richarderkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf overview

{
  "metadata": {},
  "card_data": {
    "frontmatter": {},
    "hero_image_url": "https://netspresso-research-code-release.s3.us-east-2.amazonaws.com/compressed-llm/st_llm-cpt_results.png",
    "summary": "Shortened LLM is a depth-pruned version of large language models for efficient text generation.",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "Quantization made by Richard Erkhov.\n\n[Github](https://github.com/RichardErkhov)\n\n[Discord](https://discord.gg/pvy7H8DZMG)\n\n[Request more models](https://github.com/RichardErkhov/quant_request)\n\n\ncpt_st-vicuna-v1.3-1.5b-ppl - GGUF\n- Model creator: https://huggingface.co/nota-ai/\n- Original model: https://huggingface.co/nota-ai/cpt_st-vicuna-v1.3-1.5b-ppl/\n\n\n| Name | Quant method | Size |\n| ---- | ---- | ---- |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q2_K.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q2_K.gguf) | Q2_K | 0.56GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.IQ3_XS.gguf) | IQ3_XS | 0.61GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.IQ3_S.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.IQ3_S.gguf) | IQ3_S | 0.64GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q3_K_S.gguf) | Q3_K_S | 0.64GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.IQ3_M.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.IQ3_M.gguf) | IQ3_M | 0.66GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q3_K.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q3_K.gguf) | Q3_K | 0.7GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q3_K_M.gguf) | Q3_K_M | 0.7GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q3_K_L.gguf) | Q3_K_L | 0.75GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.IQ4_XS.gguf) | IQ4_XS | 0.77GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q4_0.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q4_0.gguf) | Q4_0 | 0.81GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.IQ4_NL.gguf) | IQ4_NL | 0.81GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q4_K_S.gguf) | Q4_K_S | 0.81GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q4_K.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q4_K.gguf) | Q4_K | 0.84GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q4_K_M.gguf) | Q4_K_M | 0.84GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q4_1.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q4_1.gguf) | Q4_1 | 0.88GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q5_0.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q5_0.gguf) | Q5_0 | 0.96GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q5_K_S.gguf) | Q5_K_S | 0.96GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q5_K.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q5_K.gguf) | Q5_K | 0.98GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q5_K_M.gguf) | Q5_K_M | 0.98GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q5_1.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q5_1.gguf) | Q5_1 | 1.04GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q6_K.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q6_K.gguf) | Q6_K | 1.13GB |\n| [cpt_st-vicuna-v1.3-1.5b-ppl.Q8_0.gguf](https://huggingface.co/RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf/blob/main/cpt_st-vicuna-v1.3-1.5b-ppl.Q8_0.gguf) | Q8_0 | 1.46GB |\n\n\n\n\nOriginal model description:\n# Shortened LLM Model Card\n\nShortened LLM is a depth-pruned version of large language models for efficient text generation.\n\n- **Developed by:** [Nota AI](https://www.nota.ai/)\n- **License:** Non-commercial license\n- **Repository:** https://github.com/Nota-NetsPresso/shortened-llm\n- **Paper:** https://arxiv.org/abs/2402.02834\n\n## Compression Method\n* After identifying unimportant Transformer blocks, we perform **one-shot pruning**.\n* In retraining pruned models for quality recovery, **continued pretraining (CPT)** on a large corpus markedly outperforms LoRA-based tuning, particularly at severe pruning ratios.\n\n## Models from Aggressive Pruning & CPT Retraining (arXiv-v2):\n  | Source<br>Model | Pruning<br>Ratio | Pruning<br>Criterion | HF Models<br>Link |\n  |:---:|:---:|:---:|:---:|\n  | Vicuna-v1.3-7B | 20% | PPL | [nota-ai/cpt_st-vicuna-v1.3-5.5b-ppl](https://huggingface.co/nota-ai/cpt_st-vicuna-v1.3-5.5b-ppl) |\n  | Vicuna-v1.3-7B | 45% | PPL | [nota-ai/cpt_st-vicuna-v1.3-3.7b-ppl](https://huggingface.co/nota-ai/cpt_st-vicuna-v1.3-3.7b-ppl) |\n  | Vicuna-v1.3-7B | 60% | PPL | [nota-ai/cpt_st-vicuna-v1.3-2.7b-ppl](https://huggingface.co/nota-ai/cpt_st-vicuna-v1.3-2.7b-ppl) |\n  | Vicuna-v1.3-7B | 80% | PPL | [nota-ai/cpt_st-vicuna-v1.3-1.5b-ppl](https://huggingface.co/nota-ai/cpt_st-vicuna-v1.3-1.5b-ppl) |\n\n<details>\n<summary>\nClick to see the results:\n</summary>\n\n- EleutherAI/lm-evaluation-harness version [3326c54](https://github.com/EleutherAI/lm-evaluation-harness/tree/3326c547a733d598b4377e54be96e194861b964c)\n\n<img alt=\"results\" img src=\"https://netspresso-research-code-release.s3.us-east-2.amazonaws.com/compressed-llm/st_llm-cpt_results.png\" width=\"100%\">\n\n</details>\n\n#### Experimental Setup for CPT of Pruned Vicuna-7B\n* Dataset: [SlimPajama-627B](https://huggingface.co/datasets/cerebras/SlimPajama-627B) \n* Training using 8 NVIDIA H100 GPUs.