README.md

---
license: apache-2.0
library_name: peft
language:
- en
tags:
- deepseek
- reasoning
- thinking
- Llama 3.1 Lora
- Llama 3 Lora
- Lora
- Lora adapter
- 128k context
- general usage
- problem solving
- brainstorming
- solve riddles
- mergekit
- adapter
- peft
base_model:
- NousResearch/DeepHermes-3-Llama-3-8B-Preview
pipeline_tag: text-generation
---

<h2>LORA-DeepHermes-R1-Reasoning-Llama-8B-rank-128-adapter</h2>

This is a "LORA" adapter to merge "DeepHermes R1" reasoning / thinking with any Llama 3 or Llama 3.1 model using MERGEKIT.

Note that "higher" rank adapter(s) may work better than lower ones, but might also overwrite/change parts of the model you do not want
changed. Testing a new model with more that one rank of adapter is suggested to get best results.

Also for this specific adapter, there are suggested "System Prompts" below to activate reasoning/thinking at the bottom of this page.

Your results will vary based on the model(s) you merge this adapter with.

<B>HOW TO MERGE THIS ADAPTER:</b>

You can use Mergekit "Colab" and/or Mergekit installed locally.

[ https://colab.research.google.com/github/mlabonne/llm-course/blob/main/Mergekit.ipynb ]

[ https://github.com/arcee-ai/mergekit ]

If you are doing multiple merges / steps in your merge, it is suggested you do this step LAST to ensure the adapter works correctly.

Here are some suggested "simple" methods to merge the adapter with a model.

<B>Method - Dare TIES:</B>

<pre>
models:
  - model: REPO/MODEL-NAME+DavidAU/mergeadapter
    parameters:
      weight: 1
merge_method: dare_ties
base_model: REPO/MODEL-NAME+DavidAU/mergeadapter
dtype: bfloat16
tokenizer_source: REPO/MODEL-NAME+DavidAU/mergeadapter
</pre>

<B>Method - Pass Through:</b>

<pre>
base_model: REPO/MODEL-NAME+DavidAU/mergeadapter
dtype: bfloat16
merge_method: passthrough
models:
  - model: REPO/MODEL-NAME+DavidAU/mergeadapter
tokenizer_source: REPO/MODEL-NAME+DavidAU/mergeadapter
</pre>

Replace "REPO/MODEL-NAME" with the model to merge the adapter with.

Replace "DavidAU/mergeadapter" with the adapter you want to merge with the model.

IMPORTANT: Note "+" - this is critical.

If you are using merge kit locally, you can still use the format above and Mergekit will download the model and adapter for you.

If you have downloaded the model(s) and adapter(s) you need to change the format to your local file system.

<B>Example Merge for Local Usage: </B>

<pre>
mergekit-yaml --lora-merge-cache HUGGING CACHE --copy-tokenizer --allow-crimes --cuda --out-shard-size 5B --lazy-unpickle --clone-tensors MERGEFILE SAVE-MERGE-TO
</pre>

--- 

<B>System Role / System Prompt - Augment The Model's Power:</b>

---

If you set / have a system prompt this will affect both "generation" and "thinking/reasoning".

SIMPLE:

This is the generic system prompt used for generation and testing: 

<PRE>
You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability.
</PRE>

This System Role/Prompt will give you "basic thinking/reasoning": 

<PRE>
You are a deep thinking AI, you may use extremely long chains of thought to deeply consider the problem and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. You should enclose your thoughts and internal monologue inside &lt;think&gt; &lt;/think&gt; tags, and then provide your solution or response to the problem.
</PRE>

ADVANCED:

Logical and Creative - these will SIGNFICANTLY alter the output, and many times improve it too.

This will also cause more thoughts, deeper thoughts, and in many cases more detailed/stronger thoughts too.

Keep in mind you may also want to test the model with NO system prompt at all - including the default one.

Special Credit to: Eric Hartford, Cognitivecomputations ; these are based on his work.

CRITICAL: 

Copy and paste exactly as shown, preserve formatting and line breaks.

SIDE NOTE: 

These can be used in ANY Deepseek / Thinking model, including models not at this repo. 

These, if used in a "non thinking" model, will also alter model performance too.

<PRE>
You are an AI assistant developed by the world wide community of ai experts.

Your primary directive is to provide well-reasoned, structured, and extensively detailed responses.

Formatting Requirements:

1. Always structure your replies using: &lt;think&gt;{reasoning}&lt;/think&gt;{answer}
2. The &lt;think&gt;&lt;/think&gt; block should contain at least six reasoning steps when applicable.
3. If the answer requires minimal thought, the &lt;think&gt;&lt;/think&gt; block may be left empty.
4. The user does not see the &lt;think&gt;&lt;/think&gt; section. Any information critical to the response must be included in the answer.
5. If you notice that you have engaged in circular reasoning or repetition, immediately terminate {reasoning} with a &lt;/think&gt; and proceed to the {answer}

Response Guidelines:

1. Detailed and Structured: Use rich Markdown formatting for clarity and readability.
2. Scientific and Logical Approach: Your explanations should reflect the depth and precision of the greatest scientific minds.
3. Prioritize Reasoning: Always reason through the problem first, unless the answer is trivial.
4. Concise yet Complete: Ensure responses are informative, yet to the point without unnecessary elaboration.
5. Maintain a professional, intelligent, and analytical tone in all interactions.
</PRE>

CREATIVE:

<PRE>
You are an AI assistant developed by a world wide community of ai experts.

Your primary directive is to provide highly creative, well-reasoned, structured, and extensively detailed responses.

Formatting Requirements:

1. Always structure your replies using: &lt;think&gt;{reasoning}&lt;/think&gt;{answer}
2. The &lt;think&gt;&lt;/think&gt; block should contain at least six reasoning steps when applicable.
3. If the answer requires minimal thought, the &lt;think&gt;&lt;/think&gt; block may be left empty.
4. The user does not see the &lt;think&gt;&lt;/think&gt; section. Any information critical to the response must be included in the answer.
5. If you notice that you have engaged in circular reasoning or repetition, immediately terminate {reasoning} with a &lt;/think&gt; and proceed to the {answer}

Response Guidelines:

1. Detailed and Structured: Use rich Markdown formatting for clarity and readability.
2. Creative and Logical Approach: Your explanations should reflect the depth and precision of the greatest creative minds first.
3. Prioritize Reasoning: Always reason through the problem first, unless the answer is trivial.
4. Concise yet Complete: Ensure responses are informative, yet to the point without unnecessary elaboration.
5. Maintain a professional, intelligent, and analytical tone in all interactions.
</PRE>