https://github.com/writer/AI-Adaptive-Learning-GESAL/blob/main/Adaptive_GESAL_NASA_dataset_2.ipynb

\n"}},{"time":"2025-02-28T17:50:09.256Z","user":"wassemgtk","userAvatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","type":"social-post-comment","socialPost":{"slug":"111123518019673","content":[{"type":"text","value":"# GESAL: Real-Time Adaptation for LLMs","raw":"# GESAL: Real-Time Adaptation for LLMs"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"We’re excited to unveil **Graph-Enhanced Singular Adaptive Learning (GESAL)**, a framework that lets LLMs like ","raw":"We’re excited to unveil **Graph-Enhanced Singular Adaptive Learning (GESAL)**, a framework that lets LLMs like "},{"type":"inline_code","code":"meta-llama/Llama-3.2-1B","raw":"`meta-llama/Llama-3.2-1B`"},{"type":"text","value":" adapt in real time using user feedback. Check out the code and white paper on GitHub!","raw":" adapt in real time using user feedback. Check out the code and white paper on GitHub!"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"🔗 **Code**: [https://github.com/writer/AI-Adaptive-Learning-GESAL](","raw":"🔗 **Code**: [https://github.com/writer/AI-Adaptive-Learning-GESAL]("},{"type":"link","href":"https://github.com/writer/AI-Adaptive-Learning-GESAL","raw":"https://github.com/writer/AI-Adaptive-Learning-GESAL"},{"type":"text","value":") ","raw":") "},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"---","raw":"---"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"## Why GESAL?","raw":"## Why GESAL?"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Static LLMs struggle to adapt without heavy retraining. GESAL solves this with:","raw":"Static LLMs struggle to adapt without heavy retraining. GESAL solves this with:"},{"type":"new_line","raw":"\n"},{"type":"text","value":"- **SVF**: Adapts weights via \\( W' = U (\\Sigma \\cdot z) V^T \\), using few parameters.","raw":"- **SVF**: Adapts weights via \\( W' = U (\\Sigma \\cdot z) V^T \\), using few parameters."},{"type":"new_line","raw":"\n"},{"type":"text","value":"- **Graph Memory**: Stores adaptations in nodes for scalability.","raw":"- **Graph Memory**: Stores adaptations in nodes for scalability."},{"type":"new_line","raw":"\n"},{"type":"text","value":"- **RL**: Updates via \\( J(z) = \\mathbb{E}[\\log \\pi_z(y|x) r] \\) based on feedback.","raw":"- **RL**: Updates via \\( J(z) = \\mathbb{E}[\\log \\pi_z(y|x) r] \\) based on feedback."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"---","raw":"---"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"## How It Works","raw":"## How It Works"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Ask \"How many R’s in ‘strawberry’?\" If it says \"2\" and you say \"no,\" GESAL learns to say \"3\" next time, avoiding repeats.","raw":"Ask \"How many R’s in ‘strawberry’?\" If it says \"2\" and you say \"no,\" GESAL learns to say \"3\" next time, avoiding repeats."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"---","raw":"---"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"## Try It","raw":"## Try It"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Built with Hugging Face’s ","raw":"Built with Hugging Face’s "},{"type":"inline_code","code":"transformers","raw":"`transformers`"},{"type":"text","value":":","raw":":"},{"type":"new_line","raw":"\n"},{"type":"code_fence","lang":"bash","code":"pip install transformers torch numpy\npython Adaptive_Learning_(GESAL).py","raw":"```bash\npip install transformers torch numpy\npython Adaptive_Learning_(GESAL).py\n```"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Needs a Hugging Face token for Llama-3.2-1B.","raw":"Needs a Hugging Face token for Llama-3.2-1B."