\n  * 5.5B parameters: 37B training tokens (for 6 days)\n  * 3.7B parameters: 74B tokens (for 8 days)\n  * 2.7B parameters: 150B tokens (for 12 days)\n  * 1.5B parameters: 271B tokens (for 11 days)\n* AdamW optimizer with (β1, β2)=(0.9, 0.95); a learning rate of 0.0001; a weight decay of 0.1.\n* Global batch size: 512 (micro-batch size of 2 × 32 gradient accumulation steps × 8 GPUs).\n\n<details>\n<summary>\nClick to see the learning curve:\n</summary>\n\n**Zero-shot performance over the course of training for models from Vicuna-7B-v1.3 at different pruning ratios.** For each model size, the CPT duration was limited to a two-week period, but additional training could further improve the quality.\n\n<img alt=\"results\" img src=\"https://netspresso-research-code-release.s3.us-east-2.amazonaws.com/compressed-llm/st_llm-cpt_learning-curve.png\" width=\"100%\">\n\n</details>\n\n\n\n## Models from Moderate Pruning & LoRA Retraining (arXiv-v1):\n  | Source<br>Model | Pruning<br>Ratio | Pruning<br>Criterion | HF Models<br>Link |\n  |:---:|:---:|:---:|:---:|\n  | LLaMA-1-7B | 20% | PPL | [nota-ai/st-llama-1-5.5b-ppl](https://huggingface.co/nota-ai/st-llama-1-5.5b-ppl) |\n  | LLaMA-1-7B | 20% | Taylor+ | [nota-ai/st-llama-1-5.5b-taylor](https://huggingface.co/nota-ai/st-llama-1-5.5b-taylor) |\n  | Vicuna-v1.3-7B | 20% | PPL | [nota-ai/st-vicuna-v1.3-5.5b-ppl](https://huggingface.co/nota-ai/st-vicuna-v1.3-5.5b-ppl) |\n  | Vicuna-v1.3-7B | 20% | Taylor+ | [nota-ai/st-vicuna-v1.3-5.5b-taylor](https://huggingface.co/nota-ai/st-vicuna-v1.3-5.5b-taylor) |\n  | Vicuna-v1.3-13B | 21% | PPL | [nota-ai/st-vicuna-v1.3-10.5b-ppl](https://huggingface.co/nota-ai/st-vicuna-v1.3-10.5b-ppl) |\n  | Vicuna-v1.3-13B | 21% | Taylor+ | [nota-ai/st-vicuna-v1.3-10.5b-taylor](https://huggingface.co/nota-ai/st-vicuna-v1.3-10.5b-taylor) |\n\n<details>\n\n<summary>\nClick to see the results:\n</summary>\n\n- EleutherAI/lm-evaluation-harness version [3326c54](https://github.com/EleutherAI/lm-evaluation-harness/tree/3326c547a733d598b4377e54be96e194861b964c)\n\n<img alt=\"results\" img src=\"https://netspresso-research-code-release.s3.us-east-2.amazonaws.com/compressed-llm/st-llama_zero-shot_scores.png\" width=\"100%\">\n\n</details>\n\n## License\n- All rights related to this repository and the compressed models are reserved by Nota Inc.\n- The intended use is strictly limited to research and non-commercial projects.\n\n## Acknowledgments\n- [Microsoft for Startups Founders Hub](https://www.microsoft.com/en-us/startups) and [Gwangju AICA](http://www.aica-gj.kr/main.php) for generously providing GPU resources.\n- [LLM-Pruner](https://github.com/horseee/LLM-Pruner), which utilizes [LM Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness), [PEFT](https://github.com/huggingface/peft), and [Alpaca-LoRA](https://github.com/tloen/alpaca-lora). Thanks for the pioneering work on structured pruning of LLMs! \n- Meta AI's [LLaMA](https://github.com/facebookresearch/llama) and  LMSYS Org's [Vicuna](https://github.com/lm-sys/FastChat/blob/main/docs/vicuna_weights_version.md). Thanks for the open-source LLMs!\n\n## Citation\n```bibtex\n@article{kim2024shortened,\n  title={Shortened LLaMA: Depth Pruning for Large Language Models with Comparison of Retraining Methods},\n  author={Kim, Bo-Kyeong and Kim, Geonmin and Kim, Tae-Ho and Castells, Thibault and Choi, Shinkook and Shin, Junho and Song, Hyoung-Kyu},\n  journal={arXiv preprint arXiv:2402.02834},      \n  year={2024},\n  url={https://arxiv.org/abs/2402.02834}\n}\n```\n```bibtex\n@article{kim2024mefomo,\n  title={Shortened LLaMA: A Simple Depth Pruning for Large Language Models},\n  author={Kim, Bo-Kyeong and Kim, Geonmin and Kim, Tae-Ho and Castells, Thibault and Choi, Shinkook and Shin, Junho and Song, Hyoung-Kyu},\n  journal={ICLR Workshop on Mathematical and Empirical Understanding of Foundation Models (ME-FoMo)},\n  year={2024},\n  url={https://openreview.net/forum?id=18VGxuOdpu}\n}\n```\n\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "arxiv:2402.02834",
    "endpoints_compatible",
    "region:us"
  ],
  "likes": 0,
  "downloads": 96,
  "gated": false,
  "private": false,
  "last_modified": "2024-07-05T13:50:12.000Z",
  "created_at": "2024-07-05T13:42:33.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "6687f849060faa8ce6b88ded",
  "id": "RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf",
  "modelId": "RichardErkhov/nota-ai_-_cpt_st-vicuna-v1.3-1.5b-ppl-gguf",
  "sha": "bc17dbbab5dfd6b20f57b9cd770882c1821fc7fb",
  "createdAt": "2024-07-05T13:42:33.000Z",
  "lastModified": "2024-07-05T13:50:12.000Z",
  "author": "RichardErkhov",
  "downloads": 96,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 24
}