richarderkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf overview

{
  "metadata": {},
  "card_data": {
    "frontmatter": {},
    "hero_image_url": "https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png",
    "summary": "This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the None dataset. It achieves the following results on the evaluation set:",
    "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\nshisa-v1-llama3-8b - GGUF\n- Model creator: https://huggingface.co/shisa-ai/\n- Original model: https://huggingface.co/shisa-ai/shisa-v1-llama3-8b/\n\n\n| Name | Quant method | Size |\n| ---- | ---- | ---- |\n| [shisa-v1-llama3-8b.Q2_K.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q2_K.gguf) | Q2_K | 2.96GB |\n| [shisa-v1-llama3-8b.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.IQ3_XS.gguf) | IQ3_XS | 3.28GB |\n| [shisa-v1-llama3-8b.IQ3_S.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.IQ3_S.gguf) | IQ3_S | 3.43GB |\n| [shisa-v1-llama3-8b.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q3_K_S.gguf) | Q3_K_S | 3.41GB |\n| [shisa-v1-llama3-8b.IQ3_M.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.IQ3_M.gguf) | IQ3_M | 3.52GB |\n| [shisa-v1-llama3-8b.Q3_K.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q3_K.gguf) | Q3_K | 3.74GB |\n| [shisa-v1-llama3-8b.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q3_K_M.gguf) | Q3_K_M | 3.74GB |\n| [shisa-v1-llama3-8b.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q3_K_L.gguf) | Q3_K_L | 4.03GB |\n| [shisa-v1-llama3-8b.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.IQ4_XS.gguf) | IQ4_XS | 4.18GB |\n| [shisa-v1-llama3-8b.Q4_0.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q4_0.gguf) | Q4_0 | 4.34GB |\n| [shisa-v1-llama3-8b.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.IQ4_NL.gguf) | IQ4_NL | 4.38GB |\n| [shisa-v1-llama3-8b.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q4_K_S.gguf) | Q4_K_S | 4.37GB |\n| [shisa-v1-llama3-8b.Q4_K.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q4_K.gguf) | Q4_K | 4.58GB |\n| [shisa-v1-llama3-8b.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q4_K_M.gguf) | Q4_K_M | 4.58GB |\n| [shisa-v1-llama3-8b.Q4_1.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q4_1.gguf) | Q4_1 | 4.78GB |\n| [shisa-v1-llama3-8b.Q5_0.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q5_0.gguf) | Q5_0 | 5.21GB |\n| [shisa-v1-llama3-8b.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q5_K_S.gguf) | Q5_K_S | 5.21GB |\n| [shisa-v1-llama3-8b.Q5_K.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q5_K.gguf) | Q5_K | 5.34GB |\n| [shisa-v1-llama3-8b.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q5_K_M.gguf) | Q5_K_M | 5.34GB |\n| [shisa-v1-llama3-8b.Q5_1.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q5_1.gguf) | Q5_1 | 5.65GB |\n| [shisa-v1-llama3-8b.Q6_K.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q6_K.gguf) | Q6_K | 6.14GB |\n| [shisa-v1-llama3-8b.Q8_0.gguf](https://huggingface.co/RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf/blob/main/shisa-v1-llama3-8b.Q8_0.gguf) | Q8_0 | 7.95GB |\n\n\n\n\nOriginal model description:\n---\nlicense: llama3\nbase_model: meta-llama/Meta-Llama-3-8B-Instruct\ntags:\n- generated_from_trainer\nmodel-index:\n- name: outputs/lr-8e6\n  results: []\ndatasets:\n- augmxnt/ultra-orca-boros-en-ja-v1\n---\n*Per the Llama 3 Community License Agreement, the official name of this model is \"LLama 3 shisa-v1-llama3-8b\"*\n\n8e6 moved in as it is a slightly superior model, will do some cleanup and renaming soon...