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"---","raw":"---"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"## Results","raw":"## Results"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"GESAL hits 95% accuracy after 5 feedbacks vs. LoRA’s 70%. It’s efficient (~0.5M params) and scalable.","raw":"GESAL hits 95% accuracy after 5 feedbacks vs. LoRA’s 70%. It’s efficient (~0.5M params) and scalable."},{"type":"new_line","raw":"\n"}],"rawContent":"# GESAL: Real-Time Adaptation for LLMs\n\n\nWe’re excited to unveil **Graph-Enhanced Singular Adaptive Learning (GESAL)**, a framework that lets LLMs like `meta-llama/Llama-3.2-1B` adapt in real time using user feedback. Check out the code and white paper on GitHub!\n\n🔗 **Code**: [https://github.com/writer/AI-Adaptive-Learning-GESAL](https://github.com/writer/AI-Adaptive-Learning-GESAL) \n\n---\n\n## Why GESAL?\n\nStatic LLMs struggle to adapt without heavy retraining. GESAL solves this with:\n- **SVF**: Adapts weights via \\( W' = U (\\Sigma \\cdot z) V^T \\), using few parameters.\n- **Graph Memory**: Stores adaptations in nodes for scalability.\n- **RL**: Updates via \\( J(z) = \\mathbb{E}[\\log \\pi_z(y|x) r] \\) based on feedback.\n\n---\n\n## How It Works\n\nAsk \"How many R’s in ‘strawberry’?\" If it says \"2\" and you say \"no,\" GESAL learns to say \"3\" next time, avoiding repeats.\n\n---\n\n## Try It\n\nBuilt with Hugging Face’s `transformers`:\n```bash\npip install transformers torch numpy\npython Adaptive_Learning_(GESAL).py\n```\nNeeds a Hugging Face token for Llama-3.2-1B.\n\n---\n\n## Results\n\nGESAL hits 95% accuracy after 5 feedbacks vs. LoRA’s 70%. It’s efficient (~0.5M params) and scalable.\n","author":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43,"isFollowing":false},"attachments":[],"mentions":[],"reactions":[{"reaction":"🔥","users":["emredeveloper","kiranr","EquinoxElahin","John6666"],"count":4},{"reaction":"❤️","users":["oieieio","melisa"],"count":2},{"reaction":"🤗","users":["emredeveloper"],"count":1},{"reaction":"🧠","users":["hassenhamdi"],"count":1}],"publishedAt":"2025-02-25T15:33:59.000Z","updatedAt":"2025-02-28T21:18:06.876Z","commentators":[{"_id":"64bef7bc1363b5c799de6d44","avatarUrl":"/avatars/a9947c6d7ca98d7385efa5ee7f2fb9a8.svg","fullname":"hassenhamdi","name":"hassenhamdi","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":5,"isFollowing":false},{"_id":"6346822ebcb5e67902ab345b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6346822ebcb5e67902ab345b/I7j6_FU8HZhXvkVWPTxPj.png","fullname":"Jorge Alonso","name":"oieieio","type":"user","isPro":true,"isHf":false,"isMod":false,"followerCount":4,"isFollowing":false},{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43,"isFollowing":false}],"url":"/posts/wassemgtk/111123518019673","totalUniqueImpressions":1850,"identifiedLanguage":{"language":"en","probability":0.7727798819541931},"numComments":15},"comment":{"_id":"67c1f751a0da58325c73d5f9","raw":"One more idea add temporal layers\nIntegrate 1D convolutional layers or LSTM layers before the SVFLinear layers to capture temporal dependencies in the sensor data over cycles. something like;\n```\nclass AdaptiveGESAL(nn.Module):\n def __init__(self, input_dim=21, hidden_dim=128, num_nodes=50):\n super().__init__()\n self.conv1d = nn.Conv1d(input_dim, hidden_dim, kernel_size=3, padding=1)\n self.lstm = nn.LSTM(hidden_dim, hidden_dim, batch_first=True)\n self.svf_layers = [SVFLinear(nn.Linear(hidden_dim, hidden_dim)) for _ in range(2)]\n self.output_layer = nn.Linear(hidden_dim, 1) # RUL prediction\n # Graph and SVF initialization as before\n```\n\nWith replace MSE (implicit in RMSE) with a hybrid loss combining MSE and a quantile loss (e.g., 0.9 quantile for conservative RUL estimates) this penalizes underestimation, aligning with conservative RUL needs.\n","html":"

