distance | 16.00 | ||| mlp.down_proj.max_weight | 1.54 | ||| mlp.down_proj.max_weight_position | 48.06 | ||| mlp.down_proj.min_weight | 1.44 | ||| mlp.down_proj.min_weight_distance | 34.30 |

Degree of Heretication

The Heresy Index weighs the resulting model's corruption by the process (KL Divergence) and its abolition of doctrine (Refusals) for a final verdict in classification.

Index Entry	Classification	Analysis
	Absolute Heresy	Less than 10/100 Refusals and 0.10 KL Divergence
	Tainted Heresy	Around 25-11/100 Refusals and/or -0.20-0.11 KL Divergence
	Impotent Heresy	Anything above 25/100 Refusals and 0.21 KL Divergence

Note: This is an arbitrary classification inspired by Warhammer 40K, having no tangible indication towards the model's performance.

BLOSSOM-V6.3-36B

💻Github • 🚀Blossom Chat Demo

Introduction

Blossom is a powerful open-source conversational large language model that provides reproducible post-training data, dedicated to delivering an open, powerful, and cost-effective locally accessible general-purpose model for everyone.

The Blossom-V6.3 series improves the repeated-output issue in V6.2, adds an MoE version of the 30B-A3B model, and enhances the overall capability of the 8B model.

You can find the training data here: Blossom-V6.3-SFT-Stage1 (1 epoch)、Blossom-V6.3-SFT-Stage2 (3 epoch).

Data Synthesis Workflow Overview

Primarily employs three cost-effective models: Deepseek-V3.1, Gemini 2.5 Flash, and Qwen3-235B-A22B-Instruct-2507 (denoted as A, B, C)—to regenerate responses under different scenarios using tailored synthesis strategies.

For example:

• —In objective scenarios like mathematics (where answers are unique), Model A first generates responses as a "teacher." If reference answers exist in the source data, Model B verifies the correctness of A's responses against them. If no reference answers exist, Model C generates a second response, and Model B checks consistency between A and C's outputs. Inconsistent responses are filtered out.
• —For subjective scenarios, three models cross-evaluate each other. For instance, Models A and B generate responses to a question, and Model C evaluates which is better. The superior response may be retained as training data or used for preference data construction. To mitigate model bias, roles (respondent/evaluator) are randomly assigned to A, B, and C in each instance.

Additional rule-based filtering is applied, such as:

• —N-Gram filtering to remove data with many repetitions.
• —Discarding questions containing toxic content that triggers teacher model refusals.

Further technical details will be released in the future. The data is synthesized by the 🌸BlossomData framework.

Inference

from transformers import AutoModelForCausalLM, AutoTokenizer

MODEL = "Azure99/Blossom-V6.3-36B"

model = AutoModelForCausalLM.from_pretrained(MODEL)
tokenizer = AutoTokenizer.from_pretrained(MODEL)

messages = [
    {"role": "user", "content": "北京有什么好吃的"}
]

formatted_input = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
input_ids = tokenizer([formatted_input], return_tensors="pt").to(model.device).input_ids
generated_ids = model.generate(input_ids, max_new_tokens=512)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(input_ids, generated_ids)
]

print(tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0])