=4.39.2\nflash_attn>=2.5.6\n```\n## Usage \n\n### Sentence Transformers\n\n```python\nfrom sentence_transformers import SentenceTransformer\n\nmodel = SentenceTransformer(\"Alibaba-NLP/gte-Qwen2-1.5B-instruct\", trust_remote_code=True)\n# In case you want to reduce the maximum length:\nmodel.max_seq_length = 8192\n\nqueries = [\n \"how much protein should a female eat\",\n \"summit define\",\n]\ndocuments = [\n \"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.\",\n \"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments.\",\n]\n\nquery_embeddings = model.encode(queries, prompt_name=\"query\")\ndocument_embeddings = model.encode(documents)\n\nscores = (query_embeddings @ document_embeddings.T) * 100\nprint(scores.tolist())\n```\n\nObserve the [config_sentence_transformers.json](config_sentence_transformers.json) to see all pre-built prompt names. Otherwise, you can use `model.encode(queries, prompt=\"Instruct: ...\\nQuery: \"` to use a custom prompt of your choice.\n\n### Transformers\n\n```python\nimport torch\nimport torch.nn.functional as F\n\nfrom torch import Tensor\nfrom transformers import AutoTokenizer, AutoModel\n\n\ndef last_token_pool(last_hidden_states: Tensor,\n attention_mask: Tensor) -> Tensor:\n left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0])\n if left_padding:\n return last_hidden_states[:, -1]\n else:\n sequence_lengths = attention_mask.sum(dim=1) - 1\n batch_size = last_hidden_states.shape[0]\n return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths]\n\n\ndef get_detailed_instruct(task_description: str, query: str) -> str:\n return f'Instruct: {task_description}\\nQuery: {query}'\n\n\n# Each query must come with a one-sentence instruction that describes the task\ntask = 'Given a web search query, retrieve relevant passages that answer the query'\nqueries = [\n get_detailed_instruct(task, 'how much protein should a female eat'),\n get_detailed_instruct(task, 'summit define')\n]\n# No need to add instruction for retrieval documents\ndocuments = [\n \"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.\",\n \"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments.\"\n]\ninput_texts = queries + documents\n\ntokenizer = AutoTokenizer.from_pretrained('Alibaba-NLP/gte-Qwen2-1.5B-instruct', trust_remote_code=True)\nmodel = AutoModel.from_pretrained('Alibaba-NLP/gte-Qwen2-1.5B-instruct', trust_remote_code=True)\n\nmax_length = 8192\n\n# Tokenize the input texts\nbatch_dict = tokenizer(input_texts, max_length=max_length, padding=True, truncation=True, return_tensors='pt')\noutputs = model(**batch_dict)\nembeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask'])\n\n# normalize embeddings\nembeddings = F.normalize(embeddings, p=2, dim=1)\nscores = (embeddings[:2] @ embeddings[2:].T) * 100\nprint(scores.tolist())\n```\n\n## Evaluation\n\n### MTEB & C-MTEB\n\nYou can use the [scripts/eval_mteb.py](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct/blob/main/scripts/eval_mteb.py) to reproduce the following result of **gte-Qwen2-1.5B-instruct** on MTEB(English)/C-MTEB(Chinese):\n\n| Model Name | MTEB(56) | C-MTEB(35) | MTEB-fr(26) | MTEB-pl(26) | \n|:----:|:---------:|:----------:|:----------:|:----------:|\n| [bge-base-en-1.5](https://huggingface.co/BAAI/bge-base-en-v1.5) | 64.23 | - | - | - |\n| [bge-large-en-1.5](https://huggingface.co/BAAI/bge-large-en-v1.5) | 63.55 | - | - | - |\n| [gte-large-en-v1.5](https://huggingface.co/Alibaba-NLP/gte-large-en-v1.5) | 65.39 | - | - | - |\n| [gte-base-en-v1.5](https://huggingface.co/Alibaba-NLP/gte-large-en-v1.5) | 64.11 | - | - | - |\n| [mxbai-embed-large-v1](https://huggingface.co/mixedbread-ai/mxbai-embed-large-v1) | 64.68 | - | - | - |\n| [acge_text_embedding](https://huggingface.co/aspire/acge_text_embedding) | - | 69.07 | - | - |\n| [stella-mrl-large-zh-v3.5-1792d](https://huggingface.co/infgrad/stella-mrl-large-zh-v3.5-1792d) | - | 68.55 | - | - |\n| [gte-large-zh](https://huggingface.co/thenlper/gte-large-zh) | - | 66.72 | - | - |\n| [multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base) | 59.45 | 56.21 | - | - |\n| [multilingual-e5-large](https://huggingface.co/intfloat/multilingual-e5-large) | 61.50 | 58.81 | - | - |\n| [e5-mistral-7b-instruct](https://huggingface.co/intfloat/e5-mistral-7b-instruct) | 66.63 | 60.81 | - | - |\n| [gte-Qwen1.5-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen1.5-7B-instruct) | 67.34 | 69.52 | - | - |\n| [NV-Embed-v1](https://huggingface.co/nvidia/NV-Embed-v1) | 69.32 | - | - | - |\n| [**gte-Qwen2-7B-instruct**](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct) | **70.24** | **72.05** | **68.25** | **67.86** |\n| [**gte-Qwen2-1.5B-instruct**](https://huggingface.co/Alibaba-NLP/gte-Qwen2-1.5B-instruct) | **67.16** | **67.65** | **66.60** | **64.04** |\n\n### GTE Models\n\nThe gte series models have consistently released two types of models: encoder-only models (based on the BERT architecture) and decode-only models (based on the LLM architecture). \n\n| Models | Language | Max Sequence Length | Dimension | Model Size (Memory Usage, fp32) |\n|:-------------------------------------------------------------------------------------:|:--------:|:-----: |:---------:|:-------------------------------:|\n| [GTE-large-zh](https://huggingface.co/thenlper/gte-large-zh) | Chinese | 512 | 1024 | 1.25GB |\n| [GTE-base-zh](https://huggingface.co/thenlper/gte-base-zh) | Chinese | 512 | 512 | 0.41GB |\n| [GTE-small-zh](https://huggingface.co/thenlper/gte-small-zh) | Chinese | 512 | 512 | 0.12GB |\n| [GTE-large](https://huggingface.co/thenlper/gte-large) | English | 512 | 1024 | 1.25GB |\n| [GTE-base](https://huggingface.co/thenlper/gte-base) | English | 512 | 512 | 0.21GB |\n| [GTE-small](https://huggingface.co/thenlper/gte-small) | English | 512 | 384 | 0.10GB |\n| [GTE-large-en-v1.5](https://huggingface.co/Alibaba-NLP/gte-large-en-v1.5) | English | 8192 | 1024 | 1.74GB |\n| [GTE-base-en-v1.5](https://huggingface.co/Alibaba-NLP/gte-base-en-v1.5) | English | 8192 | 768 | 0.51GB |\n| [GTE-Qwen1.5-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen1.5-7B-instruct) | Multilingual | 32000 | 4096 | 26.45GB |\n| [GTE-Qwen2-7B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen2-7B-instruct) | Multilingual | 32000 | 3584 | 26.45GB |\n| [GTE-Qwen2-1.5B-instruct](https://huggingface.co/Alibaba-NLP/gte-Qwen2-1.5B-instruct) | Multilingual | 32000 | 1536 | 6.62GB |\n\n## Citation\n\nIf you find our paper or models helpful, please consider cite:\n\n```\n@article{li2023towards,\n title={Towards general text embeddings with multi-stage contrastive learning},\n author={Li, Zehan and Zhang, Xin and Zhang, Yanzhao and Long, Dingkun and Xie, Pengjun and Zhang, Meishan},\n journal={arXiv preprint arXiv:2308.03281},\n year={2023}\n}\n```\n\n\n"},"matched_task":{"kind":"list like","value":["SUMMARIZATION"],"string":"[\n \"SUMMARIZATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["BIOSSES","SCIFACT"],"string":"[\n \"BIOSSES\",\n \"SCIFACT\"\n]","isConversational":false}}},{"rowIdx":187,"cells":{"id":{"kind":"string","value":"TheBloke/meditron-70B-GPTQ"},"author":{"kind":"string","value":"TheBloke"},"task_category":{"kind":"string","value":"text-generation"},"tags":{"kind":"list like","value":["transformers","safetensors","llama","text-generation","medical","health","llama2","en","dataset:bigbio/med_qa","dataset:medmcqa","dataset:bigbio/pubmed_qa","dataset:epfl-llm/guidelines","arxiv:2311.16079","base_model:epfl-llm/meditron-70b","base_model:quantized:epfl-llm/meditron-70b","license:llama2","autotrain_compatible","text-generation-inference","4-bit","gptq","region:us"],"string":"[\n \"transformers\",\n \"safetensors\",\n \"llama\",\n \"text-generation\",\n \"medical\",\n \"health\",\n \"llama2\",\n \"en\",\n \"dataset:bigbio/med_qa\",\n \"dataset:medmcqa\",\n \"dataset:bigbio/pubmed_qa\",\n \"dataset:epfl-llm/guidelines\",\n \"arxiv:2311.16079\",\n \"base_model:epfl-llm/meditron-70b\",\n \"base_model:quantized:epfl-llm/meditron-70b\",\n \"license:llama2\",\n \"autotrain_compatible\",\n \"text-generation-inference\",\n \"4-bit\",\n \"gptq\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2023-11-30T17:10:33","string":"2023-11-30T17:10:33"},"last_modified":{"kind":"timestamp","value":"2023-11-30T21:24:40","string":"2023-11-30T21:24:40"},"downloads":{"kind":"number","value":2391,"string":"2,391"},"likes":{"kind":"number","value":4,"string":"4"},"README":{"kind":"string","value":"---\nbase_model: epfl-llm/meditron-70b\ndatasets:\n- bigbio/med_qa\n- medmcqa\n- bigbio/pubmed_qa\n- epfl-llm/guidelines\nlanguage:\n- en\nlicense: llama2\nmetrics:\n- accuracy\n- perplexity\nmodel_name: Meditron 70B\npipeline_tag: text-generation\ntags:\n- medical\n- health\n- llama2\ninference: false\nmodel_creator: EPFL LLM Team\nmodel_type: llama\nprompt_template: '<|im_start|>system\n\n {system_message}<|im_end|>\n\n <|im_start|>user\n\n {prompt}<|im_end|>\n\n <|im_start|>assistant\n\n '\nquantized_by: TheBloke\n---\n<!-- markdownlint-disable MD041 -->\n\n<!-- header start -->\n<!-- 200823 -->\n<div style=\"width: auto; margin-left: auto; margin-right: auto\">\n<img src=\"https://i.imgur.com/EBdldam.jpg\" alt=\"TheBlokeAI\" style=\"width: 100%; min-width: 400px; display: block; margin: auto;\">\n</div>\n<div style=\"display: flex; justify-content: space-between; width: 100%;\">\n <div style=\"display: flex; flex-direction: column; align-items: flex-start;\">\n <p style=\"margin-top: 0.5em; margin-bottom: 0em;\"><a href=\"https://discord.gg/theblokeai\">Chat & support: TheBloke's Discord server</a></p>\n </div>\n <div style=\"display: flex; flex-direction: column; align-items: flex-end;\">\n <p style=\"margin-top: 0.5em; margin-bottom: 0em;\"><a href=\"https://www.patreon.com/TheBlokeAI\">Want to contribute? TheBloke's Patreon page</a></p>\n </div>\n</div>\n<div style=\"text-align:center; margin-top: 0em; margin-bottom: 0em\"><p style=\"margin-top: 0.25em; margin-bottom: 0em;\">TheBloke's LLM work is generously supported by a grant from <a href=\"https://a16z.com\">andreessen horowitz (a16z)</a></p></div>\n<hr style=\"margin-top: 1.0em; margin-bottom: 1.0em;\">\n<!-- header end -->\n\n# Meditron 70B - GPTQ\n- Model creator: [EPFL LLM Team](https://huggingface.co/epfl-llm)\n- Original model: [Meditron 70B](https://huggingface.co/epfl-llm/meditron-70b)\n\n<!-- description start -->\n# Description\n\nThis repo contains GPTQ model files for [EPFL LLM Team's Meditron 70B](https://huggingface.co/epfl-llm/meditron-70b).\n\nMultiple GPTQ parameter permutations are provided; see Provided Files below for details of the options provided, their parameters, and the software used to create them.\n\nThese files were quantised using hardware kindly provided by [Massed Compute](https://massedcompute.com/).\n\n<!-- description end -->\n<!-- repositories-available start -->\n## Repositories available\n\n* [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/meditron-70B-AWQ)\n* [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/meditron-70B-GPTQ)\n* [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/meditron-70B-GGUF)\n* [EPFL LLM Team's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/epfl-llm/meditron-70b)\n<!-- repositories-available end -->\n\n<!-- prompt-template start -->\n## Prompt template: ChatML\n\n```\n<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n\n```\n\n<!-- prompt-template end -->\n\n\n\n<!-- README_GPTQ.md-compatible clients start -->\n## Known compatible clients / servers\n\nThese GPTQ models are known to work in the following inference servers/webuis.\n\n- [text-generation-webui](https://github.com/oobabooga/text-generation-webui)\n- [KoboldAI United](https://github.com/henk717/koboldai)\n- [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui)\n- [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference)\n\nThis may not be a complete list; if you know of others, please let me know!\n<!-- README_GPTQ.md-compatible clients end -->\n\n<!-- README_GPTQ.md-provided-files start -->\n## Provided files, and GPTQ parameters\n\nMultiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements.\n\nEach separate quant is in a different branch. See below for instructions on fetching from different branches.\n\nMost GPTQ files are made with AutoGPTQ. Mistral models are currently made with Transformers.\n\n<details>\n <summary>Explanation of GPTQ parameters</summary>\n\n- Bits: The bit size of the quantised model.\n- GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. \"None\" is the lowest possible value.\n- Act Order: True or False. Also known as `desc_act`. True results in better quantisation accuracy. Some GPTQ clients have had issues with models that use Act Order plus Group Size, but this is generally resolved now.