richarderkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf overview

{
  "metadata": {},
  "card_data": {
    "frontmatter": {},
    "hero_image_url": "",
    "summary": "Quantization made by Richard Erkhov. Github Discord Request more models llama-3-typhoon-v1.5x-70b-instruct - GGUF | Name | Quant method | Size | | ---- | ---- | ---- | | llama-3-typhoon-v1.5x-70b-instruct.Q2_K.gguf | Q2_K | 24.56GB | | llama-3-typhoon-v1.5x-70b-instruct.IQ3_XS.gguf | IQ3_XS | 27.29GB | | llama-3-typhoon-v1.5x-70b-instruct.IQ3_S.gguf | IQ3_S | 28.79GB | | llama-3-typhoon-v1.5x-70b-instruct.Q3_K_S.gguf | Q3_K_S | 28.79GB | | llama-3-typhoon-v1.5x-70b-instruct.IQ3_M.gguf | IQ3_M | 29.74GB | | llama-3-typhoon-v1.5x-70b-instruct.Q3_K.gguf | Q3_K | 31.91GB | | llama-3-typhoon-v1.5x-70b-instruct.Q3_K_M.gguf | Q3_K_M | 31.91GB | | llama-3-typhoon-v1.5x-70b-instruct.Q3_K_L.gguf | Q3_K_L | 34.59GB | | llama-3-typhoon-v1.5x-70b-instruct.IQ4_XS.gguf | IQ4_XS | 35.64GB | | llama-3-typhoon-v1.5x-70b-instruct.Q4_0.gguf | Q4_0 | 37.22GB | | llama-3-typhoon-v1.5x-70b-instruct.IQ4_NL.gguf | IQ4_NL | 37.58GB | | llama-3-typhoon-v1.5x-70b-instruct.Q4_K_S.gguf | Q4_K_S | 37.58GB | | llama-3-typhoon-v1.5x-70b-instruct.Q4_K.gguf | Q4_K | 39.6GB | | llama-3-typhoon-v1.5x-70b-instruct.Q4_K_M.gguf | Q4_K_M | 39.6GB | | llama-3-typhoon-v1.5x-70b-instruct.Q4_1.gguf | Q4_1 | 41.27GB | | llama-3-typhoon-v1.5x-70b-instruct.Q5_0.gguf | Q5_0 | 45.32GB | | llama-3-typhoon-v1.5x-70b-instruct.Q5_K_S.gguf | Q5_K_S | 45.32GB | | llama-3-typhoon-v1.5x-70b-instruct.Q5_K.gguf | Q5_K | 46.52GB | | llama-3-typhoon-v1.5x-70b-instruct.Q5_K_M.gguf | Q5_K_M | 46.52GB | | llama-3-typhoon-v1.5x-70b-instruct.Q5_1.gguf | Q5_1 | 49.36GB | | llama-3-typhoon-v1.5x-70b-instruct.Q6_K.gguf | Q6_K | 53.91GB | | llama-3-typhoon-v1.5x-70b-instruct.Q8_0.gguf | Q8_0 | 69.83GB | Original model description: --- language: pipeline_tag: text-generation license: llama3 --- **Llama-3-Typhoon-1.5X-70B-instruct: Thai Large Language Model (Instruct)** **Llama-3-Typhoon-1.5X-70B-instruct** is a 70 billion parameter instruct model designed for Thai 🇹🇭 language. It demonstrates competitive performance with GPT-4-0612, and is optimized for **application** use cases, **Retrieval-Augmented Generation (RAG), constrained generation**, and **reasoning** tasks. Built on Typhoon 1.5 70B (not yet released) and Llama 3 70B Instruct. this model is a result of our experiment on **cross-lingual transfer**. It utilizes the task-arithmetic model editing technique, combining the Thai understanding capability of Typhoon with the human alignment performance of Llama 3 Instruct. Remark: To acknowledge Meta's efforts in creating the foundation model and comply with the license, we explicitly include \"llama-3\" in the model name.",
    "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\nllama-3-typhoon-v1.5x-70b-instruct - GGUF\n- Model creator: https://huggingface.co/scb10x/\n- Original model: https://huggingface.co/scb10x/llama-3-typhoon-v1.5x-70b-instruct/\n\n\n| Name | Quant method | Size |\n| ---- | ---- | ---- |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q2_K.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.Q2_K.gguf) | Q2_K | 24.56GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.IQ3_XS.gguf) | IQ3_XS | 27.29GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.IQ3_S.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.IQ3_S.gguf) | IQ3_S | 28.79GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.Q3_K_S.gguf) | Q3_K_S | 28.79GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.IQ3_M.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.IQ3_M.gguf) | IQ3_M | 29.74GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q3_K.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.Q3_K.gguf) | Q3_K | 31.91GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.Q3_K_M.gguf) | Q3_K_M | 31.91GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.Q3_K_L.gguf) | Q3_K_L | 34.59GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.IQ4_XS.gguf) | IQ4_XS | 35.64GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q4_0.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/blob/main/llama-3-typhoon-v1.5x-70b-instruct.Q4_0.gguf) | Q4_0 | 37.22GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | IQ4_NL | 37.58GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q4_K_S | 37.58GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q4_K.