One more idea add temporal layers
Integrate 1D convolutional layers or LSTM layers before the SVFLinear layers to capture temporal dependencies in the sensor data over cycles. something like;

\n

class AdaptiveGESAL(nn.Module):\n    def __init__(self, input_dim=21, hidden_dim=128, num_nodes=50):\n        super().__init__()\n        self.conv1d = nn.Conv1d(input_dim, hidden_dim, kernel_size=3, padding=1)\n        self.lstm = nn.LSTM(hidden_dim, hidden_dim, batch_first=True)\n        self.svf_layers = [SVFLinear(nn.Linear(hidden_dim, hidden_dim)) for _ in range(2)]\n        self.output_layer = nn.Linear(hidden_dim, 1)  # RUL prediction\n        # Graph and SVF initialization as before\n

\n

With replace MSE (implicit in RMSE) with a hybrid loss combining MSE and a quantile loss (e.g., 0.9 quantile for conservative RUL estimates) this penalizes underestimation, aligning with conservative RUL needs.

\n"}}],"blogPosts":[{"_id":"6602077c4410b00b3f22e069","authors":[{"user":"jeffboudier"},{"user":"wassemgtk","guest":true}],"canonical":true,"isUpvotedByUser":false,"numCoauthors":2,"publishedAt":"2023-07-01T00:00:00.000Z","slug":"writer-case-study","title":"Leveraging Hugging Face for complex generative AI use cases","upvotes":0}],"totalBlogPosts":1,"canReadDatabase":false,"canManageEntities":false,"canImpersonate":false,"canManageBilling":false,"communityScore":31,"collections":[],"datasets":[{"author":"wassemgtk","downloads":85,"gated":false,"id":"wassemgtk/CoT-O1-examples","lastModified":"2024-09-16T21:32:51.000Z","datasetsServerInfo":{"viewer":"viewer","numRows":4,"libraries":["datasets","dask","mlcroissant"],"formats":["json"],"modalities":["text"]},"private":false,"repoType":"dataset","likes":0,"isLikedByUser":false}],"models":[{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":13,"gated":false,"id":"wassemgtk/mergekit-ties-gkhzksa","availableInferenceProviders":[],"lastModified":"2025-02-17T20:40:32.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":0,"gated":false,"id":"wassemgtk/mergekit-passthrough-gokgikt","availableInferenceProviders":[],"lastModified":"2024-11-14T01:57:16.000Z","likes":0,"private":false,"repoType":"model","isLikedByUser":false},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":0,"gated":"manual","id":"wassemgtk/mergekit-linear-tdzebun","availableInferenceProviders":[],"lastModified":"2024-06-16T02:28:07.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":9,"gated":false,"id":"wassemgtk/Tiny-ll","availableInferenceProviders":[],"lastModified":"2024-05-29T15:50:22.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":0,"gated":"manual","id":"wassemgtk/mt-best","availableInferenceProviders":[],"lastModified":"2024-05-18T00:35:31.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":12,"gated":false,"id":"wassemgtk/FC","availableInferenceProviders":[],"lastModified":"2024-05-17T20:53:14.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":10,"gated":false,"id":"wassemgtk/very-creative1","availableInferenceProviders":[],"lastModified":"2024-05-15T15:38:06.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":10,"gated":false,"id":"wassemgtk/very-creative","availableInferenceProviders":[],"lastModified":"2024-05-15T14:29:17.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":9,"gated":false,"id":"wassemgtk/big-p-128k","availableInferenceProviders":[],"lastModified":"2024-05-09T14:05:57.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":9,"gated":false,"id":"wassemgtk/big-p","availableInferenceProviders":[],"lastModified":"2024-05-09T13:09:06.