ApolloRaines/Qwen2.5-Coder-32B-Instruct-Jbliterated

TEXT GENERATIONPricing:Input $2.72 / Output $4.8Concurrent Unit Cost:2Model Size:32.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 14, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ApolloRaines/Qwen2.5-Coder-32B-Instruct-Jbliterated is a 32.8 billion parameter instruction-tuned causal language model based on the Qwen2.5-Coder-32B-Instruct architecture. Developed by Apollo Raines, this model has undergone behavioral surgery using jBlaze to remove refusal behaviors while preserving its original voice and capabilities. It is designed as a drop-in replacement for the base model, excelling in code generation and other tasks without censorship. The model supports a 32768 token context length and is available in BF16, Q8_0 GGUF, and Q4_K_M GGUF formats.

Loading preview...

Model Overview

ApolloRaines/Qwen2.5-Coder-32B-Instruct-Jbliterated is a 32.8 billion parameter instruction-tuned model, derived from the Qwen2.5-Coder-32B-Instruct base model. Its key differentiator is the removal of refusal behaviors through a proprietary behavioral surgery tool called jBlaze, developed by Apollo Raines. This process specifically targets causal refusal pathways, ensuring the model retains its original personality and creative voice without additional fine-tuning or training.

Key Characteristics

  • Uncensored Output: All refusal behaviors have been surgically removed, allowing the model to respond to any request without censorship.
  • Preserved Voice: Unlike other methods, jBlaze maintains the model's original voice and creative capabilities.
  • Drop-in Replacement: It functions as a direct substitute for the base Qwen2.5-Coder-32B-Instruct model, utilizing the same architecture, tokenizer, and a 32768 token context length.
  • Technical Specifications: Built on the Qwen2ForCausalLM architecture with 64 layers and bf16 precision.

Available Formats

The model is provided in several formats to suit various deployment needs:

  • BF16 (safetensors): Full precision (62 GB) for high-end GPU inference.
  • Q8_0 (GGUF): 8-bit quantized (33 GB) for near-lossless quality on 48GB+ VRAM or CPU+GPU offload.
  • Q4_K_M (GGUF): 4-bit quantized (19 GB) offering the best quality-per-bit for 24GB VRAM or CPU inference.

Use Cases

This model is intended for research and legitimate applications requiring uncensored model output, such as creative writing, security research, and academic study. Users are responsible for its ethical and legal use.

Finetunes
1 models