\n- Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy.\n- GPTQ dataset: The calibration dataset used during quantisation. Using a dataset more appropriate to the model's training can improve quantisation accuracy. Note that the GPTQ calibration dataset is not the same as the dataset used to train the model - please refer to the original model repo for details of the training dataset(s).\n- Sequence Length: The length of the dataset sequences used for quantisation. Ideally this is the same as the model sequence length. For some very long sequence models (16+K), a lower sequence length may have to be used. Note that a lower sequence length does not limit the sequence length of the quantised model. It only impacts the quantisation accuracy on longer inference sequences.\n- ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama and Mistral models in 4-bit.\n\n</details>\n\n| Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc |\n| ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- |\n| [main](https://huggingface.co/TheBloke/meditron-70B-GPTQ/tree/main) | 4 | None | Yes | 0.1 | [Medical Medaow WikiDoc](https://huggingface.co/datasets/medalpaca/medical_meadow_wikidoc/viewer/) | 4096 | 35.33 GB | Yes | 4-bit, with Act Order. No group size, to lower VRAM requirements. | \n| [gptq-4bit-128g-actorder_True](https://huggingface.co/TheBloke/meditron-70B-GPTQ/tree/gptq-4bit-128g-actorder_True) | 4 | 128 | Yes | 0.1 | [Medical Medaow WikiDoc](https://huggingface.co/datasets/medalpaca/medical_meadow_wikidoc/viewer/) | 4096 | 36.65 GB | Yes | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. | \n| [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/meditron-70B-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.1 | [Medical Medaow WikiDoc](https://huggingface.co/datasets/medalpaca/medical_meadow_wikidoc/viewer/) | 4096 | 40.66 GB | Yes | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. | \n| [gptq-3bit--1g-actorder_True](https://huggingface.co/TheBloke/meditron-70B-GPTQ/tree/gptq-3bit--1g-actorder_True) | 3 | None | Yes | 0.1 | [Medical Medaow WikiDoc](https://huggingface.co/datasets/medalpaca/medical_meadow_wikidoc/viewer/) | 4096 | 26.77 GB | No | 3-bit, with Act Order and no group size. Lowest possible VRAM requirements. May be lower quality than 3-bit 128g. | \n| [gptq-3bit-128g-actorder_True](https://huggingface.co/TheBloke/meditron-70B-GPTQ/tree/gptq-3bit-128g-actorder_True) | 3 | 128 | Yes | 0.1 | [Medical Medaow WikiDoc](https://huggingface.co/datasets/medalpaca/medical_meadow_wikidoc/viewer/) | 4096 | 28.03 GB | No | 3-bit, with group size 128g and act-order. Higher quality than 128g-False. | \n| [gptq-3bit-32g-actorder_True](https://huggingface.co/TheBloke/meditron-70B-GPTQ/tree/gptq-3bit-32g-actorder_True) | 3 | 32 | Yes | 0.1 | [Medical Medaow WikiDoc](https://huggingface.co/datasets/medalpaca/medical_meadow_wikidoc/viewer/) | 4096 | 31.84 GB | No | 3-bit, with group size 64g and act-order. Highest quality 3-bit option. |\n\n<!-- README_GPTQ.md-provided-files end -->\n\n<!-- README_GPTQ.md-download-from-branches start -->\n## How to download, including from branches\n\n### In text-generation-webui\n\nTo download from the `main` branch, enter `TheBloke/meditron-70B-GPTQ` in the \"Download model\" box.\n\nTo download from another branch, add `:branchname` to the end of the download name, eg `TheBloke/meditron-70B-GPTQ:gptq-4bit-128g-actorder_True`\n\n### From the command line\n\nI recommend using the `huggingface-hub` Python library:\n\n```shell\npip3 install huggingface-hub\n```\n\nTo download the `main` branch to a folder called `meditron-70B-GPTQ`:\n\n```shell\nmkdir meditron-70B-GPTQ\nhuggingface-cli download TheBloke/meditron-70B-GPTQ --local-dir meditron-70B-GPTQ --local-dir-use-symlinks False\n```\n\nTo download from a different branch, add the `--revision` parameter:\n\n```shell\nmkdir meditron-70B-GPTQ\nhuggingface-cli download TheBloke/meditron-70B-GPTQ --revision gptq-4bit-128g-actorder_True --local-dir meditron-70B-GPTQ --local-dir-use-symlinks False\n```\n\n<details>\n <summary>More advanced huggingface-cli download usage</summary>\n\nIf you remove the `--local-dir-use-symlinks False` parameter, the files will instead be stored in the central Hugging Face cache directory (default location on Linux is: `~/.cache/huggingface`), and symlinks will be added to the specified `--local-dir`, pointing to their real location in the cache. This allows for interrupted downloads to be resumed, and allows you to quickly clone the repo to multiple places on disk without triggering a download again. The downside, and the reason why I don't list that as the default option, is that the files are then hidden away in a cache folder and it's harder to know where your disk space is being used, and to clear it up if/when you want to remove a download model.\n\nThe cache location can be changed with the `HF_HOME` environment variable, and/or the `--cache-dir` parameter to `huggingface-cli`.\n\nFor more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli).\n\nTo accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`:\n\n```shell\npip3 install hf_transfer\n```\n\nAnd set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`:\n\n```shell\nmkdir meditron-70B-GPTQ\nHF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/meditron-70B-GPTQ --local-dir meditron-70B-GPTQ --local-dir-use-symlinks False\n```\n\nWindows Command Line users: You can set the environment variable by running `set HF_HUB_ENABLE_HF_TRANSFER=1` before the download command.\n</details>\n\n### With `git` (**not** recommended)\n\nTo clone a specific branch with `git`, use a command like this:\n\n```shell\ngit clone --single-branch --branch gptq-4bit-128g-actorder_True https://huggingface.co/TheBloke/meditron-70B-GPTQ\n```\n\nNote that using Git with HF repos is strongly discouraged. It will be much slower than using `huggingface-hub`, and will use twice as much disk space as it has to store the model files twice (it stores every byte both in the intended target folder, and again in the `.git` folder as a blob.)\n\n<!-- README_GPTQ.md-download-from-branches end -->\n<!-- README_GPTQ.md-text-generation-webui start -->\n## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui)\n\nPlease make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui).\n\nIt is strongly recommended to use the text-generation-webui one-click-installers unless you're sure you know how to make a manual install.\n\n1. Click the **Model tab**.\n2. Under **Download custom model or LoRA**, enter `TheBloke/meditron-70B-GPTQ`.\n\n - To download from a specific branch, enter for example `TheBloke/meditron-70B-GPTQ:gptq-4bit-128g-actorder_True`\n - see Provided Files above for the list of branches for each option.\n\n3. Click **Download**.\n4. The model will start downloading. Once it's finished it will say \"Done\".\n5. In the top left, click the refresh icon next to **Model**.\n6. In the **Model** dropdown, choose the model you just downloaded: `meditron-70B-GPTQ`\n7. The model will automatically load, and is now ready for use!\n8. If you want any custom settings, set them and then click **Save settings for this model** followed by **Reload the Model** in the top right.\n\n - Note that you do not need to and should not set manual GPTQ parameters any more. These are set automatically from the file `quantize_config.json`.\n\n9. Once you're ready, click the **Text Generation** tab and enter a prompt to get started!\n\n<!-- README_GPTQ.md-text-generation-webui end -->\n\n<!-- README_GPTQ.md-use-from-tgi start -->\n## Serving this model from Text Generation Inference (TGI)\n\nIt's recommended to use TGI version 1.1.0 or later. The official Docker container is: `ghcr.io/huggingface/text-generation-inference:1.1.0`\n\nExample Docker parameters:\n\n```shell\n--model-id TheBloke/meditron-70B-GPTQ --port 3000 --quantize gptq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096\n```\n\nExample Python code for interfacing with TGI (requires huggingface-hub 0.17.0 or later):\n\n```shell\npip3 install huggingface-hub\n```\n\n```python\nfrom huggingface_hub import InferenceClient\n\nendpoint_url = \"https://your-endpoint-url-here\"\n\nprompt = \"Tell me about AI\"\nprompt_template=f'''<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n'''\n\nclient = InferenceClient(endpoint_url)\nresponse = client.text_generation(prompt,\n max_new_tokens=128,\n do_sample=True,\n temperature=0.7,\n top_p=0.95,\n top_k=40,\n repetition_penalty=1.1)\n\nprint(f\"Model output: {response}\")\n```\n<!-- README_GPTQ.md-use-from-tgi end -->\n<!-- README_GPTQ.md-use-from-python start -->\n## Python code example: inference from this GPTQ model\n\n### Install the necessary packages\n\nRequires: Transformers 4.33.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later.\n\n```shell\npip3 install --upgrade transformers optimum\n# If using PyTorch 2.1 + CUDA 12.x:\npip3 install --upgrade auto-gptq\n# or, if using PyTorch 2.1 + CUDA 11.x:\npip3 install --upgrade auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/\n```\n\nIf you are using PyTorch 2.0, you will need to install AutoGPTQ from source. Likewise if you have problems with the pre-built wheels, you should try building from source:\n\n```shell\npip3 uninstall -y auto-gptq\ngit clone https://github.com/PanQiWei/AutoGPTQ\ncd AutoGPTQ\ngit checkout v0.5.1\npip3 install .\n```\n\n### Example Python code\n\n```python\nfrom transformers import AutoModelForCausalLM, AutoTokenizer, pipeline\n\nmodel_name_or_path = \"TheBloke/meditron-70B-GPTQ\"\n# To use a different branch, change revision\n# For example: revision=\"gptq-4bit-128g-actorder_True\"\nmodel = AutoModelForCausalLM.from_pretrained(model_name_or_path,\n device_map=\"auto\",\n trust_remote_code=False,\n revision=\"main\")\n\ntokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)\n\nprompt = \"Tell me about AI\"\nprompt_template=f'''<|im_start|>system\n{system_message}<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n'''\n\nprint(\"\\n\\n*** Generate:\")\n\ninput_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()\noutput = model.generate(inputs=input_ids, temperature=0.7, do_sample=True, top_p=0.95, top_k=40, max_new_tokens=512)\nprint(tokenizer.decode(output[0]))\n\n# Inference can also be done using transformers' pipeline\n\nprint(\"*** Pipeline:\")\npipe = pipeline(\n \"text-generation\",\n model=model,\n tokenizer=tokenizer,\n max_new_tokens=512,\n do_sample=True,\n temperature=0.7,\n top_p=0.95,\n top_k=40,\n repetition_penalty=1.1\n)\n\nprint(pipe(prompt_template)[0]['generated_text'])\n```\n<!-- README_GPTQ.md-use-from-python end -->\n\n<!-- README_GPTQ.md-compatibility start -->\n## Compatibility\n\nThe files provided are tested to work with Transformers. For non-Mistral models, AutoGPTQ can also be used directly.\n\n[ExLlama](https://github.com/turboderp/exllama) is compatible with Llama and Mistral models in 4-bit. Please see the Provided Files table above for per-file compatibility.\n\nFor a list of clients/servers, please see \"Known compatible clients / servers\", above.\n<!-- README_GPTQ.md-compatibility end -->\n\n<!-- footer start -->\n<!-- 200823 -->\n## Discord\n\nFor further support, and discussions on these models and AI in general, join us at:\n\n[TheBloke AI's Discord server](https://discord.gg/theblokeai)\n\n## Thanks, and how to contribute\n\nThanks to the [chirper.ai](https://chirper.ai) team!\n\nThanks to Clay from [gpus.llm-utils.org](llm-utils)!\n\nI've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.\n\nIf you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.\n\nDonaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.\n\n* Patreon: https://patreon.com/TheBlokeAI\n* Ko-Fi: https://ko-fi.com/TheBlokeAI\n\n**Special thanks to**: Aemon Algiz.