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q4_K | 39.6GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q4_K_M | 39.6GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q4_1.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q4_1 | 41.27GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q5_0.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q5_0 | 45.32GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q5_K_S | 45.32GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q5_K.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q5_K | 46.52GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q5_K_M | 46.52GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q5_1.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q5_1 | 49.36GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q6_K.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q6_K | 53.91GB |\n| [llama-3-typhoon-v1.5x-70b-instruct.Q8_0.gguf](https://huggingface.co/RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf/tree/main/) | Q8_0 | 69.83GB |\n\n\n\n\nOriginal model description:\n---\nlanguage:\n- th\n- en\npipeline_tag: text-generation\nlicense: llama3\n---\n**Llama-3-Typhoon-1.5X-70B-instruct: Thai Large Language Model (Instruct)**\n\n**Llama-3-Typhoon-1.5X-70B-instruct** is a 70 billion parameter instruct model designed for Thai 🇹🇭 language. It demonstrates competitive performance with GPT-4-0612, and is optimized for **application** use cases, **Retrieval-Augmented Generation (RAG), constrained generation**, and **reasoning** tasks.\n\nBuilt on Typhoon 1.5 70B (not yet released) and Llama 3 70B Instruct. this model is a result of our experiment on **cross-lingual transfer**. It utilizes the [task-arithmetic model editing](https://arxiv.org/abs/2212.04089) technique, combining the Thai understanding capability of Typhoon with the human alignment performance of Llama 3 Instruct.\n\nRemark: To acknowledge Meta's efforts in creating the foundation model and comply with the license, we explicitly include \"llama-3\" in the model name.\n\n## **Model Description**\n\n- **Model type**: A 70B instruct decoder-only model based on the Llama architecture\n- **Requirement**: Transformers 4.38.0 or newer\n- **Primary Language(s)**: Thai 🇹🇭 and English 🇬🇧\n- **License**: [**Llama 3 Community License**](https://llama.meta.com/llama3/license/)\n\n## **Performance**\n\nWe evaluated the model's performance in **Language & Knowledge Capabilities** and **Instruction Following Capabilities**.\n\n- **Language & Knowledge Capabilities**:\n    - Assessed using multiple-choice question-answering datasets such as ThaiExam and MMLU.\n- **Instruction Following Capabilities**:\n    - Evaluated based on beta users' feedback, focusing on two factors:\n        - **Human Alignment & Reasoning**: Ability to generate responses that are clear and logically structured across multiple steps.\n            - Evaluated using [MT-Bench](https://arxiv.org/abs/2306.05685) — How LLMs can align with human needs.\n        - **Instruction-following**: Ability to adhere to specified constraints in the instructions.\n            - Evaluated using [IFEval](https://arxiv.org/abs/2311.07911) — How LLMs can follow specified constraints, such as formatting and brevity.\n- **Agentic Capabilities**:\n    - Evaluated in agent use-cases using [Hugging Face's Transformer Agents](https://huggingface.co/blog/agents) and the associated [benchmark](https://huggingface.co/blog/open-source-llms-as-agents).\n\nRemark: We developed the Thai (TH) pairs by translating the original datasets into Thai through machine and human methods.\n\n### ThaiExam\n\n| Model | ONET | IC | TGAT | TPAT-1 | A-Level | Average (ThaiExam) | MMLU |\n| --- | --- | --- | --- | --- | --- | --- | --- |\n| Typhoon-1.5X 70B | **0.565** | 0.68 | **0.778** | **0.517** | 0.56 | **0.620** | 0.7945 |\n| gpt-4-0612 | 0.493 | **0.69** | 0.744 | 0.509 | **0.616** | 0.610 | **0.864**** |\n| --- | --- | --- | --- | --- | --- | --- | --- |\n| gpt-4o | 0.62 | 0.63 | 0.789 | 0.56 | 0.623 | 0.644 | 0.887** |\n\n** We report the MMLU score that is reported in [GPT-4o Tech Report](https://openai.com/index/hello-gpt-4o/). \n\n### MT-Bench\n\n| Model | MT-Bench Thai | MT-Bench English |\n| --- | --- | --- |\n| Typhoon-1.5X 70B | **8.029** | **8.797** |\n| gpt-4-0612 | 7.801 | 8.671 |\n| --- | --- | --- |\n| gpt-4o | 8.514 | 9.184 |\n\n### IFEval\n\n| Model | IFEval Thai | IFEval English |\n| --- | --- | --- |\n| Typhoon-1.5X 70B | **0.645** | **0.810** |\n| gpt-4-0612 | 0.612 | 0.793* |\n| --- | --- | --- |\n| gpt-4o | 0.737 | 0.871 |\n\n* We report the number from IFEval paper.