\n\n\n\nI ran the tests for 2 runs just to try to lower variance. These are all using temp 0.2, min_p 0.1, freq penalty 0.5\n\n| Model                       | AVG Score | ELYZA100 | JA MT-Bench | Rakuda | Tengu-Bench | JA Char % |\n|-----------------------------|-----------|----------|-------------|--------|-------------|-----------|\n| shisa-v1-llama3-8b.lr-2e4   | 3.97      | 4.60     | 4.54        | 3.33   | 3.42        | 92.42%    |\n| shisa-v1-llama3-8b.lr-5e5   | 5.73      | 6.28     | 6.45        | 5.37   | 4.81        | 90.93%    |\n| shisa-v1-llama3-8b.2e5      | 6.33      | 6.51     | 6.66        | 6.68   | 5.48        | 91.51%    |\n| shisa-v1-llama3-8b (8-e6)   | 6.59      | 6.67     | 6.95        | 7.05   | 5.68        | 91.30%    |\n| shisa-v1-llama3-8b.5e6      | 6.42      | 6.33     | 6.76        | 7.15   | 5.45        | 91.56%    |\n| shisa-v1-llama3-8b.2e6      | 6.31      | 6.26     | 6.88        | 6.73   | 5.38        | 92.00%    |\n* The 2e-4 and 5e-5 are definitely overtrained and perform significantly worse.\n* 2e-5 is on the edge since weightwacher shows the embed as slightly overtrained for 2e-5, but NEFTune version is not\n* 8e-6 performs the best, and 5e-6 also performed slightly better than 2e-5\n\nFor a comparison of where it sits vs other models:\n\n| Model                                  | Average | ELYZA-tasks-100 | MT-Bench | Rakuda | Tengu-Bench |\n|----------------------------------------|---------|-----------------|----------|--------|-------------|\n| gpt-4-turbo-2024-04-09                 | 8.75    | 8.78            | 8.74     | 9.18   | 8.31        |\n| gpt-4o-2024-05-13                      | 8.72    | 8.88            | 8.69     | 9.15   | 8.16        |\n| gemini-1.5-pro                         | 8.58    | 8.58            | 8.93     | 9.20   | 7.61        |\n| claude-3-opus-20240229                 | 8.55    | 8.64            | 8.58     | 8.75   | 8.23        |\n| CohereForAI/c4ai-command-r-plus        | 7.69    | 7.50            | 7.43     | 9.05   | 6.79        |\n| **shisa-ai/shisa-v1-llama3-70b**       | **7.30**| **7.34**        | **7.67** | **8.15** | **6.04**  |\n| gpt-3.5-turbo-0125                     | 7.17    | 7.24            | 6.98     | 7.64   | 6.82        |\n| **shisa-ai/shisa-v1-llama3-70b.2e5**   | **7.17**| **7.16**        | **7.45** | **7.98** | **6.09**  |\n| karakuri-ai/karakuri-lm-8x7b-chat-v0.1 | 7.00    | 7.18            | 6.30     | 7.98   | 6.55        |\n| karakuri-ai/karakuri-lm-70b-chat-v0.1  | 6.84    | 6.86            | 6.43     | 7.85   | 6.23        |\n| lightblue/ao-karasu-72B                | 6.81    | 7.19            | 6.54     | 7.25   | 6.27        |\n| **shisa-ai/shisa-v1-llama3-8b**        | **6.59**| **6.67**        | **6.95** | **7.05**| **5.68**   |\n| **shisa-ai/shisa-swallowmx-13a47b-v1** | **6.17**| **6.48**        | **6.07** | **7.11**| **5.03**   |\n| lightblue/suzume-llama-3-8B-japanese   | 5.96    | 6.68            | 4.96     | 6.68   | 5.53        |\n| augmxnt/shisa-gamma-7b-v1              | 5.82    | 5.96            | 5.02     | 6.85   | 5.47        |\n| **shisa-ai/shisa-v1-phi3-14b**         | **5.77**| **6.28**        | **5.26** | **6.55**| **5.01**   |\n| **shisa-ai/shisa-v1-gemma-8b**         | **5.64**| **6.50**        | **5.42** | **5.10**| **5.55**   |\n| Rakuten/RakutenAI-7B-chat              | 5.58    | 5.92            | 4.60     | 6.58   | 5.24        |\n| lightblue/qarasu-14B-chat-plus-unleashed | 5.20  | 5.58            | 4.74     | 5.46   | 5.01        |\n| **shisa-ai/shisa-v1-mistral0.3-7b**    | **5.11**| **5.64**        | **6.10** | **3.83**|**4.86**    |\n| cyberagent/calm2-7b-chat               | 4.76    | 4.90            | 3.58     | 5.75   | 4.81        |\n| mistralai/Mistral-7B-Instruct-v0.2     | 4.69    | 5.78            | 4.65     | 3.80   | 4.53        |\n| **shisa-ai/shisa-v1-yi1.5-9b**         | **4.63**| **5.98**        | **4.28** | **3.26**|**5.00**    |\n| augmxnt/shisa-7b-v1                    | 4.50    | 4.63            | 3.95     | 4.89   | 4.53        |\n\n\n\n<!