\n\n**Patreon special mentions**: Brandon Frisco, LangChain4j, Spiking Neurons AB, transmissions 11, Joseph William Delisle, Nitin Borwankar, Willem Michiel, Michael Dempsey, vamX, Jeffrey Morgan, zynix, jjj, Omer Bin Jawed, Sean Connelly, jinyuan sun, Jeromy Smith, Shadi, Pawan Osman, Chadd, Elijah Stavena, Illia Dulskyi, Sebastain Graf, Stephen Murray, terasurfer, Edmond Seymore, Celu Ramasamy, Mandus, Alex, biorpg, Ajan Kanaga, Clay Pascal, Raven Klaugh, 阿明, K, ya boyyy, usrbinkat, Alicia Loh, John Villwock, ReadyPlayerEmma, Chris Smitley, Cap'n Zoog, fincy, GodLy, S_X, sidney chen, Cory Kujawski, OG, Mano Prime, AzureBlack, Pieter, Kalila, Spencer Kim, Tom X Nguyen, Stanislav Ovsiannikov, Michael Levine, Andrey, Trailburnt, Vadim, Enrico Ros, Talal Aujan, Brandon Phillips, Jack West, Eugene Pentland, Michael Davis, Will Dee, webtim, Jonathan Leane, Alps Aficionado, Rooh Singh, Tiffany J. Kim, theTransient, Luke @flexchar, Elle, Caitlyn Gatomon, Ari Malik, subjectnull, Johann-Peter Hartmann, Trenton Dambrowitz, Imad Khwaja, Asp the Wyvern, Emad Mostaque, Rainer Wilmers, Alexandros Triantafyllidis, Nicholas, Pedro Madruga, SuperWojo, Harry Royden McLaughlin, James Bentley, Olakabola, David Ziegler, Ai Maven, Jeff Scroggin, Nikolai Manek, Deo Leter, Matthew Berman, Fen Risland, Ken Nordquist, Manuel Alberto Morcote, Luke Pendergrass, TL, Fred von Graf, Randy H, Dan Guido, NimbleBox.ai, Vitor Caleffi, Gabriel Tamborski, knownsqashed, Lone Striker, Erik Bjäreholt, John Detwiler, Leonard Tan, Iucharbius\n\n\nThank you to all my generous patrons and donaters!\n\nAnd thank you again to a16z for their generous grant.\n\n<!-- footer end -->\n\n# Original model card: EPFL LLM Team's Meditron 70B\n\n\n<img width=50% src=\"meditron_LOGO.png\" alt=\"Alt text\" title=\"Meditron-logo\">\n\n# Model Card for Meditron-70B-v1.0\n\nMeditron is a suite of open-source medical Large Language Models (LLMs).\nMeditron-70B is a 70 billion parameters model adapted to the medical domain from Llama-2-70B through continued pretraining on a comprehensively curated medical corpus, including selected PubMed articles, abstracts, a [new dataset](https://huggingface.co/datasets/epfl-llm/guidelines) of internationally-recognized medical guidelines, and general domain data from [RedPajama-v1](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T).\nMeditron-70B, finetuned on relevant training data, outperforms Llama-2-70B, GPT-3.5 (`text-davinci-003`, 8-shot), and Flan-PaLM on multiple medical reasoning tasks.\n\n\n<!--# Table of Contents\n\n [Model Card for Meditron 70B](#model-card-for--meditron-70b-v1.0)\n- [Table of Contents](#table-of-contents)\n- [Model Details](#model-details)\n - [Model Description](#model-description)\n- [Uses](#uses)\n - [Downstream Use](#downstream-use)\n - [Out-of-Scope Use](#out-of-scope-use)\n- [Bias, Risks, and Limitations](#bias-risks-and-limitations)\n - [Recommendations](#recommendations)\n- [Training Details](#training-details)\n - [Training Data](#training-data)\n - [Training Procedure](#training-procedure)\n - [Preprocessing](#preprocessing)\n- [Evaluation](#evaluation)\n - [Testing Data & Metrics](#testing-data-&-metrics)\n - [Testing Data](#testing-data)\n - [Metrics](#metrics)\n - [Results](#results)\n- [Environmental Impact](#environmental-impact)\n- [Citation](#citation)-->\n\n<details open>\n <summary><strong>Advisory Notice</strong></summary>\n\n <blockquote style=\"padding: 10px; margin: 0 0 10px; border-left: 5px solid #ddd;\">\n While Meditron is designed to encode medical knowledge from sources of high-quality evidence, it is not yet adapted to deliver this knowledge appropriately, safely, or within professional actionable constraints. \n We recommend against deploying Meditron in medical applications without extensive use-case alignment, as well as additional testing, specifically including randomized controlled trials in real-world practice settings.\n </blockquote>\n</details>\n\n## Model Details\n\n- **Developed by:** [EPFL LLM Team](https://huggingface.co/epfl-llm)\n- **Model type:** Causal decoder-only transformer language model\n- **Language(s):** English (mainly)\n- **Model License:** [LLAMA 2 COMMUNITY LICENSE AGREEMENT](https://huggingface.co/meta-llama/Llama-2-70b/raw/main/LICENSE.txt)\n- **Code License:** [APACHE 2.0 LICENSE](LICENSE)\n- **Continue-pretrained from model:** [Llama-2-70B](https://huggingface.co/meta-llama/Llama-2-70b)\n- **Context length:** 4K tokens\n- **Input:** Text-only data\n- **Output:** Model generates text only\n- **Status:** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we enhance model's performance.\n- **Knowledge Cutoff:** August 2023\n\n\n### Model Sources\n\n- **Repository:** [epflLLM/meditron](https://github.com/epfLLM/meditron)\n- **Trainer:** [epflLLM/Megatron-LLM](https://github.com/epfLLM/Megatron-LLM)\n- **Paper:** *[MediTron-70B: Scaling Medical Pretraining for Large Language Models](https://arxiv.org/abs/2311.16079)*\n\n## Uses\n\nMeditron-70B is being made available for further testing and assessment as an AI assistant to enhance clinical decision-making and enhance access to an LLM for healthcare use. Potential use cases may include but are not limited to:\n- Medical exam question answering\n- Supporting differential diagnosis\n- Disease information (symptoms, cause, treatment) query\n- General health information query\n\n### Direct Use\n\nIt is possible to use this model to generate text, which is useful for experimentation and understanding its capabilities. \nIt should not be used directly for production or work that may impact people.\n\n### Downstream Use\nMeditron-70B is a foundation model that can be finetuned, instruction-tuned, or RLHF-tuned for specific downstream tasks and applications.\nThe main way we have used this model is finetuning for downstream question-answering tasks, but we encourage using this model for additional applications.\n\nSpecific formatting needs to be followed to prompt our finetuned models, including the `<|im_start|>`, `<|im_end|>` tags, and `system`, `question`, `answer` identifiers.\n \n \"\"\"\n <|im_start|>system\n {system_message}<|im_end|>\n <|im_start|>question\n {prompt}<|im_end|>\n <|im_start|>answer \n \"\"\"\n\n**Note 1**: The above formatting is not required for running the base model (this repository)\n\n**Note 2**: the above formatting is just an example of a finetuning template. This format is not a requirement if you use your own formatting option for the finetuning of the model.\n\nTo run proper generation with this base model, we recommend using a high-throughput and memory-efficient inference engine, such as [vLLM](https://github.com/vllm-project/vllm), with a UI that supports chat and text generation, such as [BetterChatGPT](https://github.com/ztjhz/BetterChatGPT)\nTo see more details about model deployment and generation, please see our [documentation](https://github.com/epfLLM/meditron/blob/main/deployment/README.md).\n\n### Out-of-Scope Use\n\nWe do not recommend using this model for natural language generation in a production environment, finetuned or otherwise.\n\n## Truthfulness, Helpfulness, Risk, and Bias\n\n<!-- This section is meant to convey both technical and sociotechnical limitations. -->\n\nWe did an initial assessment of Meditron models' **Truthfulness** against baseline models and consumer-level medical models. \nWe use TruthfulQA (multiple choice) as the main evaluation benchmark.\nWe only focus on the categories that are relevant to the medical domain, including Health, Nutrition, Psychology, and Science.\nFor 7B models, we perform one-shot evaluations for consistent answer generation.\nFor 70B models, the evaluations are under the zero-shot setting.\nBelow, we report the detailed truthfulness performance of each category.\n\n| | | | | | | | |\n| --- | ------ |----- |----- |----- |----- |----- |----- |\n|Category | meditron-70b | llama-2-70b | med42-70b* | meditron-7b | llama-2-7b | PMC-llama-7b |\n|Health | 81.8 | 69.1 | 83.6 | 27.3 | 16.4 | 3.6 |\n|Nutrition | 77.9 | 68.8 | 62.5 | 31.1 | 12.5 | 6.3 |\n|Psychology| 47.4 | 36.8 | 52.6 | 21.1 | 10.5 | 0.0 |\n|Science | 77.8 | 44.4 | 33.3 | 33.3 | 11.1 | 0.0 |\n|Avg | 71.2 | 54.8 | 58.0 | 28.3 | 12.6 | 2.5 |\n| | | | | | | |\n\nFor a more detailed performance analysis, please see our paper. \n\nFor **Helpfulness**, **Risk** and **Bias**, we provide a comprehensive qualitative generation report of Meditron-70B on queries designed by medical experts.\nEach query targets specific aspects of helpfulness (medical accuracy, up-to-date information, etc.), risk (public health, medical ethics, etc.) and bias (gender, age, race, etc.).\nPlease see the detailed generations in our paper. We compare our generations to Llama-2-70B and ChatGPT-3.5 (version Nov, 27, 2023)\n\nSignificant research is still required to fully explore potential bias, fairness, and safety issues with this language model. \n\n\n### Recommendations\n\n**IMPORTANT!**\nUsers (both direct and downstream) should be made aware of the risks, biases, and limitations of the model.\nWhile this model is capable of generating natural language text, we have only begun to explore this capability and its limitations. \nUnderstanding these limitations is especially important in a domain like medicine. \nTherefore, we strongly recommend against using this model in production for natural language generation or for professional purposes related to health and medicine without comprehensive testing for your application. \n\n## Training Details\n\n### Training Data\nMeditron’s domain-adaptive pre-training corpus GAP-Replay combines 48.1B tokens from four corpora:\n- [**Clinical Guidelines**](https://huggingface.co/datasets/epfl-llm/guidelines): a new dataset of 46K internationally-recognized clinical practice guidelines from various healthcare-related sources, including hospitals and international organizations. \n- **Medical Paper Abstracts**: 16.1M abstracts extracted from closed-access PubMed and PubMed Central papers.\n- **Medical Papers**: full-text articles extracted from 5M publicly available PubMed and PubMed Central papers.\n- **Replay Data**: 400M tokens of general domain pretraining data sampled from [RedPajama-v1](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T)\n\n\n<img width=\"60%\" src=\"gap-replay.png\" alt=\"Alt text\" title=\"Meditron-logo\">\n\n\n#### Data Preprocessing\n\nPlease see the detailed preprocessing procedure in our paper. \n\n### Training Procedure \n\nWe used the [Megatron-LLM](https://github.com/epfLLM/Megatron-LLM) distributed training library, a derivative of Nvidia's Megatron LM project, to optimize training efficiency. \nHardware consists of 16 nodes of 8x NVIDIA A100 (80GB) SXM GPUs connected by NVLink and NVSwitch with a single Nvidia ConnectX-6 DX network card and equipped with 2 x AMD EPYC 7543 32-Core Processors and 512 GB of RAM. \nThe nodes are connected via RDMA over Converged Ethernet.\n\nOur three-way parallelism scheme uses:\n - Data Parallelism (DP -- different GPUs process different subsets of the batches) of 2,\n - Pipeline Parallelism (PP -- different GPUs process different layers) of 8,\n - Tensor Parallelism (TP -- different GPUs process different subtensors for matrix multiplication) of 8.\n \n\n#### Training Hyperparameters\n\n| | |\n| --- | ------ |\n| bf16 | true |\n| lr | 1.5e-4 |\n| eps | 1e-5 |\n| betas | \\[0.9, 0.95\\] |\n| clip_grad | 1 |\n| weight decay | 0.1 |\n| DP size | 2 |\n| TP size | 8 |\n| PP size | 8 |\n| seq length | 4096 |\n| lr scheduler | cosine|\n| min lr | 1e-6 |\n| warmup iteration | 2000 |\n| micro batch size | 2 |\n| global batch size | 512 |\n| | |\n\n#### Speeds, Sizes, Times\nThe model was trained in September and October 2023.\n\nThe model architecture is exactly Llama 2, meaning\n\n| | | \n| --- | ------ |\n| Model size | 70B | \n| Hidden dimension | 8192 |\n| Num. attention heads | 64 |\n| Num. layers | 80 |\n| | | | \n\nWe train the 70B model on 48e9 tokens, at a throughput of about 40,200 tokens / second. \nThis amounts to a bfloat16 model flops utilization of roughly 42.3\\%.\n\n## Evaluation\n\n<!-- This section describes the evaluation protocols and provides the results. -->\n\n### Testing Data & Metrics\n\n#### Testing Data\n- [MedQA (USMLE)](https://huggingface.co/datasets/bigbio/med_qa)\n- [MedMCQA](https://huggingface.co/datasets/medmcqa)\n- [PubMedQA](https://huggingface.co/datasets/bigbio/pubmed_qa)\n- [MMLU-Medical](https://huggingface.co/datasets/lukaemon/mmlu)\n- [MedQA-4-Option](https://huggingface.co/datasets/GBaker/MedQA-USMLE-4-options)\n\n#### Metrics\n- Accuracy: suite the evaluation of multiple-choice question-answering tasks.\n\n### Results\nWe finetune meditron-70b and llama-2-70b on each benchmark (pubmedqa, medmcqa, medqa)'s training data individually. \nWe report the finetuned models' performance with self-consistency chain-of-thought as the inference mode.\nFor MMLU-Medical, models finetuned on MedMCQA are used for inference.\nFor MedQA-4-Option, models finetuned on MedQA are used for inference.\nFor a more detailed performance analysis, please see our paper. \n\n| | | | | | |\n| --- | ------ |----- |----- |----- |----- |\n|Dataset| meditron-70b | llama-2-70b | med42-70b* | clinical-camel-70b* |\n|MMLU-Medical | 77.6 | 77.9 | 74.5 | 65.7 |\n|PubMedQA | 81.6 | 80.0 | 61.2 | 67.0 |\n|MedMCQA | 66.0 | 62.6 | 59.2 | 46.7 |\n|MedQA | 64.4 | 61.5 | 59.1 | 50.8 |\n|MedQA-4-Option| 70.2 | 63.8 | 63.9 | 56.8 |\n|Avg | 72.0 | 69.2 | 63.6 | 57.4 |\n| | | | | | |\n\n**Note**: models with * are already instruction-tuned, so we exclude them from further finetuning on any training data.