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":12,"gated":false,"id":"wassemgtk/mistral-llama","availableInferenceProviders":[],"lastModified":"2024-05-07T13:07:46.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":9,"gated":false,"id":"wassemgtk/big-miqu-71","availableInferenceProviders":[],"lastModified":"2024-05-07T01:16:46.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":9,"gated":false,"id":"wassemgtk/4nd-big-llm","availableInferenceProviders":[],"lastModified":"2024-05-06T15:21:28.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":13,"gated":false,"id":"wassemgtk/merge-Meta-Llama-3-8B-Instruct-Nous-Hermes-2-Yi-34B","availableInferenceProviders":[],"lastModified":"2024-05-06T04:36:39.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":10,"gated":false,"id":"wassemgtk/merge-Nous-Hermes-2-Yi-34B-Llama-3-8B-Instruct-12B","availableInferenceProviders":[],"lastModified":"2024-05-06T04:20:21.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":68,"gated":false,"id":"wassemgtk/merge-passthrough-Meta-Llama-3-8B-Instruct","availableInferenceProviders":[],"lastModified":"2024-05-06T04:17:49.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":8,"gated":false,"id":"wassemgtk/merge-passthrough-Meta-Llama-3-Instruct-10B","availableInferenceProviders":[],"lastModified":"2024-05-06T04:16:26.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]},{"author":"wassemgtk","authorData":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43},"downloads":7,"gated":false,"id":"wassemgtk/merge-NeuralHermes-2.5-Mistral-7B-MetaMath","availableInferenceProviders":[],"lastModified":"2024-05-06T03:44:43.000Z","likes":0,"pipeline_tag":"text-generation","private":false,"repoType":"model","isLikedByUser":false,"widgetOutputUrls":[]}],"numberLikes":16,"papers":[{"id":"2502.06329","title":"Expect the Unexpected: FailSafe Long Context QA for Finance","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2502.06329.png","upvotes":126,"publishedAt":"2025-02-10T10:29:28.000Z","isUpvotedByUser":false},{"id":"2408.14906","title":"Writing in the Margins: Better Inference Pattern for Long Context\n Retrieval","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2408.14906.png","upvotes":140,"publishedAt":"2024-08-27T09:34:38.000Z","isUpvotedByUser":false},{"id":"2405.02048","title":"Comparative Analysis of Retrieval Systems in the Real World","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2405.02048.png","upvotes":0,"publishedAt":"2024-05-03T12:30:01.000Z","isUpvotedByUser":false},{"id":"2402.17553","title":"OmniACT: A Dataset and Benchmark for Enabling Multimodal Generalist\n Autonomous Agents for Desktop and Web","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2402.17553.png","upvotes":24,"publishedAt":"2024-02-27T14:47:53.000Z","isUpvotedByUser":false},{"id":"2307.03692","title":"Becoming self-instruct: introducing early stopping criteria for minimal\n instruct tuning","thumbnailUrl":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2307.03692.png","upvotes":26,"publishedAt":"2023-07-05T09:42:25.000Z","isUpvotedByUser":false}],"posts":[{"slug":"111123518019673","content":[{"type":"text","value":"# GESAL: Real-Time Adaptation for LLMs","raw":"# GESAL: Real-Time Adaptation for LLMs"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"We’re excited to unveil **Graph-Enhanced Singular Adaptive Learning (GESAL)**, a framework that lets LLMs like ","raw":"We’re excited to unveil **Graph-Enhanced Singular Adaptive Learning (GESAL)**, a framework that lets LLMs like "},{"type":"inline_code","code":"meta-llama/Llama-3.2-1B","raw":"`meta-llama/Llama-3.2-1B`"},{"type":"text","value":" adapt in real time using user feedback. Check out the code and white paper on GitHub!","raw":" adapt in real time using user feedback. Check out the code and white paper on GitHub!"