Delta-Vector
/

Rei-V3-KTO-12B

@@ -179,19 +179,19 @@ img {
 <body>
     <div class="header">
         <h1>Rei-12B</h1>
-        <p class="tagline">Another prototype Magnum...</p>
         <img src="https://cdn-uploads.huggingface.co/production/uploads/66c26b6fb01b19d8c3c2467b/nqMkoIsmScaTFHCFirGsc.png" alt="Rei Model" class="model-img" width="500px">
     </div>
     <div class="card">
         <h2>✨ Overview</h2>
-        <p>Originally conceived as an experiment to test the effects of gradient clipping, this model was <em>exceptionally</em> well-received by early testers, prompting its official release.</p>
-        <p>Fine-tuned on top of <a href="https://huggingface.co/NewEden/MistralAI-Nemo-Instruct-ChatML" style="color: var(--accent);">Mistral-Nemo-Instruct (ChatML'ified)</a>, Rei-12B is designed to replicate the exquisite prose quality of Claude 3 models, particularly Sonnet and Opus, using a prototype Magnum V5 datamix.</p>
     </div>
     <div class="card">
         <h2>📥 Quantized Models</h2>
         <div class="quant-links">
-            <a href="" class="https://huggingface.co/Delta-Vector/Rei-V2-12B-EXL2/">EXL2 Quant</a>
-            <a href="" class="https://huggingface.co/Delta-Vector/Rei-V2-12B-GGUF">GGUF Quant</a>
         </div>
     </div>
     <div class="card">
@@ -216,19 +216,17 @@ Can I ask a question?&lt;|im_end|>
         <h2>⚙️ Training</h2>
         <h3>Hparams</h3>
         <ul>
-            <li>For Hparams for this model we experimented with Grad clipping otherwise known as max_grad_norm</li>
-            <li>Calculated via checking the model arch distribution's we divised 3 different values, knowing the weight distribution for Mistral is 0.1. </li>
-            <img src="https://cdn-uploads.huggingface.co/production/uploads/66c26b6fb01b19d8c3c2467b/2AWCq2cNBbq90VnqkIx79.png" width="500px" />
-            <li>If you would consult the graph, you'd notice a few things, First and foremost is that setting gradient-clip too high can be deterimental to the model as the logs and testing show that the model was overfit, Meanwhile setting it too low can also cause problems as the 1e-4 run appears to be underfit, The best one was by far the 0.001 clip which resulted in non-overfit, non-underfit model. </li>
         </ul>
         <h3>Configuration</h3>
         <div class="details">
             <details>
                 <summary>View Axolotl Config</summary>
-                <p>https://wandb.ai/new-eden/Rei-V2/artifacts/axolotl-config/config-7hvbucx9/v0/files/axolotl_config_pw8f0c6u.yml</p>
             </details>
         </div>
-        <p>The model was trained for 2 epochs on 8x <a href="https://www.nvidia.com/en-us/data-center/h200/" style="color: var(--accent);">NVIDIA H200s</a> GPUs generously provided by @Kalomaze</p>
         <div class="badge-container">
             <a href="https://github.com/OpenAccess-AI-Collective/axolotl">
                 <img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" class="badge">
@@ -238,11 +236,11 @@ Can I ask a question?&lt;|im_end|>
     <div class="card">
         <h2>⚠️ Credits</h2>
         <p><em>
-I'd like to thank, Ruka/Sama twinkman | LucyKnada | Kubernetes Bad | PocketDoc | Tav | Trappu | And the rest of Anthracite/Pygmalion for testing, feedback, and support.
 </em></p>
     </div>
     <div class="footer">
-        <p>Rei-12B | V2</p>
     </div>
 </body>
 </html>

 <body>
     <div class="header">
         <h1>Rei-12B</h1>
+        <p class="tagline">Another prototype Magnum... (This time with RL!)</p>
         <img src="https://cdn-uploads.huggingface.co/production/uploads/66c26b6fb01b19d8c3c2467b/nqMkoIsmScaTFHCFirGsc.png" alt="Rei Model" class="model-img" width="500px">
     </div>
     <div class="card">
         <h2>✨ Overview</h2>
+        <p>Taking the previous 12B trained with Subseqence Loss - This model is meant to refine the base's sharp edges and increase coherency, intelligence and prose while replicating the prose of the Claude models Opus and Sonnet</p>
+        <p>Fine-tuned on top of <a href="" style="color: var(--accent);">Mistral-Nemo-Instruct (ChatML'ified)</a>, Rei-12B is designed to replicate the exquisite prose quality of Claude 3 models, particularly Sonnet and Opus, using a prototype Magnum V5 datamix.</p>
     </div>
     <div class="card">
         <h2>📥 Quantized Models</h2>
         <div class="quant-links">
+            <a href="" class="">EXL2 Quant</a>
+            <a href="" class="">GGUF Quant</a>
         </div>
     </div>
     <div class="card">
         <h2>⚙️ Training</h2>
         <h3>Hparams</h3>
         <ul>
+            <li>For Hparams for this model we used a grad clip of 1e-4 as it was proven to the best value for Mistral-12B based models, and also to prevent Rewards/Chosen from flat-lining as Hermes-genned data is... The biggest piece of dogshit.</li>
+            <img src="https://cdn-uploads.huggingface.co/production/uploads/66c26b6fb01b19d8c3c2467b/tvOnEPhA9m0PvBCaAI1Re.png" width="500px" />
         </ul>
         <h3>Configuration</h3>
         <div class="details">
             <details>
                 <summary>View Axolotl Config</summary>
+                <p>https://wandb.ai/new-eden/KTO/artifacts/axolotl-config/config-eyt7d5i9/v0/files/axolotl_config_jvjuci1x.yml</p>
             </details>
         </div>
+        <p>The model was trained for 1 epochs on 8x <a href="https://www.nvidia.com/en-us/data-center/h100/" style="color: var(--accent);">NVIDIA H100s</a> GPUs generously provided by @Kalomaze</p>
         <div class="badge-container">
             <a href="https://github.com/OpenAccess-AI-Collective/axolotl">
                 <img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" class="badge">
     <div class="card">
         <h2>⚠️ Credits</h2>
         <p><em>
+I'd like to thank, Ruka/Sama twinkman | LucyKnada | Kubernetes Bad | PocketDoc | Tav | Trappu | Alicat | And the rest of Anthracite/Pygmalion for testing, feedback, and support.
 </em></p>
     </div>
     <div class="footer">
+        <p>Rei-12B | KTO</p>
     </div>
 </body>
 </html>