arliai/mistral-nemo-12b-arliai-rpmax-v1.2-gguf overview

{
  "metadata": {},
  "card_data": {
    "license": "apache-2.0",
    "frontmatter": {
      "license": "apache-2.0"
    },
    "hero_image_url": "",
    "summary": "# ArliAI-RPMax-12B-v1.2 =====================================",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nlicense: apache-2.0\n---\n# ArliAI-RPMax-12B-v1.2\n=====================================\n\n## UPDATE: For those getting gibberish results, it was merged wrongly to base after LORA training. Reuploaded all the files so it should work properly now.\n\n## RPMax Series Overview\n\n| [2B](https://huggingface.co/ArliAI/Gemma-2-2B-ArliAI-RPMax-v1.1) |\n[3.8B](https://huggingface.co/ArliAI/Phi-3.5-mini-3.8B-ArliAI-RPMax-v1.1) | \n[8B](https://huggingface.co/ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.2) | \n[9B](https://huggingface.co/ArliAI/Gemma-2-9B-ArliAI-RPMax-v1.1) |\n[12B](https://huggingface.co/ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2) | \n[20B](https://huggingface.co/ArliAI/InternLM2_5-20B-ArliAI-RPMax-v1.1) | \n[22B](https://huggingface.co/ArliAI/Mistral-Small-22B-ArliAI-RPMax-v1.1) | \n[70B](https://huggingface.co/ArliAI/Llama-3.1-70B-ArliAI-RPMax-v1.1) | \n\nRPMax is a series of models that are trained on a diverse set of curated creative writing and RP datasets with a focus on variety and deduplication. This model is designed to be highly creative and non-repetitive by making sure no two entries in the dataset have repeated characters or situations, which makes sure the model does not latch on to a certain personality and be capable of understanding and acting appropriately to any characters or situations.\n\nEarly tests by users mentioned that these models does not feel like any other RP models, having a different style and generally doesn't feel in-bred.\n\nYou can access the model at https://arliai.com and ask questions at https://www.reddit.com/r/ArliAI/\n\nWe also have a models ranking page at https://www.arliai.com/models-ranking\n\nAsk questions in our new Discord Server! https://discord.com/invite/t75KbPgwhk\n\n## Model Description\n\nArliAI-RPMax-12B-v1.2 is a variant based on Mistral Nemo 12B Instruct 2407.\n\nThis is arguably the most successful RPMax model due to how Mistral is already very uncensored in the first place.\n\nv1.2 update completely removes non-creative/RP examples in the dataset and is also an incremental improvement of the RPMax dataset which dedups the dataset even more and better filtering to cutout irrelevant description text that came from card sharing sites.\n\n### Specs\n\n* **Context Length**: 128K\n* **Parameters**: 12B\n\n### Training Details\n\n* **Sequence Length**: 8192\n* **Training Duration**: Approximately 2 days on 2x3090Ti\n* **Epochs**: 1 epoch training for minimized repetition sickness\n* **LORA**: 64-rank 128-alpha, resulting in ~2% trainable weights\n* **Learning Rate**: 0.00001\n* **Gradient accumulation**: Very low 32 for better learning.\n\n## Quantization\n\nThe model is available in quantized formats:\n\n* **FP16**: https://huggingface.co/ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2\n* **GGUF**: https://huggingface.co/ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2-GGUF\n\n\n## Suggested Prompt Format\n\nMistral Instruct Prompt Format",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "license:apache-2.0",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 20,
  "downloads": 596,
  "gated": false,
  "private": false,
  "last_modified": "2024-10-12T21:09:04.000Z",
  "created_at": "2024-10-12T18:47:17.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "670ac435623c91990f3c7e05",
  "id": "ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2-GGUF",
  "modelId": "ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.2-GGUF",
  "sha": "4b1fa90ea7c556ec3f5c27b507295048534ad94a",
  "createdAt": "2024-10-12T18:47:17.000Z",
  "lastModified": "2024-10-12T21:09:04.000Z",
  "author": "ArliAI",
  "downloads": 596,
  "likes": 20,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 11
}