Solshine/reflection-llama-3.1-8B-Solshine-trainround3-16bit

Uploaded model

• Developed by: Solshine (Caleb DeLeeuw)
• License: LLama 3.1 License
• Finetuned from model : Solshine/reflection-llama-3.1-8B-Solshine-trainround2-16bit

Inspired by and featuring the Reflection Tuning technique pioneered by Matt Shumer (possibly earlier innovated by the team at Anthropic.)

To the authors' knowledge, this is V3 of the first "reflection tuned" Llama 3.1 8B LLM

As per the inspiring model "mattshumer/Reflection-Llama-3.1-70B" (this mode was not used in the training process nor as a foundational model, but only served as inspiration) :

During sampling, the model will start by outputting reasoning inside <thinking> and </thinking> tags, and then once it is satisfied with its reasoning, it will output the final answer inside <output> and </output> tags. Each of these tags are special tokens, trained into the model.

This enables the model to separate its internal thoughts and reasoning from its final answer, improving the experience for the user.

Inside the <thinking> section, the model may output one or more <reflection> tags, which signals the model has caught an error in its reasoning and will attempt to correct it before providing a final answer.

System Prompt:
The system prompt used for training this model is:

"You are a world-class AI system, capable of complex reasoning and reflection. Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags. If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags."

We recommend using this exact system prompt to get the best results from Reflection Llama-3.1 70B. You may also want to experiment combining this system prompt with your own custom instructions to customize the behavior of the model.

Chat Format:
As mentioned above, the model uses the standard Llama 3.1 chat format. Here’s an example:

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

You are a world-class AI system, capable of complex reasoning and reflection. Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags. If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags.<|eot_id|><|start_header_id|>user<|end_header_id|>

what is 2+2?<|eot_id|><|start_header_id|>assistant<|end_header_id|>

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.