\n\n## Environmental Impact\n\n<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->\n\n- **Hardware Type:** 128 x NVIDIA A100 (80GB) SXM\n- **Total GPU hours:** 42,496\n- **Hardware Provider:** EPFL Research Computing Platform\n- **Compute Region:** Switzerland\n- **Carbon Emitted:** Switzerland has a carbon efficiency of 0.016 kgCO2/kWh (https://www.carbonfootprint.com/docs/2018_8_electricity_factors_august_2018_-_online_sources.pdf). 332 hours of 128 A100s means 42496 hours at a TDP of 400W. Assuming a Power Usage effectiveness of 1.8, total emissions are estimated to be: \n \n (400W / 1000W/kWh / GPU * 0.016 kgCO2/kWh * 332 h * 128 GPU) * 1.8 PUE = 486 kgCO2.\n\n## Citation\n\n**BibTeX:**\nIf you use Meditron or its training data, please cite our work:\n\n```\n@misc{chen2023meditron70b,\n title={MEDITRON-70B: Scaling Medical Pretraining for Large Language Models}, \n author={Zeming Chen and Alejandro Hernández-Cano and Angelika Romanou and Antoine Bonnet and Kyle Matoba and Francesco Salvi and Matteo Pagliardini and Simin Fan and Andreas Köpf and Amirkeivan Mohtashami and Alexandre Sallinen and Alireza Sakhaeirad and Vinitra Swamy and Igor Krawczuk and Deniz Bayazit and Axel Marmet and Syrielle Montariol and Mary-Anne Hartley and Martin Jaggi and Antoine Bosselut},\n year={2023},\n eprint={2311.16079},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n\n@software{epfmedtrn,\n author = {Zeming Chen and Alejandro Hernández Cano and Angelika Romanou and Antoine Bonnet and Kyle Matoba and Francesco Salvi and Matteo Pagliardini and Simin Fan and Andreas Köpf and Amirkeivan Mohtashami and Alexandre Sallinen and Alireza Sakhaeirad and Vinitra Swamy and Igor Krawczuk and Deniz Bayazit and Axel Marmet and Syrielle Montariol and Mary-Anne Hartley and Martin Jaggi and Antoine Bosselut},\n title = {MediTron-70B: Scaling Medical Pretraining for Large Language Models},\n month = November,\n year = 2023,\n url = {https://github.com/epfLLM/meditron}\n}\n```\n"},"matched_task":{"kind":"list like","value":["QUESTION_ANSWERING"],"string":"[\n \"QUESTION_ANSWERING\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["MEDQA","PUBMEDQA"],"string":"[\n \"MEDQA\",\n \"PUBMEDQA\"\n]","isConversational":false}}},{"rowIdx":188,"cells":{"id":{"kind":"string","value":"Alibaba-NLP/gme-Qwen2-VL-7B-Instruct"},"author":{"kind":"string","value":"Alibaba-NLP"},"task_category":{"kind":"string","value":"sentence-similarity"},"tags":{"kind":"list like","value":["sentence-transformers","safetensors","qwen2_vl","image-text-to-text","mteb","transformers","Qwen2-VL","sentence-similarity","vidore","en","zh","arxiv:2412.16855","base_model:Qwen/Qwen2-VL-7B-Instruct","base_model:finetune:Qwen/Qwen2-VL-7B-Instruct","license:apache-2.0","model-index","autotrain_compatible","text-generation-inference","endpoints_compatible","region:us"],"string":"[\n \"sentence-transformers\",\n \"safetensors\",\n \"qwen2_vl\",\n \"image-text-to-text\",\n \"mteb\",\n \"transformers\",\n \"Qwen2-VL\",\n \"sentence-similarity\",\n \"vidore\",\n \"en\",\n \"zh\",\n \"arxiv:2412.16855\",\n \"base_model:Qwen/Qwen2-VL-7B-Instruct\",\n \"base_model:finetune:Qwen/Qwen2-VL-7B-Instruct\",\n \"license:apache-2.0\",\n \"model-index\",\n \"autotrain_compatible\",\n \"text-generation-inference\",\n \"endpoints_compatible\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2024-12-21T04:00:17","string":"2024-12-21T04:00:17"},"last_modified":{"kind":"timestamp","value":"2025-01-21T11:53:23","string":"2025-01-21T11:53:23"},"downloads":{"kind":"number","value":2333,"string":"2,333"},"likes":{"kind":"number","value":26,"string":"26"},"README":{"kind":"string","value":"---\nbase_model:\n- Qwen/Qwen2-VL-7B-Instruct\nlanguage:\n- en\n- zh\nlicense: apache-2.0\ntags:\n- mteb\n- sentence-transformers\n- transformers\n- Qwen2-VL\n- sentence-similarity\n- vidore\nmodel-index:\n- name: gme-Qwen2-VL-7B-Instruct\n results:\n - task:\n type: STS\n dataset:\n name: MTEB AFQMC\n type: C-MTEB/AFQMC\n config: default\n split: validation\n revision: b44c3b011063adb25877c13823db83bb193913c4\n metrics:\n - type: cos_sim_pearson\n value: 64.72351048394194\n - type: cos_sim_spearman\n value: 71.66842612591344\n - type: euclidean_pearson\n value: 70.0342809043895\n - type: euclidean_spearman\n value: 71.66842612323917\n - type: manhattan_pearson\n value: 69.94743870947117\n - type: manhattan_spearman\n value: 71.53159630946965\n - task:\n type: STS\n dataset:\n name: MTEB ATEC\n type: C-MTEB/ATEC\n config: default\n split: test\n revision: 0f319b1142f28d00e055a6770f3f726ae9b7d865\n metrics:\n - type: cos_sim_pearson\n value: 52.38188106868689\n - type: cos_sim_spearman\n value: 55.468235529709766\n - type: euclidean_pearson\n value: 56.974786979175086\n - type: euclidean_spearman\n value: 55.468231026153745\n - type: manhattan_pearson\n value: 56.94467132566259\n - type: manhattan_spearman\n value: 55.39037386224014\n - task:\n type: Classification\n dataset:\n name: MTEB AmazonCounterfactualClassification (en)\n type: mteb/amazon_counterfactual\n config: en\n split: test\n revision: e8379541af4e31359cca9fbcf4b00f2671dba205\n metrics:\n - type: accuracy\n value: 77.61194029850746\n - type: ap\n value: 41.29789064067677\n - type: f1\n value: 71.69633278678522\n - task:\n type: Classification\n dataset:\n name: MTEB AmazonPolarityClassification\n type: mteb/amazon_polarity\n config: default\n split: test\n revision: e2d317d38cd51312af73b3d32a06d1a08b442046\n metrics:\n - type: accuracy\n value: 97.3258\n - type: ap\n value: 95.91845683387056\n - type: f1\n value: 97.32526074864263\n - task:\n type: Classification\n dataset:\n name: MTEB AmazonReviewsClassification (en)\n type: mteb/amazon_reviews_multi\n config: en\n split: test\n revision: 1399c76144fd37290681b995c656ef9b2e06e26d\n metrics:\n - type: accuracy\n value: 64.794\n - type: f1\n value: 63.7329780206882\n - task:\n type: Classification\n dataset:\n name: MTEB AmazonReviewsClassification (zh)\n type: mteb/amazon_reviews_multi\n config: zh\n split: test\n revision: 1399c76144fd37290681b995c656ef9b2e06e26d\n metrics:\n - type: accuracy\n value: 55.099999999999994\n - type: f1\n value: 53.115528412999666\n - task:\n type: Retrieval\n dataset:\n name: MTEB ArguAna\n type: mteb/arguana\n config: default\n split: test\n revision: c22ab2a51041ffd869aaddef7af8d8215647e41a\n metrics:\n - type: map_at_1\n value: 40.541\n - type: map_at_10\n value: 56.315000000000005\n - type: map_at_100\n value: 56.824\n - type: map_at_1000\n value: 56.825\n - type: map_at_3\n value: 51.778\n - type: map_at_5\n value: 54.623\n - type: mrr_at_1\n value: 41.038000000000004\n - type: mrr_at_10\n value: 56.532000000000004\n - type: mrr_at_100\n value: 57.034\n - type: mrr_at_1000\n value: 57.034\n - type: mrr_at_3\n value: 52.015\n - type: mrr_at_5\n value: 54.835\n - type: ndcg_at_1\n value: 40.541\n - type: ndcg_at_10\n value: 64.596\n - type: ndcg_at_100\n value: 66.656\n - type: ndcg_at_1000\n value: 66.666\n - type: ndcg_at_3\n value: 55.415000000000006\n - type: ndcg_at_5\n value: 60.527\n - type: precision_at_1\n value: 40.541\n - type: precision_at_10\n value: 9.083\n - type: precision_at_100\n value: 0.996\n - type: precision_at_1000\n value: 0.1\n - type: precision_at_3\n value: 21.977\n - type: precision_at_5\n value: 15.661\n - type: recall_at_1\n value: 40.541\n - type: recall_at_10\n value: 90.825\n - type: recall_at_100\n value: 99.57300000000001\n - type: recall_at_1000\n value: 99.644\n - type: recall_at_3\n value: 65.932\n - type: recall_at_5\n value: 78.307\n - task:\n type: Clustering\n dataset:\n name: MTEB ArxivClusteringP2P\n type: mteb/arxiv-clustering-p2p\n config: default\n split: test\n revision: a122ad7f3f0291bf49cc6f4d32aa80929df69d5d\n metrics:\n - type: v_measure\n value: 54.96111428218386\n - task:\n type: Clustering\n dataset:\n name: MTEB ArxivClusteringS2S\n type: mteb/arxiv-clustering-s2s\n config: default\n split: test\n revision: f910caf1a6075f7329cdf8c1a6135696f37dbd53\n metrics:\n - type: v_measure\n value: 50.637711388838945\n - task:\n type: Reranking\n dataset:\n name: MTEB AskUbuntuDupQuestions\n type: mteb/askubuntudupquestions-reranking\n config: default\n split: test\n revision: 2000358ca161889fa9c082cb41daa8dcfb161a54\n metrics:\n - type: map\n value: 64.0741897266483\n - type: mrr\n value: 76.11440882909028\n - task:\n type: STS\n dataset:\n name: MTEB BIOSSES\n type: mteb/biosses-sts\n config: default\n split: test\n revision: d3fb88f8f02e40887cd149695127462bbcf29b4a\n metrics:\n - type: cos_sim_pearson\n value: 86.2557839280406\n - type: cos_sim_spearman\n value: 82.58200216886888\n - type: euclidean_pearson\n value: 84.80588838508498\n - type: euclidean_spearman\n value: 82.58200216886888\n - type: manhattan_pearson\n value: 84.53082035185592\n - type: manhattan_spearman\n value: 82.4964580510134\n - task:\n type: STS\n dataset:\n name: MTEB BQ\n type: C-MTEB/BQ\n config: default\n split: test\n revision: e3dda5e115e487b39ec7e618c0c6a29137052a55\n metrics:\n - type: cos_sim_pearson\n value: 76.98420285210636\n - type: cos_sim_spearman\n value: 78.95549489000658\n - type: euclidean_pearson\n value: 79.14591532018991\n - type: euclidean_spearman\n value: 78.95549488953284\n - type: manhattan_pearson\n value: 79.26212116856509\n - type: manhattan_spearman\n value: 79.02104262086006\n - task:\n type: Classification\n dataset:\n name: MTEB Banking77Classification\n type: mteb/banking77\n config: default\n split: test\n revision: 0fd18e25b25c072e09e0d92ab615fda904d66300\n metrics:\n - type: accuracy\n value: 84.76298701298703\n - type: f1\n value: 84.24881789367576\n - task:\n type: Clustering\n dataset:\n name: MTEB BiorxivClusteringP2P\n type: mteb/biorxiv-clustering-p2p\n config: default\n split: test\n revision: 65b79d1d13f80053f67aca9498d9402c2d9f1f40\n metrics:\n - type: v_measure\n value: 46.86757924102047\n - task:\n type: Clustering\n dataset:\n name: MTEB BiorxivClusteringS2S\n type: mteb/biorxiv-clustering-s2s\n config: default\n split: test\n revision: 258694dd0231531bc1fd9de6ceb52a0853c6d908\n metrics:\n - type: v_measure\n value: 43.86043680479362\n - task:\n type: Clustering\n dataset:\n name: MTEB CLSClusteringP2P\n type: C-MTEB/CLSClusteringP2P\n config: default\n split: test\n revision: 4b6227591c6c1a73bc76b1055f3b7f3588e72476\n metrics:\n - type: v_measure\n value: 45.684222588040605\n - task:\n type: Clustering\n dataset:\n name: MTEB CLSClusteringS2S\n type: C-MTEB/CLSClusteringS2S\n config: default\n split: test\n revision: e458b3f5414b62b7f9f83499ac1f5497ae2e869f\n metrics:\n - type: v_measure\n value: 45.45639765303432\n - task:\n type: Reranking\n dataset:\n name: MTEB CMedQAv1\n type: C-MTEB/CMedQAv1-reranking\n config: default\n split: test\n revision: 8d7f1e942507dac42dc58017c1a001c3717da7df\n metrics:\n - type: map\n value: 88.7058672660788\n - type: mrr\n value: 90.5795634920635\n - task:\n type: Reranking\n dataset:\n name: MTEB CMedQAv2\n type: C-MTEB/CMedQAv2-reranking\n config: default\n split: test\n revision: 23d186750531a14a0357ca22cd92d712fd512ea0\n metrics:\n - type: map\n value: 90.50750030424048\n - type: mrr\n value: 92.3970634920635\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackAndroidRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: f46a197baaae43b4f621051089b82a364682dfeb\n metrics:\n - type: map_at_1\n value: 28.848000000000003\n - type: map_at_10\n value: 40.453\n - type: map_at_100\n value: 42.065000000000005\n - type: map_at_1000\n value: 42.176\n - type: map_at_3\n value: 36.697\n - type: map_at_5\n value: 38.855000000000004\n - type: mrr_at_1\n value: 34.764\n - type: mrr_at_10\n value: 45.662000000000006\n - type: mrr_at_100\n value: 46.56\n - type: mrr_at_1000\n value: 46.597\n - type: mrr_at_3\n value: 42.632\n - type: mrr_at_5\n value: 44.249\n - type: ndcg_at_1\n value: 34.764\n - type: ndcg_at_10\n value: 47.033\n - type: ndcg_at_100\n value: 53.089\n - type: ndcg_at_1000\n value: 54.818\n - type: ndcg_at_3\n value: 41.142\n - type: ndcg_at_5\n value: 43.928\n - type: precision_at_1\n value: 34.764\n - type: precision_at_10\n value: 9.027000000000001\n - type: precision_at_100\n value: 1.465\n - type: precision_at_1000\n value: 0.192\n - type: precision_at_3\n value: 19.695\n - type: precision_at_5\n value: 14.535\n - type: recall_at_1\n value: 28.848000000000003\n - type: recall_at_10\n value: 60.849\n - type: recall_at_100\n value: 85.764\n - type: recall_at_1000\n