\n\n### Agent\n\n| Model | GAIA - Thai/English | GSM8K - Thai/English | HotpotQA - Thai/English |\n| --- | --- | --- | --- |\n| gpt-3.5-turbo-0125 | **18.42**/37.5 | 70/80 | 39.56/59 |\n| Typhoon-1.5X 70B | 17.10/36.25 | 80/95 | 52.7/65.83 |\n| gpt-4-0612 | 17.10/**38.75** | **90**/**100** | **56.41**/**76.25** |\n| --- | --- | --- | --- |\n| gpt-4o | 44.73/57.5 | 100/100 | 71.64/76.58 |\n\n## Insight\n\nWe utilized **model editing** techniques and found that the most critical feature for generating accurate Thai answers is located in the backend (the upper layers of the transformer block). Accordingly, we incorporated a high ratio of Typhoon components in these backend layers to enhance our model’s performance.\n\n## **Usage Example**\n\n```python\nfrom transformers import AutoTokenizer, AutoModelForCausalLM\nimport torch\n\nmodel_id = \"scb10x/llama-3-typhoon-v1.5x-70b-instruct\"\n\ntokenizer = AutoTokenizer.from_pretrained(model_id)\nmodel = AutoModelForCausalLM.from_pretrained(\n    model_id,\n    torch_dtype=torch.bfloat16,\n    device_map=\"auto\",\n) # We don't recommend using BNB 4-bit (load_in_4bit) here. Instead, use AWQ, as detailed here: https://huggingface.co/scb10x/llama-3-typhoon-v1.5x-70b-instruct-awq.\n\nmessages = [...] # add message here\n\ninput_ids = tokenizer.apply_chat_template(\n    messages,\n    add_generation_prompt=True,\n    return_tensors=\"pt\"\n).to(model.device)\n\nterminators = [\n    tokenizer.eos_token_id,\n    tokenizer.convert_tokens_to_ids(\"<|eot_id|>\")\n]\n\noutputs = model.generate(\n    input_ids,\n    max_new_tokens=512,\n    eos_token_id=terminators,\n    do_sample=True,\n    temperature=0.4,\n    top_p=0.95,\n)\nresponse = outputs[0][input_ids.shape[-1]:]\nprint(tokenizer.decode(response, skip_special_tokens=True))\n```\n\n## **Chat Template**\n\nWe use the Llama 3 chat template.\n\n```python\n{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>\\n\\n'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{% if add_generation_prompt %}{{ '<|start_header_id|>assistant<|end_header_id|>\\n\\n' }}{% endif %}\n```\n\n## **Intended Uses & Limitations**\n\nThis model is experimental and might not be fully evaluated for all use cases. Developers should assess risks in the context of their specific applications.\n\n## **Follow us**\n\n[**https://twitter.com/opentyphoon**](https://twitter.com/opentyphoon)\n\n## **Support**\n\n[**https://discord.gg/CqyBscMFpg**](https://discord.gg/CqyBscMFpg)\n\n## **SCB 10X Typhoon Team**\n\n- Kunat Pipatanakul, Potsawee Manakul, Sittipong Sripaisarnmongkol, Natapong Nitarach, Pathomporn Chokchainant, Kasima Tharnpipitchai\n- If you find Typhoon-1.5X useful for your work, please cite it using:\n\n```\n@article{pipatanakul2023typhoon,\n    title={Typhoon: Thai Large Language Models}, \n    author={Kunat Pipatanakul and Phatrasek Jirabovonvisut and Potsawee Manakul and Sittipong Sripaisarnmongkol and Ruangsak Patomwong and Pathomporn Chokchainant and Kasima Tharnpipitchai},\n    year={2023},\n    journal={arXiv preprint arXiv:2312.13951},\n    url={https://arxiv.org/abs/2312.13951}\n}\n```\n\n## **Contact Us**\n\n- General & Collaboration: [**kasima@scb10x.com**](mailto:kasima@scb10x.com), [**pathomporn@scb10x.com**](mailto:pathomporn@scb10x.com)\n- Technical: [**kunat@scb10x.com**](mailto:kunat@scb10x.com)\n\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "arxiv:2212.04089",
    "arxiv:2306.05685",
    "arxiv:2311.07911",
    "arxiv:2312.13951",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 0,
  "downloads": 1761,
  "gated": false,
  "private": false,
  "last_modified": "2024-08-03T08:16:53.000Z",
  "created_at": "2024-08-02T08:54:11.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "66ac9eb359f04cc05410246a",
  "id": "RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf",
  "modelId": "RichardErkhov/scb10x_-_llama-3-typhoon-v1.5x-70b-instruct-gguf",
  "sha": "8e58d746262e621afd8538c16b53eef6f827733b",
  "createdAt": "2024-08-02T08:54:11.000Z",
  "lastModified": "2024-08-03T08:16:53.000Z",
  "author": "RichardErkhov",
  "downloads": 1761,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 33
}