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"🔗 **Code**: [https://github.com/writer/AI-Adaptive-Learning-GESAL](","raw":"🔗 **Code**: [https://github.com/writer/AI-Adaptive-Learning-GESAL]("},{"type":"link","href":"https://github.com/writer/AI-Adaptive-Learning-GESAL","raw":"https://github.com/writer/AI-Adaptive-Learning-GESAL"},{"type":"text","value":") ","raw":") "},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"---","raw":"---"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"## Why GESAL?","raw":"## Why GESAL?"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Static LLMs struggle to adapt without heavy retraining. GESAL solves this with:","raw":"Static LLMs struggle to adapt without heavy retraining. GESAL solves this with:"},{"type":"new_line","raw":"\n"},{"type":"text","value":"- **SVF**: Adapts weights via \\( W' = U (\\Sigma \\cdot z) V^T \\), using few parameters.","raw":"- **SVF**: Adapts weights via \\( W' = U (\\Sigma \\cdot z) V^T \\), using few parameters."},{"type":"new_line","raw":"\n"},{"type":"text","value":"- **Graph Memory**: Stores adaptations in nodes for scalability.","raw":"- **Graph Memory**: Stores adaptations in nodes for scalability."},{"type":"new_line","raw":"\n"},{"type":"text","value":"- **RL**: Updates via \\( J(z) = \\mathbb{E}[\\log \\pi_z(y|x) r] \\) based on feedback.","raw":"- **RL**: Updates via \\( J(z) = \\mathbb{E}[\\log \\pi_z(y|x) r] \\) based on feedback."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"---","raw":"---"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"## How It Works","raw":"## How It Works"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Ask \"How many R’s in ‘strawberry’?\" If it says \"2\" and you say \"no,\" GESAL learns to say \"3\" next time, avoiding repeats.","raw":"Ask \"How many R’s in ‘strawberry’?\" If it says \"2\" and you say \"no,\" GESAL learns to say \"3\" next time, avoiding repeats."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"---","raw":"---"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"## Try It","raw":"## Try It"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Built with Hugging Face’s ","raw":"Built with Hugging Face’s "},{"type":"inline_code","code":"transformers","raw":"`transformers`"},{"type":"text","value":":","raw":":"},{"type":"new_line","raw":"\n"},{"type":"code_fence","lang":"bash","code":"pip install transformers torch numpy\npython Adaptive_Learning_(GESAL).py","raw":"```bash\npip install transformers torch numpy\npython Adaptive_Learning_(GESAL).py\n```"},{"type":"new_line","raw":"\n"},{"type":"text","value":"Needs a Hugging Face token for Llama-3.2-1B.","raw":"Needs a Hugging Face token for Llama-3.2-1B."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"---","raw":"---"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"## Results","raw":"## Results"},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"GESAL hits 95% accuracy after 5 feedbacks vs. LoRA’s 70%. It’s efficient (~0.5M params) and scalable.","raw":"GESAL hits 95% accuracy after 5 feedbacks vs. LoRA’s 70%. It’s efficient (~0.5M params) and scalable."},{"type":"new_line","raw":"\n"}],"rawContent":"# GESAL: Real-Time Adaptation for LLMs\n\n\nWe’re excited to unveil **Graph-Enhanced Singular Adaptive Learning (GESAL)**, a framework that lets LLMs like `meta-llama/Llama-3.2-1B` adapt in real time using user feedback. Check out the code and white paper on GitHub!\n\n🔗 **Code**: [https://github.com/writer/AI-Adaptive-Learning-GESAL](https://github.com/writer/AI-Adaptive-Learning-GESAL) \n\n---\n\n## Why GESAL?