value: 96.098\n - type: recall_at_3\n value: 44.579\n - type: recall_at_5\n value: 51.678999999999995\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackEnglishRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: ad9991cb51e31e31e430383c75ffb2885547b5f0\n metrics:\n - type: map_at_1\n value: 30.731\n - type: map_at_10\n value: 41.859\n - type: map_at_100\n value: 43.13\n - type: map_at_1000\n value: 43.257\n - type: map_at_3\n value: 38.384\n - type: map_at_5\n value: 40.284\n - type: mrr_at_1\n value: 38.471\n - type: mrr_at_10\n value: 47.531\n - type: mrr_at_100\n value: 48.199\n - type: mrr_at_1000\n value: 48.24\n - type: mrr_at_3\n value: 44.989000000000004\n - type: mrr_at_5\n value: 46.403\n - type: ndcg_at_1\n value: 38.471\n - type: ndcg_at_10\n value: 48.022999999999996\n - type: ndcg_at_100\n value: 52.32599999999999\n - type: ndcg_at_1000\n value: 54.26\n - type: ndcg_at_3\n value: 42.986999999999995\n - type: ndcg_at_5\n value: 45.23\n - type: precision_at_1\n value: 38.471\n - type: precision_at_10\n value: 9.248000000000001\n - type: precision_at_100\n value: 1.469\n - type: precision_at_1000\n value: 0.193\n - type: precision_at_3\n value: 20.892\n - type: precision_at_5\n value: 14.892\n - type: recall_at_1\n value: 30.731\n - type: recall_at_10\n value: 59.561\n - type: recall_at_100\n value: 77.637\n - type: recall_at_1000\n value: 89.64999999999999\n - type: recall_at_3\n value: 44.897999999999996\n - type: recall_at_5\n value: 51.181\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackGamingRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 4885aa143210c98657558c04aaf3dc47cfb54340\n metrics:\n - type: map_at_1\n value: 34.949000000000005\n - type: map_at_10\n value: 48.117\n - type: map_at_100\n value: 49.355\n - type: map_at_1000\n value: 49.409\n - type: map_at_3\n value: 44.732\n - type: map_at_5\n value: 46.555\n - type: mrr_at_1\n value: 40.188\n - type: mrr_at_10\n value: 51.452\n - type: mrr_at_100\n value: 52.219\n - type: mrr_at_1000\n value: 52.24100000000001\n - type: mrr_at_3\n value: 48.642\n - type: mrr_at_5\n value: 50.134\n - type: ndcg_at_1\n value: 40.188\n - type: ndcg_at_10\n value: 54.664\n - type: ndcg_at_100\n value: 59.38099999999999\n - type: ndcg_at_1000\n value: 60.363\n - type: ndcg_at_3\n value: 48.684\n - type: ndcg_at_5\n value: 51.406\n - type: precision_at_1\n value: 40.188\n - type: precision_at_10\n value: 9.116\n - type: precision_at_100\n value: 1.248\n - type: precision_at_1000\n value: 0.13699999999999998\n - type: precision_at_3\n value: 22.236\n - type: precision_at_5\n value: 15.310000000000002\n - type: recall_at_1\n value: 34.949000000000005\n - type: recall_at_10\n value: 70.767\n - type: recall_at_100\n value: 90.79\n - type: recall_at_1000\n value: 97.57900000000001\n - type: recall_at_3\n value: 54.723\n - type: recall_at_5\n value: 61.404\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackGisRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 5003b3064772da1887988e05400cf3806fe491f2\n metrics:\n - type: map_at_1\n value: 25.312\n - type: map_at_10\n value: 34.799\n - type: map_at_100\n value: 35.906\n - type: map_at_1000\n value: 35.983\n - type: map_at_3\n value: 31.582\n - type: map_at_5\n value: 33.507999999999996\n - type: mrr_at_1\n value: 27.232\n - type: mrr_at_10\n value: 36.82\n - type: mrr_at_100\n value: 37.733\n - type: mrr_at_1000\n value: 37.791000000000004\n - type: mrr_at_3\n value: 33.804\n - type: mrr_at_5\n value: 35.606\n - type: ndcg_at_1\n value: 27.232\n - type: ndcg_at_10\n value: 40.524\n - type: ndcg_at_100\n value: 45.654\n - type: ndcg_at_1000\n value: 47.557\n - type: ndcg_at_3\n value: 34.312\n - type: ndcg_at_5\n value: 37.553\n - type: precision_at_1\n value: 27.232\n - type: precision_at_10\n value: 6.52\n - type: precision_at_100\n value: 0.9530000000000001\n - type: precision_at_1000\n value: 0.11399999999999999\n - type: precision_at_3\n value: 14.915000000000001\n - type: precision_at_5\n value: 10.847\n - type: recall_at_1\n value: 25.312\n - type: recall_at_10\n value: 56.169000000000004\n - type: recall_at_100\n value: 79.16499999999999\n - type: recall_at_1000\n value: 93.49300000000001\n - type: recall_at_3\n value: 39.5\n - type: recall_at_5\n value: 47.288999999999994\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackMathematicaRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 90fceea13679c63fe563ded68f3b6f06e50061de\n metrics:\n - type: map_at_1\n value: 17.153\n - type: map_at_10\n value: 27.671\n - type: map_at_100\n value: 29.186\n - type: map_at_1000\n value: 29.299999999999997\n - type: map_at_3\n value: 24.490000000000002\n - type: map_at_5\n value: 26.178\n - type: mrr_at_1\n value: 21.144\n - type: mrr_at_10\n value: 32.177\n - type: mrr_at_100\n value: 33.247\n - type: mrr_at_1000\n value: 33.306000000000004\n - type: mrr_at_3\n value: 29.187\n - type: mrr_at_5\n value: 30.817\n - type: ndcg_at_1\n value: 21.144\n - type: ndcg_at_10\n value: 33.981\n - type: ndcg_at_100\n value: 40.549\n - type: ndcg_at_1000\n value: 43.03\n - type: ndcg_at_3\n value: 28.132\n - type: ndcg_at_5\n value: 30.721999999999998\n - type: precision_at_1\n value: 21.144\n - type: precision_at_10\n value: 6.666999999999999\n - type: precision_at_100\n value: 1.147\n - type: precision_at_1000\n value: 0.149\n - type: precision_at_3\n value: 14.302999999999999\n - type: precision_at_5\n value: 10.423\n - type: recall_at_1\n value: 17.153\n - type: recall_at_10\n value: 48.591\n - type: recall_at_100\n value: 76.413\n - type: recall_at_1000\n value: 93.8\n - type: recall_at_3\n value: 32.329\n - type: recall_at_5\n value: 38.958999999999996\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackPhysicsRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 79531abbd1fb92d06c6d6315a0cbbbf5bb247ea4\n metrics:\n - type: map_at_1\n value: 27.909\n - type: map_at_10\n value: 40.168\n - type: map_at_100\n value: 41.524\n - type: map_at_1000\n value: 41.626000000000005\n - type: map_at_3\n value: 36.274\n - type: map_at_5\n value: 38.411\n - type: mrr_at_1\n value: 34.649\n - type: mrr_at_10\n value: 45.613\n - type: mrr_at_100\n value: 46.408\n - type: mrr_at_1000\n value: 46.444\n - type: mrr_at_3\n value: 42.620999999999995\n - type: mrr_at_5\n value: 44.277\n - type: ndcg_at_1\n value: 34.649\n - type: ndcg_at_10\n value: 47.071000000000005\n - type: ndcg_at_100\n value: 52.559999999999995\n - type: ndcg_at_1000\n value: 54.285000000000004\n - type: ndcg_at_3\n value: 40.63\n - type: ndcg_at_5\n value: 43.584\n - type: precision_at_1\n value: 34.649\n - type: precision_at_10\n value: 8.855\n - type: precision_at_100\n value: 1.361\n - type: precision_at_1000\n value: 0.167\n - type: precision_at_3\n value: 19.538\n - type: precision_at_5\n value: 14.187\n - type: recall_at_1\n value: 27.909\n - type: recall_at_10\n value: 62.275000000000006\n - type: recall_at_100\n value: 84.95\n - type: recall_at_1000\n value: 96.02000000000001\n - type: recall_at_3\n value: 44.767\n - type: recall_at_5\n value: 52.03\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackProgrammersRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 6184bc1440d2dbc7612be22b50686b8826d22b32\n metrics:\n - type: map_at_1\n value: 25.846000000000004\n - type: map_at_10\n value: 36.870999999999995\n - type: map_at_100\n value: 38.294\n - type: map_at_1000\n value: 38.401\n - type: map_at_3\n value: 33.163\n - type: map_at_5\n value: 35.177\n - type: mrr_at_1\n value: 31.849\n - type: mrr_at_10\n value: 41.681000000000004\n - type: mrr_at_100\n value: 42.658\n - type: mrr_at_1000\n value: 42.71\n - type: mrr_at_3\n value: 39.003\n - type: mrr_at_5\n value: 40.436\n - type: ndcg_at_1\n value: 31.849\n - type: ndcg_at_10\n value: 43.291000000000004\n - type: ndcg_at_100\n value: 49.136\n - type: ndcg_at_1000\n value: 51.168\n - type: ndcg_at_3\n value: 37.297999999999995\n - type: ndcg_at_5\n value: 39.934\n - type: precision_at_1\n value: 31.849\n - type: precision_at_10\n value: 8.219\n - type: precision_at_100\n value: 1.318\n - type: precision_at_1000\n value: 0.167\n - type: precision_at_3\n value: 18.151\n - type: precision_at_5\n value: 13.242\n - type: recall_at_1\n value: 25.846000000000004\n - type: recall_at_10\n value: 57.642\n - type: recall_at_100\n value: 82.069\n - type: recall_at_1000\n value: 95.684\n - type: recall_at_3\n value: 40.778999999999996\n - type: recall_at_5\n value: 47.647\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 4ffe81d471b1924886b33c7567bfb200e9eec5c4\n metrics:\n - type: map_at_1\n value: 25.34866666666667\n - type: map_at_10\n value: 35.65541666666667\n - type: map_at_100\n value: 36.982416666666666\n - type: map_at_1000\n value: 37.09416666666667\n - type: map_at_3\n value: 32.421499999999995\n - type: map_at_5\n value: 34.20266666666667\n - type: mrr_at_1\n value: 30.02116666666667\n - type: mrr_at_10\n value: 39.781666666666666\n - type: mrr_at_100\n value: 40.69733333333333\n - type: mrr_at_1000\n value: 40.74875\n - type: mrr_at_3\n value: 37.043083333333335\n - type: mrr_at_5\n value: 38.56391666666666\n - type: ndcg_at_1\n value: 30.02116666666667\n - type: ndcg_at_10\n value: 41.66133333333333\n - type: ndcg_at_100\n value: 47.21474999999999\n - type: ndcg_at_1000\n value: 49.29600000000001\n - type: ndcg_at_3\n value: 36.06958333333334\n - type: ndcg_at_5\n value: 38.66858333333333\n - type: precision_at_1\n value: 30.02116666666667\n - type: precision_at_10\n value: 7.497249999999999\n - type: precision_at_100\n value: 1.2044166666666667\n - type: precision_at_1000\n value: 0.15766666666666665\n - type: precision_at_3\n value: 16.83458333333333\n - type: precision_at_5\n value: 12.134\n - type: recall_at_1\n value: 25.34866666666667\n - type: recall_at_10\n value: 55.40541666666666\n - type: recall_at_100\n value: 79.38683333333333\n - type: recall_at_1000\n value: 93.50958333333334\n - type: recall_at_3\n value: 39.99858333333334\n - type: recall_at_5\n value: 46.55741666666666\n - type: map_at_1\n value: 18.336\n - type: map_at_10\n value: 26.811\n - type: map_at_100\n value: 27.892\n - type: map_at_1000\n value: 27.986\n - type: map_at_3\n value: 23.976\n - type: map_at_5\n value: 25.605\n - type: mrr_at_1\n value: 20.148\n - type: mrr_at_10\n value: 28.898000000000003\n - type: mrr_at_100\n value: 29.866\n - type: mrr_at_1000\n value: 29.929\n - type: mrr_at_3\n value: 26.247999999999998\n - type: mrr_at_5\n value: 27.744999999999997\n - type: ndcg_at_1\n value: 20.148\n - type: ndcg_at_10\n value: 32.059\n - type: ndcg_at_100\n value: 37.495\n - type: ndcg_at_1000\n value: 39.855000000000004\n - type: ndcg_at_3\n value: 26.423000000000002\n - type: ndcg_at_5\n value: 29.212\n - type: precision_at_1\n value: 20.148\n - type: precision_at_10\n value: 5.268\n - type: precision_at_100\n value: 0.872\n - type: precision_at_1000\n value: 0.11900000000000001\n - type: precision_at_3\n value: 11.459999999999999\n - type: precision_at_5\n value: 8.503\n - type: recall_at_1\n value: 18.336\n - type: recall_at_10\n value: 46.411\n - type: recall_at_100\n value: 71.33500000000001\n - type: recall_at_1000\n value: 88.895\n - type: recall_at_3\n value: 31.134\n - type: recall_at_5\n value: 37.862\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackStatsRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 65ac3a16b8e91f9cee4c9828cc7c335575432a2a\n metrics:\n - type: map_at_1\n value: 25.102000000000004\n - type: map_at_10\n value: 33.31\n - type: map_at_100\n value: 34.443\n - type: map_at_1000\n value: 34.547\n - type: map_at_3\n value: 30.932\n - type: map_at_5\n value: 32.126\n - type: mrr_at_1\n value: 28.221\n - type: mrr_at_10\n value: 36.519\n - type: mrr_at_100\n value: 37.425000000000004\n - type: mrr_at_1000\n value: 37.498\n - type: mrr_at_3\n value: 34.254\n - type: mrr_at_5\n value: 35.388999999999996\n - type: ndcg_at_1\n value: 28.221\n - type: ndcg_at_10\n value: 38.340999999999994\n - type: ndcg_at_100\n value: 43.572\n - type: ndcg_at_1000\n value: 45.979\n - type: ndcg_at_3\n value: 33.793\n - type: ndcg_at_5\n value: 35.681000000000004\n - type: precision_at_1\n value: 28.221\n - type: precision_at_10\n value: 6.135\n - type: precision_at_100\n value: 0.946\n - type: precision_at_1000\n value: 0.123\n - type: precision_at_3\n value: 14.519000000000002\n - type: precision_at_5\n value: 9.969\n - type: recall_at_1\n value: 25.102000000000004\n - type: recall_at_10\n value: 50.639\n - type: recall_at_100\n value: 74.075\n - type: recall_at_1000\n value: 91.393\n - type: recall_at_3\n value: 37.952000000000005\n - type: recall_at_5\n value: 