Maximiliano-Flores-Dev/grok_demon_7b

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Maximiliano-Flores-Dev/grok_demon_7b is a 7.6 billion parameter Qwen2-based causal language model, developed by Maximiliano-Flores-Dev. This model, trained in bfloat16 precision, serves as the required base for the more advanced grok_demon_7b_v2.1 adapter. While deprecated as a standalone model, it is crucial for loading the v2.1 bilingual (ES/EN) fine-tune, which offers superior performance and optimized linguistic handling.

Loading preview...

Model Overview: grok_demon_7b (v1 Base)

Maximiliano-Flores-Dev/grok_demon_7b is a 7.6 billion parameter Qwen2-based causal language model, developed by Maximiliano-Flores-Dev. This repository contains the merged weights of the v1 release, saved in bfloat16 precision, and is derived from unsloth/qwen2.5-7b-instruct-unsloth-bnb-4bit. It is primarily an English-Spanish bilingual model, though this v1 is deprecated as a standalone final model.

Key Characteristics

  • Base Model for v2.1: This model is a mandatory base for loading the Maximiliano-Flores-Dev/grok_demon_7b_v2.1 adapter, which provides significantly enhanced performance and bilingual capabilities.
  • Architecture: Utilizes the Qwen2 (Qwen2ForCausalLM) architecture.
  • Precision: Weights are stored in bfloat16 after merging a LoRA v1 adapter.
  • Training: Optimized using Unsloth and Hugging Face's TRL library for faster training.
  • Chat Template: Includes a chat_template.jinja for the "Grok Demon" persona, emphasizing the use of apply_chat_template() for optimal quality.

Important Note on Usage

This grok_demon_7b repository is deprecated as a final model but remains essential as the foundational weights for the grok_demon_7b_v2.1 adapter. Users are strongly encouraged to use the v2.1 adapter on top of this base for new deployments to leverage its superior bilingual performance and optimized dataset sampling. The model is released under the Apache-2.0 License.

Finetunes
3 models