\n\nStatic LLMs struggle to adapt without heavy retraining. GESAL solves this with:\n- **SVF**: Adapts weights via \\( W' = U (\\Sigma \\cdot z) V^T \\), using few parameters.\n- **Graph Memory**: Stores adaptations in nodes for scalability.\n- **RL**: Updates via \\( J(z) = \\mathbb{E}[\\log \\pi_z(y|x) r] \\) based on feedback.\n\n---\n\n## How It Works\n\nAsk \"How many R’s in ‘strawberry’?\" If it says \"2\" and you say \"no,\" GESAL learns to say \"3\" next time, avoiding repeats.\n\n---\n\n## Try It\n\nBuilt with Hugging Face’s `transformers`:\n```bash\npip install transformers torch numpy\npython Adaptive_Learning_(GESAL).py\n```\nNeeds a Hugging Face token for Llama-3.2-1B.\n\n---\n\n## Results\n\nGESAL hits 95% accuracy after 5 feedbacks vs. LoRA’s 70%. It’s efficient (~0.5M params) and scalable.\n","author":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43,"isFollowing":false},"attachments":[],"mentions":[],"reactions":[{"reaction":"🔥","users":["emredeveloper","kiranr","EquinoxElahin","John6666"],"count":4},{"reaction":"❤️","users":["oieieio","melisa"],"count":2},{"reaction":"🤗","users":["emredeveloper"],"count":1},{"reaction":"🧠","users":["hassenhamdi"],"count":1}],"publishedAt":"2025-02-25T15:33:59.000Z","updatedAt":"2025-02-28T21:18:06.876Z","commentators":[{"_id":"64bef7bc1363b5c799de6d44","avatarUrl":"/avatars/a9947c6d7ca98d7385efa5ee7f2fb9a8.svg","fullname":"hassenhamdi","name":"hassenhamdi","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":5,"isFollowing":false},{"_id":"6346822ebcb5e67902ab345b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6346822ebcb5e67902ab345b/I7j6_FU8HZhXvkVWPTxPj.png","fullname":"Jorge Alonso","name":"oieieio","type":"user","isPro":true,"isHf":false,"isMod":false,"followerCount":4,"isFollowing":false},{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43,"isFollowing":false}],"url":"/posts/wassemgtk/111123518019673","totalUniqueImpressions":1850,"identifiedLanguage":{"language":"en","probability":0.7727798819541931},"numComments":15},{"slug":"204283859077004","content":[{"type":"text","value":"Writer team had the opportunity to run an eval for Mixtral-8x22b, results were interesting.","raw":"Writer team had the opportunity to run an eval for Mixtral-8x22b, results were interesting."},{"type":"new_line","raw":"\n"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| ---------------------------- |","raw":"| ---------------------------- |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| #mmlu 77.26 |","raw":"| #mmlu 77.26 |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| ---------------------------- |","raw":"| ---------------------------- |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| #hellaswag 88.81 |","raw":"| #hellaswag 88.81 |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| ---------------------------- |","raw":"| ---------------------------- |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| #truthfulqa 52.05 |","raw":"| #truthfulqa 52.05 |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| ---------------------------- |","raw":"| ---------------------------- |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| #arc_challenge 70.31 |","raw":"| #arc_challenge 70.31 |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| ---------------------------- |","raw":"| ---------------------------- |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| #winogrande 84.93 | ","raw":"| #winogrande 84.93 | "},{"type":"new_line","raw":"\n"},{"type":"text","value":"| ---------------------------- |","raw":"| ---------------------------- |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| #gsm8k 76.65 |","raw":"| #gsm8k 76.65 |"},{"type":"new_line","raw":"\n"},{"type":"text","value":"| ---------------------------- |","raw":"| ---------------------------- |"}],"rawContent":"Writer team had the opportunity to run an eval for Mixtral-8x22b, results were interesting.