-- This model card has been generated automatically according to the information the Trainer had access to. You\nshould probably proofread and complete it, then remove this comment. -->\n\n[<img src=\"https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png\" alt=\"Built with Axolotl\" width=\"200\" height=\"32\"/>](https://github.com/OpenAccess-AI-Collective/axolotl)\n<details><summary>See axolotl config</summary>\n\naxolotl version: `0.4.0`\n```yaml\nbase_model: meta-llama/Meta-Llama-3-8B-Instruct\nmodel_type: LlamaForCausalLM\ntokenizer_type: AutoTokenizer\n\nload_in_8bit: false\nload_in_4bit: false\nstrict: false\n\nchat_template: llama3\ndatasets:\n  - path: augmxnt/ultra-orca-boros-en-ja-v1\n    type: sharegpt\ndataset_prepared_path: last_run_prepared\nval_set_size: 0.05\noutput_dir: ./outputs/lr-8e6\n\nsequence_len: 8192\nsample_packing: true\npad_to_sequence_len: true\n\nuse_wandb: true\nwandb_project: shisa-v2\nwandb_entity: augmxnt\nwandb_name: shisa-v1-llama3-8b.lr-8e6\n\ngradient_accumulation_steps: 8\nmicro_batch_size: 1\nnum_epochs: 3\noptimizer: paged_adamw_8bit\nlr_scheduler: linear\nlearning_rate: 8e-6\n\ntrain_on_inputs: false\ngroup_by_length: false\nbf16: auto\nfp16:\ntf32: false\n\ngradient_checkpointing: true\ngradient_checkpointing_kwargs:\n  use_reentrant: false\nearly_stopping_patience:\nresume_from_checkpoint:\nlogging_steps: 1\nxformers_attention:\nflash_attention: true\n\nwarmup_steps: 100\nevals_per_epoch: 2\neval_table_size:\nsaves_per_epoch: 0\ndebug:\ndeepspeed: axolotl/deepspeed_configs/zero3_bf16.json\nweight_decay: 0.00\nfsdp:\nfsdp_config:\nspecial_tokens:\n  pad_token: <|end_of_text|>\n\n```\n\n</details><br>\n\n# outputs/lr-8e6\n\nThis model is a fine-tuned version of [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) on the None dataset.\nIt achieves the following results on the evaluation set:\n- Loss: 0.4983\n\n## Model description\n\nMore information needed\n\n## Intended uses & limitations\n\nMore information needed\n\n## Training and evaluation data\n\nMore information needed\n\n## Training procedure\n\n### Training hyperparameters\n\nThe following hyperparameters were used during training:\n- learning_rate: 8e-06\n- train_batch_size: 1\n- eval_batch_size: 1\n- seed: 42\n- distributed_type: multi-GPU\n- num_devices: 8\n- gradient_accumulation_steps: 8\n- total_train_batch_size: 64\n- total_eval_batch_size: 8\n- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08\n- lr_scheduler_type: linear\n- lr_scheduler_warmup_steps: 100\n- num_epochs: 3\n\n### Training results\n\n| Training Loss | Epoch  | Step | Validation Loss |\n|:-------------:|:------:|:----:|:---------------:|\n| 1.3951        | 0.0064 | 1    | 0.8645          |\n| 0.8731        | 0.5020 | 79   | 0.5577          |\n| 0.8405        | 1.0040 | 158  | 0.5138          |\n| 0.6888        | 1.4853 | 237  | 0.4982          |\n| 0.6674        | 1.9873 | 316  | 0.4870          |\n| 0.5859        | 2.4694 | 395  | 0.4983          |\n\n\n### Framework versions\n\n- Transformers 4.40.2\n- Pytorch 2.3.0+cu121\n- Datasets 2.19.1\n- Tokenizers 0.19.1\n\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 0,
  "downloads": 83,
  "gated": false,
  "private": false,
  "last_modified": "2024-08-21T07:24:18.000Z",
  "created_at": "2024-08-21T05:26:03.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "66c57a6b15e26b87d61ffe52",
  "id": "RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf",
  "modelId": "RichardErkhov/shisa-ai_-_shisa-v1-llama3-8b-gguf",
  "sha": "5438dabd4c0187db19da9a062e311f31e14a2b79",
  "createdAt": "2024-08-21T05:26:03.000Z",
  "lastModified": "2024-08-21T07:24:18.000Z",
  "author": "RichardErkhov",
  "downloads": 83,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 24
}