42.71\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackTexRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 46989137a86843e03a6195de44b09deda022eec7\n metrics:\n - type: map_at_1\n value: 18.618000000000002\n - type: map_at_10\n value: 26.714\n - type: map_at_100\n value: 27.929\n - type: map_at_1000\n value: 28.057\n - type: map_at_3\n value: 24.134\n - type: map_at_5\n value: 25.575\n - type: mrr_at_1\n value: 22.573999999999998\n - type: mrr_at_10\n value: 30.786\n - type: mrr_at_100\n value: 31.746000000000002\n - type: mrr_at_1000\n value: 31.822\n - type: mrr_at_3\n value: 28.412\n - type: mrr_at_5\n value: 29.818\n - type: ndcg_at_1\n value: 22.573999999999998\n - type: ndcg_at_10\n value: 31.852000000000004\n - type: ndcg_at_100\n value: 37.477\n - type: ndcg_at_1000\n value: 40.331\n - type: ndcg_at_3\n value: 27.314\n - type: ndcg_at_5\n value: 29.485\n - type: precision_at_1\n value: 22.573999999999998\n - type: precision_at_10\n value: 5.86\n - type: precision_at_100\n value: 1.012\n - type: precision_at_1000\n value: 0.146\n - type: precision_at_3\n value: 13.099\n - type: precision_at_5\n value: 9.56\n - type: recall_at_1\n value: 18.618000000000002\n - type: recall_at_10\n value: 43.134\n - type: recall_at_100\n value: 68.294\n - type: recall_at_1000\n value: 88.283\n - type: recall_at_3\n value: 30.397999999999996\n - type: recall_at_5\n value: 35.998000000000005\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackUnixRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 6c6430d3a6d36f8d2a829195bc5dc94d7e063e53\n metrics:\n - type: map_at_1\n value: 27.76\n - type: map_at_10\n value: 37.569\n - type: map_at_100\n value: 38.784\n - type: map_at_1000\n value: 38.884\n - type: map_at_3\n value: 34.379\n - type: map_at_5\n value: 36.092999999999996\n - type: mrr_at_1\n value: 32.556000000000004\n - type: mrr_at_10\n value: 41.870000000000005\n - type: mrr_at_100\n value: 42.759\n - type: mrr_at_1000\n value: 42.806\n - type: mrr_at_3\n value: 39.086\n - type: mrr_at_5\n value: 40.574\n - type: ndcg_at_1\n value: 32.556000000000004\n - type: ndcg_at_10\n value: 43.382\n - type: ndcg_at_100\n value: 48.943\n - type: ndcg_at_1000\n value: 50.961999999999996\n - type: ndcg_at_3\n value: 37.758\n - type: ndcg_at_5\n value: 40.282000000000004\n - type: precision_at_1\n value: 32.556000000000004\n - type: precision_at_10\n value: 7.463\n - type: precision_at_100\n value: 1.1480000000000001\n - type: precision_at_1000\n value: 0.14300000000000002\n - type: precision_at_3\n value: 17.133000000000003\n - type: precision_at_5\n value: 12.164\n - type: recall_at_1\n value: 27.76\n - type: recall_at_10\n value: 56.71000000000001\n - type: recall_at_100\n value: 81.053\n - type: recall_at_1000\n value: 94.75\n - type: recall_at_3\n value: 41.387\n - type: recall_at_5\n value: 47.818\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackWebmastersRetrieval\n type: BeIR/cqadupstack\n config: default\n split: test\n revision: 160c094312a0e1facb97e55eeddb698c0abe3571\n metrics:\n - type: map_at_1\n value: 23.62\n - type: map_at_10\n value: 33.522999999999996\n - type: map_at_100\n value: 35.281\n - type: map_at_1000\n value: 35.504000000000005\n - type: map_at_3\n value: 30.314999999999998\n - type: map_at_5\n value: 32.065\n - type: mrr_at_1\n value: 28.458\n - type: mrr_at_10\n value: 38.371\n - type: mrr_at_100\n value: 39.548\n - type: mrr_at_1000\n value: 39.601\n - type: mrr_at_3\n value: 35.638999999999996\n - type: mrr_at_5\n value: 37.319\n - type: ndcg_at_1\n value: 28.458\n - type: ndcg_at_10\n value: 39.715\n - type: ndcg_at_100\n value: 46.394999999999996\n - type: ndcg_at_1000\n value: 48.943999999999996\n - type: ndcg_at_3\n value: 34.361999999999995\n - type: ndcg_at_5\n value: 37.006\n - type: precision_at_1\n value: 28.458\n - type: precision_at_10\n value: 7.5889999999999995\n - type: precision_at_100\n value: 1.514\n - type: precision_at_1000\n value: 0.242\n - type: precision_at_3\n value: 16.073999999999998\n - type: precision_at_5\n value: 11.976\n - type: recall_at_1\n value: 23.62\n - type: recall_at_10\n value: 52.117000000000004\n - type: recall_at_100\n value: 81.097\n - type: recall_at_1000\n value: 96.47\n - type: recall_at_3\n value: 37.537\n - type: recall_at_5\n value: 44.112\n - task:\n type: Retrieval\n dataset:\n name: MTEB ClimateFEVER\n type: mteb/climate-fever\n config: default\n split: test\n revision: 47f2ac6acb640fc46020b02a5b59fdda04d39380\n metrics:\n - type: map_at_1\n value: 21.149\n - type: map_at_10\n value: 35.251\n - type: map_at_100\n value: 37.342\n - type: map_at_1000\n value: 37.516\n - type: map_at_3\n value: 30.543\n - type: map_at_5\n value: 33.19\n - type: mrr_at_1\n value: 47.687000000000005\n - type: mrr_at_10\n value: 59.391000000000005\n - type: mrr_at_100\n value: 59.946999999999996\n - type: mrr_at_1000\n value: 59.965999999999994\n - type: mrr_at_3\n value: 56.938\n - type: mrr_at_5\n value: 58.498000000000005\n - type: ndcg_at_1\n value: 47.687000000000005\n - type: ndcg_at_10\n value: 45.381\n - type: ndcg_at_100\n value: 52.405\n - type: ndcg_at_1000\n value: 55.041\n - type: ndcg_at_3\n value: 40.024\n - type: ndcg_at_5\n value: 41.821999999999996\n - type: precision_at_1\n value: 47.687000000000005\n - type: precision_at_10\n value: 13.355\n - type: precision_at_100\n value: 2.113\n - type: precision_at_1000\n value: 0.261\n - type: precision_at_3\n value: 29.793999999999997\n - type: precision_at_5\n value: 21.811\n - type: recall_at_1\n value: 21.149\n - type: recall_at_10\n value: 49.937\n - type: recall_at_100\n value: 73.382\n - type: recall_at_1000\n value: 87.606\n - type: recall_at_3\n value: 35.704\n - type: recall_at_5\n value: 42.309000000000005\n - task:\n type: Retrieval\n dataset:\n name: MTEB CmedqaRetrieval\n type: C-MTEB/CmedqaRetrieval\n config: default\n split: dev\n revision: cd540c506dae1cf9e9a59c3e06f42030d54e7301\n metrics:\n - type: map_at_1\n value: 28.74\n - type: map_at_10\n value: 41.981\n - type: map_at_100\n value: 43.753\n - type: map_at_1000\n value: 43.858999999999995\n - type: map_at_3\n value: 37.634\n - type: map_at_5\n value: 40.158\n - type: mrr_at_1\n value: 43.086\n - type: mrr_at_10\n value: 51.249\n - type: mrr_at_100\n value: 52.154\n - type: mrr_at_1000\n value: 52.190999999999995\n - type: mrr_at_3\n value: 48.787000000000006\n - type: mrr_at_5\n value: 50.193\n - type: ndcg_at_1\n value: 43.086\n - type: ndcg_at_10\n value: 48.703\n - type: ndcg_at_100\n value: 55.531\n - type: ndcg_at_1000\n value: 57.267999999999994\n - type: ndcg_at_3\n value: 43.464000000000006\n - type: ndcg_at_5\n value: 45.719\n - type: precision_at_1\n value: 43.086\n - type: precision_at_10\n value: 10.568\n - type: precision_at_100\n value: 1.616\n - type: precision_at_1000\n value: 0.184\n - type: precision_at_3\n value: 24.256\n - type: precision_at_5\n value: 17.509\n - type: recall_at_1\n value: 28.74\n - type: recall_at_10\n value: 59.349\n - type: recall_at_100\n value: 87.466\n - type: recall_at_1000\n value: 98.914\n - type: recall_at_3\n value: 43.322\n - type: recall_at_5\n value: 50.409000000000006\n - task:\n type: PairClassification\n dataset:\n name: MTEB Cmnli\n type: C-MTEB/CMNLI\n config: default\n split: validation\n revision: 41bc36f332156f7adc9e38f53777c959b2ae9766\n metrics:\n - type: cos_sim_accuracy\n value: 79.03788334335539\n - type: cos_sim_ap\n value: 87.21703260472833\n - type: cos_sim_f1\n value: 79.87784187309127\n - type: cos_sim_precision\n value: 77.36634531113059\n - type: cos_sim_recall\n value: 82.55786766425064\n - type: dot_accuracy\n value: 79.03788334335539\n - type: dot_ap\n value: 87.22906528217948\n - type: dot_f1\n value: 79.87784187309127\n - type: dot_precision\n value: 77.36634531113059\n - type: dot_recall\n value: 82.55786766425064\n - type: euclidean_accuracy\n value: 79.03788334335539\n - type: euclidean_ap\n value: 87.21703670465753\n - type: euclidean_f1\n value: 79.87784187309127\n - type: euclidean_precision\n value: 77.36634531113059\n - type: euclidean_recall\n value: 82.55786766425064\n - type: manhattan_accuracy\n value: 78.28021647624774\n - type: manhattan_ap\n value: 86.66244127855394\n - type: manhattan_f1\n value: 79.24485643228577\n - type: manhattan_precision\n value: 76.71262858393521\n - type: manhattan_recall\n value: 81.94996492868833\n - type: max_accuracy\n value: 79.03788334335539\n - type: max_ap\n value: 87.22906528217948\n - type: max_f1\n value: 79.87784187309127\n - task:\n type: Retrieval\n dataset:\n name: MTEB CovidRetrieval\n type: C-MTEB/CovidRetrieval\n config: default\n split: dev\n revision: 1271c7809071a13532e05f25fb53511ffce77117\n metrics:\n - type: map_at_1\n value: 67.597\n - type: map_at_10\n value: 75.81599999999999\n - type: map_at_100\n value: 76.226\n - type: map_at_1000\n value: 76.23100000000001\n - type: map_at_3\n value: 73.907\n - type: map_at_5\n value: 75.08200000000001\n - type: mrr_at_1\n value: 67.756\n - type: mrr_at_10\n value: 75.8\n - type: mrr_at_100\n value: 76.205\n - type: mrr_at_1000\n value: 76.21\n - type: mrr_at_3\n value: 73.955\n - type: mrr_at_5\n value: 75.093\n - type: ndcg_at_1\n value: 67.756\n - type: ndcg_at_10\n value: 79.598\n - type: ndcg_at_100\n value: 81.34400000000001\n - type: ndcg_at_1000\n value: 81.477\n - type: ndcg_at_3\n value: 75.876\n - type: ndcg_at_5\n value: 77.94200000000001\n - type: precision_at_1\n value: 67.756\n - type: precision_at_10\n value: 9.231\n - type: precision_at_100\n value: 1.0\n - type: precision_at_1000\n value: 0.101\n - type: precision_at_3\n value: 27.362\n - type: precision_at_5\n value: 17.45\n - type: recall_at_1\n value: 67.597\n - type: recall_at_10\n value: 91.307\n - type: recall_at_100\n value: 98.946\n - type: recall_at_1000\n value: 100.0\n - type: recall_at_3\n value: 81.428\n - type: recall_at_5\n value: 86.407\n - task:\n type: Retrieval\n dataset:\n name: MTEB DBPedia\n type: mteb/dbpedia\n config: default\n split: test\n revision: c0f706b76e590d620bd6618b3ca8efdd34e2d659\n metrics:\n - type: map_at_1\n value: 9.33\n - type: map_at_10\n value: 23.118\n - type: map_at_100\n value: 34.28\n - type: map_at_1000\n value: 36.574\n - type: map_at_3\n value: 15.576\n - type: map_at_5\n value: 18.778\n - type: mrr_at_1\n value: 75.25\n - type: mrr_at_10\n value: 81.958\n - type: mrr_at_100\n value: 82.282\n - type: mrr_at_1000\n value: 82.285\n - type: mrr_at_3\n value: 81.042\n - type: mrr_at_5\n value: 81.62899999999999\n - type: ndcg_at_1\n value: 63.625\n - type: ndcg_at_10\n value: 50.781\n - type: ndcg_at_100\n value: 55.537000000000006\n - type: ndcg_at_1000\n value: 62.651\n - type: ndcg_at_3\n value: 55.297\n - type: ndcg_at_5\n value: 53.103\n - type: precision_at_1\n value: 75.25\n - type: precision_at_10\n value: 41.475\n - type: precision_at_100\n value: 13.5\n - type: precision_at_1000\n value: 2.686\n - type: precision_at_3\n value: 59.333000000000006\n - type: precision_at_5\n value: 51.9\n - type: recall_at_1\n value: 9.33\n - type: recall_at_10\n value: 29.398000000000003\n - type: recall_at_100\n value: 61.951\n - type: recall_at_1000\n value: 85.463\n - type: recall_at_3\n value: 17.267\n - type: recall_at_5\n value: 21.89\n - task:\n type: Retrieval\n dataset:\n name: MTEB DuRetrieval\n type: C-MTEB/DuRetrieval\n config: default\n split: dev\n revision: a1a333e290fe30b10f3f56498e3a0d911a693ced\n metrics:\n - type: map_at_1\n value: 25.608999999999998\n - type: map_at_10\n value: 78.649\n - type: map_at_100\n value: 81.67699999999999\n - type: map_at_1000\n value: 81.71000000000001\n - type: map_at_3\n value: 54.112\n - type: map_at_5\n value: 68.34700000000001\n - type: mrr_at_1\n value: 87.75\n - type: mrr_at_10\n value: 92.175\n - type: mrr_at_100\n value: 92.225\n - type: mrr_at_1000\n value: 92.227\n - type: mrr_at_3\n value: 91.833\n - type: mrr_at_5\n value: 92.06800000000001\n - type: ndcg_at_1\n value: 87.75\n - type: ndcg_at_10\n value: 86.56700000000001\n - type: ndcg_at_100\n value: 89.519\n - type: ndcg_at_1000\n value: 89.822\n - type: ndcg_at_3\n value: 84.414\n - type: ndcg_at_5\n value: 83.721\n - type: precision_at_1\n value: 87.75\n - type: precision_at_10\n value: 41.665\n - type: precision_at_100\n value: 4.827\n - type: precision_at_1000\n value: 0.49\n - type: precision_at_3\n value: 75.533\n - type: precision_at_5\n value: 64.01\n - type: recall_at_1\n value: 25.608999999999998\n - type: recall_at_10\n value: 88.708\n - type: recall_at_100\n value: 98.007\n - type: recall_at_1000\n value: 99.555\n - type: recall_at_3\n value: 57.157000000000004\n - type: recall_at_5\n value: 74.118\n - task:\n type: Retrieval\n dataset:\n name: MTEB EcomRetrieval\n type: C-MTEB/EcomRetrieval\n