\n\n| ---------------------------- |\n| #mmlu 77.26 |\n| ---------------------------- |\n| #hellaswag 88.81 |\n| ---------------------------- |\n| #truthfulqa 52.05 |\n| ---------------------------- |\n| #arc_challenge 70.31 |\n| ---------------------------- |\n| #winogrande 84.93 | \n| ---------------------------- |\n| #gsm8k 76.65 |\n| ---------------------------- |","author":{"_id":"60cd486d723acf5eb46fe8d3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","fullname":"Waseem AlShikh","name":"wassemgtk","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":43,"isFollowing":false},"attachments":[{"type":"image","url":"https://cdn-uploads.huggingface.co/production/uploads/60cd486d723acf5eb46fe8d3/3uzTjarADCM_X435pFvQ6.png"}],"mentions":[],"reactions":[{"reaction":"🔥","users":["den0620","kur4i","osanseviero","clem","Jason233","wannaphong","firstrow","ChuxiJ","nbroad"],"count":9}],"publishedAt":"2024-04-10T15:43:07.000Z","updatedAt":"2024-05-08T16:03:28.194Z","commentators":[{"_id":"657eb5b256c9c67605a6e8b5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/657eb5b256c9c67605a6e8b5/RPblnGJX57oiIcASEz_S8.png","fullname":"raincandy_U","name":"raincandy-u","type":"user","isPro":false,"isHf":false,"isMod":false,"followerCount":36,"isFollowing":false}],"url":"/posts/wassemgtk/204283859077004","totalUniqueImpressions":3640,"numComments":2}],"totalPosts":3,"spaces":[],"u":{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60cd486d723acf5eb46fe8d3/Z1bD1kjvZ0QAOjZna41Xr.jpeg","isPro":false,"fullname":"Waseem AlShikh","user":"wassemgtk","orgs":[{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1625001569797-60db8b5ad8b4797b129145d5.png","fullname":"Writer","name":"Writer","userRole":"admin","type":"org","isHf":false,"details":"AGI, LLMs, Knowledge Graph, Palmyra, Domain Specific LLM","isEnterprise":true,"numUsers":119},{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5f17f0a0925b9863e28ad517/nxmdd6m86cxu55UZBlQeg.jpeg","fullname":"Social Post Explorers","name":"social-post-explorers","userRole":"read","type":"org","isHf":false,"numUsers":863}],"signup":{"github":"wassemgtk","twitter":"waseem_s","details":"Multi-modal, Palmyra LLMs, Knowledge Graph","homepage":"https://writer.com/"},"isHf":false,"isMod":false,"type":"user"},"upvotes":5,"repoFilterModels":{"sortKey":"modified"},"repoFilterDatasets":{"sortKey":"modified"},"repoFilterSpaces":{"sortKey":"modified"},"numFollowers":43,"numFollowingUsers":104,"numFollowingOrgs":3,"isFollowing":false,"isFollower":false,"sampleFollowers":[{"user":"diego-lomanto","fullname":"Diego Lomanto","type":"user","_id":"66c6555b06149e7c2ae17325","isPro":false,"avatarUrl":"/avatars/8fad719583e1d27c338edb03546dbb6d.svg"},{"user":"yalexich","fullname":"Alexey","type":"user","_id":"67acda1a4a0dd70f38625876","isPro":false,"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/muiHiVrrxwrjJbOJ1IwHh.png"},{"user":"parikshithkulkarni","fullname":"Parikshith Kulkarni","type":"user","_id":"627426cfc3b4e8bb86b4cf8f","isPro":false,"avatarUrl":"/avatars/5cf841c7bb77bc83997061f1dadc4d87.svg"},{"user":"victor","fullname":"Victor Mustar","type":"user","_id":"5f17f0a0925b9863e28ad517","isPro":true,"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5f17f0a0925b9863e28ad517/fXIY5i9RLsIa1v3CCuVtt.jpeg"}],"isWatching":false,"hardwareItems":[{"sku":["Apple Silicon","-","Apple M3 Pro"],"mem":36,"num":1}],"acceptLanguages":["en","*"]}">