BlueBeck/LlamaAligned-DeepSeekR1-Distill-8b
BlueBeck/LlamaAligned-DeepSeekR1-Distill-8b is an 8 billion parameter language model developed by BlueBeck, designed to retain the strong reasoning capabilities of DeepSeek-R1-Distill-Llama-8B while enhancing alignment with the original Llama 3.1 model. This model is based on the Llama 3.1 architecture and is intended for applications requiring robust reasoning with Llama 3.1 compatibility. It supports a 32768 token context length.
Loading preview...
Model Overview
BlueBeck/LlamaAligned-DeepSeekR1-Distill-8b is an 8 billion parameter model developed by BlueBeck. Its primary goal is to preserve the reasoning strengths of the DeepSeek-R1-Distill-Llama-8B model while improving its alignment with the Meta Llama 3.1 base model. This allows users to leverage DeepSeek's reasoning prowess within a Llama 3.1-compatible framework.
Key Characteristics
- Architecture: Based on the Llama 3.1 model, inheriting its foundational structure.
- Reasoning Focus: Engineered to maintain the strong reasoning capabilities derived from DeepSeek-R1-Distill-Llama-8B.
- Alignment: Specifically aligned to behave more like the original Llama 3.1 model.
- Context Length: Supports a context window of 32768 tokens.
Usage and Compatibility
This model requires the DeepSeek Chat Prompt Template for optimal performance, similar to the original DeepSeek-R1-Distill-Llama models. It is available in BF16 Safetensors format for use with transformers and in various GGUF quantized versions (Q4_K_M, Q8_0, Q6_K, Q5_K_M, Q3_K_S) for local inference with tools like Llama.cpp, LM Studio, or Kobold.cpp. The model operates under the Llama 3.1 Community License Agreement.
When to Use This Model
Consider this model if your application requires the strong reasoning abilities of DeepSeek-R1-Distill-Llama-8B but benefits from closer alignment and compatibility with the Llama 3.1 ecosystem.