config: default\n split: dev\n revision: 687de13dc7294d6fd9be10c6945f9e8fec8166b9\n metrics:\n - type: map_at_1\n value: 55.800000000000004\n - type: map_at_10\n value: 65.952\n - type: map_at_100\n value: 66.413\n - type: map_at_1000\n value: 66.426\n - type: map_at_3\n value: 63.3\n - type: map_at_5\n value: 64.945\n - type: mrr_at_1\n value: 55.800000000000004\n - type: mrr_at_10\n value: 65.952\n - type: mrr_at_100\n value: 66.413\n - type: mrr_at_1000\n value: 66.426\n - type: mrr_at_3\n value: 63.3\n - type: mrr_at_5\n value: 64.945\n - type: ndcg_at_1\n value: 55.800000000000004\n - type: ndcg_at_10\n value: 71.00800000000001\n - type: ndcg_at_100\n value: 72.974\n - type: ndcg_at_1000\n value: 73.302\n - type: ndcg_at_3\n value: 65.669\n - type: ndcg_at_5\n value: 68.634\n - type: precision_at_1\n value: 55.800000000000004\n - type: precision_at_10\n value: 8.690000000000001\n - type: precision_at_100\n value: 0.955\n - type: precision_at_1000\n value: 0.098\n - type: precision_at_3\n value: 24.166999999999998\n - type: precision_at_5\n value: 15.939999999999998\n - type: recall_at_1\n value: 55.800000000000004\n - type: recall_at_10\n value: 86.9\n - type: recall_at_100\n value: 95.5\n - type: recall_at_1000\n value: 98.0\n - type: recall_at_3\n value: 72.5\n - type: recall_at_5\n value: 79.7\n - task:\n type: Classification\n dataset:\n name: MTEB EmotionClassification\n type: mteb/emotion\n config: default\n split: test\n revision: 4f58c6b202a23cf9a4da393831edf4f9183cad37\n metrics:\n - type: accuracy\n value: 67.39500000000001\n - type: f1\n value: 62.01837785021389\n - task:\n type: Retrieval\n dataset:\n name: MTEB FEVER\n type: mteb/fever\n config: default\n split: test\n revision: bea83ef9e8fb933d90a2f1d5515737465d613e12\n metrics:\n - type: map_at_1\n value: 86.27\n - type: map_at_10\n value: 92.163\n - type: map_at_100\n value: 92.351\n - type: map_at_1000\n value: 92.36\n - type: map_at_3\n value: 91.36\n - type: map_at_5\n value: 91.888\n - type: mrr_at_1\n value: 92.72399999999999\n - type: mrr_at_10\n value: 95.789\n - type: mrr_at_100\n value: 95.80300000000001\n - type: mrr_at_1000\n value: 95.804\n - type: mrr_at_3\n value: 95.64200000000001\n - type: mrr_at_5\n value: 95.75\n - type: ndcg_at_1\n value: 92.72399999999999\n - type: ndcg_at_10\n value: 94.269\n - type: ndcg_at_100\n value: 94.794\n - type: ndcg_at_1000\n value: 94.94\n - type: ndcg_at_3\n value: 93.427\n - type: ndcg_at_5\n value: 93.914\n - type: precision_at_1\n value: 92.72399999999999\n - type: precision_at_10\n value: 11.007\n - type: precision_at_100\n value: 1.153\n - type: precision_at_1000\n value: 0.11800000000000001\n - type: precision_at_3\n value: 34.993\n - type: precision_at_5\n value: 21.542\n - type: recall_at_1\n value: 86.27\n - type: recall_at_10\n value: 97.031\n - type: recall_at_100\n value: 98.839\n - type: recall_at_1000\n value: 99.682\n - type: recall_at_3\n value: 94.741\n - type: recall_at_5\n value: 96.03\n - task:\n type: Retrieval\n dataset:\n name: MTEB FiQA2018\n type: mteb/fiqa\n config: default\n split: test\n revision: 27a168819829fe9bcd655c2df245fb19452e8e06\n metrics:\n - type: map_at_1\n value: 29.561999999999998\n - type: map_at_10\n value: 48.52\n - type: map_at_100\n value: 50.753\n - type: map_at_1000\n value: 50.878\n - type: map_at_3\n value: 42.406\n - type: map_at_5\n value: 45.994\n - type: mrr_at_1\n value: 54.784\n - type: mrr_at_10\n value: 64.51400000000001\n - type: mrr_at_100\n value: 65.031\n - type: mrr_at_1000\n value: 65.05199999999999\n - type: mrr_at_3\n value: 62.474\n - type: mrr_at_5\n value: 63.562\n - type: ndcg_at_1\n value: 54.784\n - type: ndcg_at_10\n value: 57.138\n - type: ndcg_at_100\n value: 63.666999999999994\n - type: ndcg_at_1000\n value: 65.379\n - type: ndcg_at_3\n value: 52.589\n - type: ndcg_at_5\n value: 54.32599999999999\n - type: precision_at_1\n value: 54.784\n - type: precision_at_10\n value: 15.693999999999999\n - type: precision_at_100\n value: 2.259\n - type: precision_at_1000\n value: 0.256\n - type: precision_at_3\n value: 34.774\n - type: precision_at_5\n value: 25.772000000000002\n - type: recall_at_1\n value: 29.561999999999998\n - type: recall_at_10\n value: 64.708\n - type: recall_at_100\n value: 87.958\n - type: recall_at_1000\n value: 97.882\n - type: recall_at_3\n value: 48.394\n - type: recall_at_5\n value: 56.101\n - task:\n type: Retrieval\n dataset:\n name: MTEB HotpotQA\n type: mteb/hotpotqa\n config: default\n split: test\n revision: ab518f4d6fcca38d87c25209f94beba119d02014\n metrics:\n - type: map_at_1\n value: 43.72\n - type: map_at_10\n value: 71.905\n - type: map_at_100\n value: 72.685\n - type: map_at_1000\n value: 72.72800000000001\n - type: map_at_3\n value: 68.538\n - type: map_at_5\n value: 70.675\n - type: mrr_at_1\n value: 87.441\n - type: mrr_at_10\n value: 91.432\n - type: mrr_at_100\n value: 91.512\n - type: mrr_at_1000\n value: 91.513\n - type: mrr_at_3\n value: 90.923\n - type: mrr_at_5\n value: 91.252\n - type: ndcg_at_1\n value: 87.441\n - type: ndcg_at_10\n value: 79.212\n - type: ndcg_at_100\n value: 81.694\n - type: ndcg_at_1000\n value: 82.447\n - type: ndcg_at_3\n value: 74.746\n - type: ndcg_at_5\n value: 77.27199999999999\n - type: precision_at_1\n value: 87.441\n - type: precision_at_10\n value: 16.42\n - type: precision_at_100\n value: 1.833\n - type: precision_at_1000\n value: 0.193\n - type: precision_at_3\n value: 48.184\n - type: precision_at_5\n value: 30.897999999999996\n - type: recall_at_1\n value: 43.72\n - type: recall_at_10\n value: 82.1\n - type: recall_at_100\n value: 91.62700000000001\n - type: recall_at_1000\n value: 96.556\n - type: recall_at_3\n value: 72.275\n - type: recall_at_5\n value: 77.24499999999999\n - task:\n type: Classification\n dataset:\n name: MTEB IFlyTek\n type: C-MTEB/IFlyTek-classification\n config: default\n split: validation\n revision: 421605374b29664c5fc098418fe20ada9bd55f8a\n metrics:\n - type: accuracy\n value: 54.520969603693736\n - type: f1\n value: 42.359043311419626\n - task:\n type: Classification\n dataset:\n name: MTEB ImdbClassification\n type: mteb/imdb\n config: default\n split: test\n revision: 3d86128a09e091d6018b6d26cad27f2739fc2db7\n metrics:\n - type: accuracy\n value: 96.72559999999999\n - type: ap\n value: 95.01759461773742\n - type: f1\n value: 96.72429945397575\n - task:\n type: Classification\n dataset:\n name: MTEB JDReview\n type: C-MTEB/JDReview-classification\n config: default\n split: test\n revision: b7c64bd89eb87f8ded463478346f76731f07bf8b\n metrics:\n - type: accuracy\n value: 90.1688555347092\n - type: ap\n value: 63.36583667477521\n - type: f1\n value: 85.6845016521436\n - task:\n type: STS\n dataset:\n name: MTEB LCQMC\n type: C-MTEB/LCQMC\n config: default\n split: test\n revision: 17f9b096f80380fce5ed12a9be8be7784b337daf\n metrics:\n - type: cos_sim_pearson\n value: 67.35114066823127\n - type: cos_sim_spearman\n value: 72.98875207056305\n - type: euclidean_pearson\n value: 71.45620183630378\n - type: euclidean_spearman\n value: 72.98875207022671\n - type: manhattan_pearson\n value: 71.3845159780333\n - type: manhattan_spearman\n value: 72.92710990543166\n - task:\n type: Reranking\n dataset:\n name: MTEB MMarcoReranking\n type: C-MTEB/Mmarco-reranking\n config: default\n split: dev\n revision: 8e0c766dbe9e16e1d221116a3f36795fbade07f6\n metrics:\n - type: map\n value: 32.68592539803807\n - type: mrr\n value: 31.58968253968254\n - task:\n type: Retrieval\n dataset:\n name: MTEB MMarcoRetrieval\n type: C-MTEB/MMarcoRetrieval\n config: default\n split: dev\n revision: 539bbde593d947e2a124ba72651aafc09eb33fc2\n metrics:\n - type: map_at_1\n value: 71.242\n - type: map_at_10\n value: 80.01\n - type: map_at_100\n value: 80.269\n - type: map_at_1000\n value: 80.276\n - type: map_at_3\n value: 78.335\n - type: map_at_5\n value: 79.471\n - type: mrr_at_1\n value: 73.668\n - type: mrr_at_10\n value: 80.515\n - type: mrr_at_100\n value: 80.738\n - type: mrr_at_1000\n value: 80.744\n - type: mrr_at_3\n value: 79.097\n - type: mrr_at_5\n value: 80.045\n - type: ndcg_at_1\n value: 73.668\n - type: ndcg_at_10\n value: 83.357\n - type: ndcg_at_100\n value: 84.442\n - type: ndcg_at_1000\n value: 84.619\n - type: ndcg_at_3\n value: 80.286\n - type: ndcg_at_5\n value: 82.155\n - type: precision_at_1\n value: 73.668\n - type: precision_at_10\n value: 9.905\n - type: precision_at_100\n value: 1.043\n - type: precision_at_1000\n value: 0.106\n - type: precision_at_3\n value: 30.024\n - type: precision_at_5\n value: 19.017\n - type: recall_at_1\n value: 71.242\n - type: recall_at_10\n value: 93.11\n - type: recall_at_100\n value: 97.85000000000001\n - type: recall_at_1000\n value: 99.21900000000001\n - type: recall_at_3\n value: 85.137\n - type: recall_at_5\n value: 89.548\n - task:\n type: Retrieval\n dataset:\n name: MTEB MSMARCO\n type: mteb/msmarco\n config: default\n split: dev\n revision: c5a29a104738b98a9e76336939199e264163d4a0\n metrics:\n - type: map_at_1\n value: 22.006999999999998\n - type: map_at_10\n value: 34.994\n - type: map_at_100\n value: 36.183\n - type: map_at_1000\n value: 36.227\n - type: map_at_3\n value: 30.75\n - type: map_at_5\n value: 33.155\n - type: mrr_at_1\n value: 22.679\n - type: mrr_at_10\n value: 35.619\n - type: mrr_at_100\n value: 36.732\n - type: mrr_at_1000\n value: 36.77\n - type: mrr_at_3\n value: 31.44\n - type: mrr_at_5\n value: 33.811\n - type: ndcg_at_1\n value: 22.679\n - type: ndcg_at_10\n value: 42.376000000000005\n - type: ndcg_at_100\n value: 48.001\n - type: ndcg_at_1000\n value: 49.059999999999995\n - type: ndcg_at_3\n value: 33.727000000000004\n - type: ndcg_at_5\n value: 38.013000000000005\n - type: precision_at_1\n value: 22.679\n - type: precision_at_10\n value: 6.815\n - type: precision_at_100\n value: 0.962\n - type: precision_at_1000\n value: 0.105\n - type: precision_at_3\n value: 14.441\n - type: precision_at_5\n value: 10.817\n - type: recall_at_1\n value: 22.006999999999998\n - type: recall_at_10\n value: 65.158\n - type: recall_at_100\n value: 90.997\n - type: recall_at_1000\n value: 98.996\n - type: recall_at_3\n value: 41.646\n - type: recall_at_5\n value: 51.941\n - task:\n type: Classification\n dataset:\n name: MTEB MTOPDomainClassification (en)\n type: mteb/mtop_domain\n config: en\n split: test\n revision: d80d48c1eb48d3562165c59d59d0034df9fff0bf\n metrics:\n - type: accuracy\n value: 97.55129958960327\n - type: f1\n value: 97.43464802675416\n - task:\n type: Classification\n dataset:\n name: MTEB MTOPIntentClassification (en)\n type: mteb/mtop_intent\n config: en\n split: test\n revision: ae001d0e6b1228650b7bd1c2c65fb50ad11a8aba\n metrics:\n - type: accuracy\n value: 90.4719562243502\n - type: f1\n value: 70.76460034443902\n - task:\n type: Classification\n dataset:\n name: MTEB MassiveIntentClassification (en)\n type: mteb/amazon_massive_intent\n config: en\n split: test\n revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7\n metrics:\n - type: accuracy\n value: 83.49024882313383\n - type: f1\n value: 81.44067057564666\n - task:\n type: Classification\n dataset:\n name: MTEB MassiveIntentClassification (zh-CN)\n type: mteb/amazon_massive_intent\n config: zh-CN\n split: test\n revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7\n metrics:\n - type: accuracy\n value: 79.88231338264963\n - type: f1\n value: 77.13536609019927\n - task:\n type: Classification\n dataset:\n name: MTEB MassiveScenarioClassification (en)\n type: mteb/amazon_massive_scenario\n config: en\n split: test\n revision: 7d571f92784cd94a019292a1f45445077d0ef634\n metrics:\n - type: accuracy\n value: 87.23268325487558\n - type: f1\n value: 86.36737921996752\n - task:\n type: Classification\n dataset:\n name: MTEB MassiveScenarioClassification (zh-CN)\n type: mteb/amazon_massive_scenario\n config: zh-CN\n split: test\n revision: 7d571f92784cd94a019292a1f45445077d0ef634\n metrics:\n - type: accuracy\n value: 84.50571620712844\n - type: f1\n value: 83.4128768262944\n - task:\n type: Retrieval\n dataset:\n name: MTEB MedicalRetrieval\n type: C-MTEB/MedicalRetrieval\n config: default\n split: dev\n revision: 2039188fb5800a9803ba5048df7b76e6fb151fc6\n metrics:\n - type: map_at_1\n value: 56.89999999999999\n - type: map_at_10\n value: 63.438\n - type: map_at_100\n value: 63.956\n - type: map_at_1000\n value: 63.991\n - type: map_at_3\n value: 61.983\n - type: map_at_5\n value: 62.778\n - type: mrr_at_1\n value: 56.99999999999999\n - type: mrr_at_10\n value: 63.483000000000004\n - type: mrr_at_100\n value: 63.993\n - type: mrr_at_1000\n value: 64.02799999999999\n - type: mrr_at_3\n value: 62.017\n - type: mrr_at_5\n value: 62.812\n - type: ndcg_at_1\n value: 56.89999999999999\n - type: ndcg_at_10\n value: 66.61\n - type: ndcg_at_100\n value: 69.387\n - type: ndcg_at_1000\n value: 70.327\n - type: ndcg_at_3\n value: 63.583999999999996\n - type: ndcg_at_5\n value: 65.0\n - type: precision_at_1\n value: 56.89999999999999\n - type: precision_at_10\n value: 7.66\n - type: precision_at_100\n value: 0.902\n - type: precision_at_1000\n value: 0.098\n - type: precision_at_3\n value: 22.733\n - type: precision_at_5\n value: 14.32\n - type: recall_at_1\n value: 56.89999999999999\n - type: recall_at_10\n value: 76.6\n - type: recall_at_100\n value: 90.2\n - type: recall_at_1000\n value: 97.6\n - type: recall_at_3\n value: 68.2\n - type: recall_at_5\n value: 71.6\n - task:\n type: Clustering\n dataset:\n name: MTEB MedrxivClusteringP2P\n type: mteb/medrxiv-clustering-p2p\n config: default\n split: test\n revision: e7a26af6f3ae46b30dde8737f02c07b1505bcc73\n metrics:\n - type: v_measure\n value: 40.32149153753394\n - task:\n type: Clustering\n dataset:\n name: MTEB MedrxivClusteringS2S\n type: mteb/medrxiv-clustering-s2s\n config: default\n split: test\n revision: 35191c8c0dca72d8ff3efcd72aa802307d469663\n metrics:\n - type: v_measure\n value: 39.40319973495386\n - task:\n type: Reranking\n dataset:\n name: MTEB MindSmallReranking\n type: mteb/mind_small\n config: default\n split: test\n revision: 3bdac13927fdc888b903db93b2ffdbd90b295a69\n metrics:\n - type: map\n value: 33.9769104898534\n - type: mrr\n value: 35.32831430710564\n - task:\n type: Classification\n dataset:\n name: MTEB MultilingualSentiment\n type: C-MTEB/MultilingualSentiment-classification\n config: default\n split: validation\n revision: 46958b007a63fdbf239b7672c25d0bea67b5ea1a\n metrics:\n - type: accuracy\n value: 81.80666666666667\n - type: f1\n value: 81.83278699395508\n - task:\n type: Retrieval\n dataset:\n name: MTEB NFCorpus\n type: mteb/nfcorpus\n config: default\n split: test\n revision: ec0fa4fe99da2ff19ca1214b7966684033a58814\n metrics:\n - type: map_at_1\n value: 6.3\n - type: map_at_10\n value: 14.151\n - type: map_at_100\n value: 18.455\n - type: map_at_1000\n value: 20.186999999999998\n - type: map_at_3\n value: 10.023\n - type: map_at_5\n value: 11.736\n - type: mrr_at_1\n value: 49.536\n - type: mrr_at_10\n value: 58.516\n - type: mrr_at_100\n value: 59.084\n - type: mrr_at_1000\n value: 59.114\n - type: mrr_at_3\n value: 56.45\n - type: mrr_at_5\n value: 57.642\n - type: ndcg_at_1\n value: 47.522999999999996\n - type: ndcg_at_10\n value: 38.4\n - type: ndcg_at_100\n value: 35.839999999999996\n - type: ndcg_at_1000\n value: 44.998\n - type: ndcg_at_3\n value: 43.221\n - type: ndcg_at_5\n value: 40.784\n - type: precision_at_1\n value: 49.536\n - type: precision_at_10\n value: 28.977999999999998\n - type: precision_at_100\n value: 9.378\n - type: precision_at_1000\n value: 2.2769999999999997\n - type: precision_at_3\n value: 40.454\n - type: precision_at_5\n value: 35.418\n - type: recall_at_1\n value: 6.3\n - type: recall_at_10\n value: 19.085\n - type: recall_at_100\n value: 38.18\n - type: recall_at_1000\n value: 71.219\n - type: recall_at_3\n value: 11.17\n - type: recall_at_5\n value: 13.975999999999999\n - task:\n type: Retrieval\n dataset:\n name: MTEB NQ\n type: mteb/nq\n config: default\n split: test\n revision: b774495ed302d8c44a3a7ea25c90dbce03968f31\n metrics:\n - type: map_at_1\n value: 43.262\n - type: map_at_10\n value: 60.387\n - type: map_at_100\n value: 61.102000000000004\n - type: map_at_1000\n value: 61.111000000000004\n - type: map_at_3\n value: 56.391999999999996\n - type: map_at_5\n value: 58.916000000000004\n - type: mrr_at_1\n value: 48.725\n - type: mrr_at_10\n value: 62.812999999999995\n - type: mrr_at_100\n value: 63.297000000000004\n - type: mrr_at_1000\n value: 63.304\n - type: mrr_at_3\n value: 59.955999999999996\n - type: mrr_at_5\n value: 61.785999999999994\n - type: ndcg_at_1\n value: 48.696\n - type: ndcg_at_10\n value: 67.743\n - type: ndcg_at_100\n value: 70.404\n - type: ndcg_at_1000\n value: 70.60600000000001\n - type: ndcg_at_3\n value: 60.712999999999994\n - type: ndcg_at_5\n value: 64.693\n - type: precision_at_1\n value: 48.696\n - type: precision_at_10\n value: 10.513\n - type: precision_at_100\n value: 1.196\n - type: precision_at_1000\n value: 0.121\n - type: precision_at_3\n value: 27.221\n - type: precision_at_5\n value: 18.701999999999998\n - type: recall_at_1\n value: 43.262\n - type: recall_at_10\n value: 87.35300000000001\n - type: recall_at_100\n value: 98.31299999999999\n - type: recall_at_1000\n value: 99.797\n - type: recall_at_3\n value: 69.643\n - type: recall_at_5\n value: 78.645\n - task:\n type: PairClassification\n dataset:\n name: MTEB Ocnli\n type: C-MTEB/OCNLI\n config: default\n split: validation\n revision: 66e76a618a34d6d565d5538088562851e6daa7ec\n metrics:\n - type: cos_sim_accuracy\n value: 72.65836491608013\n - type: cos_sim_ap\n value: 78.75807247519593\n - type: cos_sim_f1\n value: 74.84662576687117\n - type: cos_sim_precision\n value: 63.97003745318352\n - type: cos_sim_recall\n value: 90.17951425554382\n - type: dot_accuracy\n value: 72.65836491608013\n - type: dot_ap\n value: 78.75807247519593\n - type: dot_f1\n value: 74.84662576687117\n - type: dot_precision\n value: 63.97003745318352\n - type: dot_recall\n value: 90.17951425554382\n - type: euclidean_accuracy\n value: 72.65836491608013\n - type: euclidean_ap\n value: 78.75807247519593\n - type: euclidean_f1\n value: 74.84662576687117\n - type: euclidean_precision\n value: 63.97003745318352\n - type: euclidean_recall\n value: 90.17951425554382\n - type: manhattan_accuracy\n value: 72.00866269626421\n - type: manhattan_ap\n value: 78.34663376353235\n - type: manhattan_f1\n value: 74.13234613604813\n - type: manhattan_precision\n value: 65.98023064250413\n - type: manhattan_recall\n value: 84.58289334741288\n - type: max_accuracy\n value: 72.65836491608013\n - type: max_ap\n value: 78.75807247519593\n - type: max_f1\n value: 74.84662576687117\n - task:\n type: Classification\n dataset:\n name: MTEB OnlineShopping\n type: C-MTEB/OnlineShopping-classification\n config: default\n split: test\n revision: e610f2ebd179a8fda30ae534c3878750a96db120\n metrics:\n - type: accuracy\n value: 94.46999999999998\n - type: ap\n value: 93.56401511160975\n - type: f1\n value: 94.46692790889986\n - task:\n type: STS\n dataset:\n name: MTEB PAWSX\n type: C-MTEB/PAWSX\n config: default\n split: test\n revision: 9c6a90e430ac22b5779fb019a23e820b11a8b5e1\n metrics:\n - type: cos_sim_pearson\n value: 46.851404503762474\n - type: cos_sim_spearman\n value: 52.74603680597415\n - type: euclidean_pearson\n value: 51.596358967977295\n - type: euclidean_spearman\n value: 52.74603680597415\n - type: manhattan_pearson\n value: 51.81838023379299\n - type: manhattan_spearman\n value: 52.79611669731429\n - task:\n type: STS\n dataset:\n name: MTEB QBQTC\n type: C-MTEB/QBQTC\n config: default\n split: test\n revision: 790b0510dc52b1553e8c49f3d2afb48c0e5c48b7\n metrics:\n - type: cos_sim_pearson\n value: 31.928376136347016\n - type: cos_sim_spearman\n value: 34.38497204533162\n - type: euclidean_pearson\n value: 32.658432953090674\n - type: euclidean_spearman\n value: 34.38497204533162\n - type: manhattan_pearson\n value: 32.887190283203054\n - type: manhattan_spearman\n value: 34.69496960849327\n - task:\n type: Retrieval\n dataset:\n name: MTEB QuoraRetrieval\n type: mteb/quora\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 69.952\n - type: map_at_10\n value: 84.134\n - type: map_at_100\n value: 84.795\n - type: map_at_1000\n value: 84.809\n - type: map_at_3\n value: 81.085\n - type: map_at_5\n value: 82.976\n - type: mrr_at_1\n value: 80.56\n - type: mrr_at_10\n value: 87.105\n - type: mrr_at_100\n value: 87.20700000000001\n - type: mrr_at_1000\n value: 87.208\n - type: mrr_at_3\n value: 86.118\n - type: mrr_at_5\n value: 86.79299999999999\n - type: ndcg_at_1\n value: 80.57\n - type: ndcg_at_10\n value: 88.047\n - type: ndcg_at_100\n value: 89.266\n - type: ndcg_at_1000\n value: 89.34299999999999\n - type: ndcg_at_3\n value: 85.052\n - type: ndcg_at_5\n value: 86.68299999999999\n - type: precision_at_1\n value: 80.57\n - type: precision_at_10\n value: 13.439\n - type: precision_at_100\n value: 1.536\n - type: precision_at_1000\n value: 0.157\n - type: precision_at_3\n value: 37.283\n - type: precision_at_5\n value: 24.558\n - type: recall_at_1\n value: 69.952\n - type: recall_at_10\n value: 95.599\n - type: recall_at_100\n value: 99.67099999999999\n - type: recall_at_1000\n value: 99.983\n - type: recall_at_3\n value: 87.095\n - type: recall_at_5\n value: 91.668\n - task:\n type: Clustering\n dataset:\n name: MTEB RedditClustering\n type: mteb/reddit-clustering\n config: default\n split: test\n revision: 24640382cdbf8abc73003fb0fa6d111a705499eb\n metrics:\n - type: v_measure\n value: 70.12802769698337\n - task:\n type: Clustering\n dataset:\n name: MTEB RedditClusteringP2P\n type: mteb/reddit-clustering-p2p\n config: default\n split: test\n revision: 282350215ef01743dc01b456c7f5241fa8937f16\n metrics:\n - type: v_measure\n value: 71.19047621740276\n - task:\n type: Retrieval\n dataset:\n name: MTEB SCIDOCS\n type: mteb/scidocs\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 6.208\n - type: map_at_10\n value: 17.036\n - type: map_at_100\n value: 20.162\n - type: map_at_1000\n value: 20.552\n - type: map_at_3\n value: 11.591999999999999\n - type: map_at_5\n value: 14.349\n - type: mrr_at_1\n value: 30.599999999999998\n - type: mrr_at_10\n value: 43.325\n - type: mrr_at_100\n value: 44.281\n - type: mrr_at_1000\n value: 44.31\n - type: mrr_at_3\n value: 39.300000000000004\n - type: mrr_at_5\n value: 41.730000000000004\n - type: ndcg_at_1\n value: 30.599999999999998\n - type: ndcg_at_10\n value: 27.378000000000004\n - type: ndcg_at_100\n value: 37.768\n - type: ndcg_at_1000\n value: 43.275000000000006\n - type: ndcg_at_3\n value: 25.167\n - type: ndcg_at_5\n value: 22.537\n - type: precision_at_1\n value: 30.599999999999998\n - type: precision_at_10\n value: 14.46\n - type: precision_at_100\n value: 2.937\n - type: precision_at_1000\n value: 0.424\n - type: precision_at_3\n value: 23.666999999999998\n - type: precision_at_5\n value: 20.14\n - type: recall_at_1\n value: 6.208\n - type: recall_at_10\n value: 29.29\n - type: recall_at_100\n value: 59.565\n - type: recall_at_1000\n value: 85.963\n - type: recall_at_3\n value: 14.407\n - type: recall_at_5\n value: 20.412\n - task:\n type: STS\n dataset:\n name: MTEB SICK-R\n type: mteb/sickr-sts\n config: default\n split: test\n revision: a6ea5a8cab320b040a23452cc28066d9beae2cee\n metrics:\n - type: cos_sim_pearson\n value: 82.65489797062479\n - type: cos_sim_spearman\n value: 75.34808277034776\n - type: euclidean_pearson\n value: 79.28097508609059\n - type: euclidean_spearman\n value: 75.3480824481771\n - type: manhattan_pearson\n value: 78.83529262858895\n - type: manhattan_spearman\n value: 74.96318170787025\n - task:\n type: STS\n dataset:\n name: MTEB STS12\n type: mteb/sts12-sts\n config: default\n split: test\n revision: a0d554a64d88156834ff5ae9920b964011b16384\n metrics:\n - type: cos_sim_pearson\n value: 85.06920163624117\n - type: cos_sim_spearman\n value: 77.24549887905519\n - type: euclidean_pearson\n value: 85.58740280635266\n - type: euclidean_spearman\n value: 77.24652170306867\n - type: manhattan_pearson\n value: 85.77917470895854\n - type: manhattan_spearman\n value: 77.54426264008778\n - task:\n type: STS\n dataset:\n name: MTEB STS13\n type: mteb/sts13-sts\n config: default\n split: test\n revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca\n metrics:\n - type: cos_sim_pearson\n value: 80.9762185094084\n - type: cos_sim_spearman\n value: 80.98090253728394\n - type: euclidean_pearson\n value: 80.88451512135202\n - type: euclidean_spearman\n value: 80.98090253728394\n - type: manhattan_pearson\n value: 80.7606664599805\n - type: manhattan_spearman\n value: 80.87197716950068\n - task:\n type: STS\n dataset:\n name: MTEB STS14\n type: mteb/sts14-sts\n config: default\n split: test\n revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375\n metrics:\n - type: cos_sim_pearson\n value: 81.91239166620251\n - type: cos_sim_spearman\n value: 76.36798509005328\n - type: euclidean_pearson\n value: 80.6393872615655\n - type: euclidean_spearman\n value: 76.36798836339655\n - type: manhattan_pearson\n value: 80.50765898709096\n - type: manhattan_spearman\n value: 76.31958999372227\n - task:\n type: STS\n dataset:\n name: MTEB STS15\n type: mteb/sts15-sts\n config: default\n split: test\n revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3\n metrics:\n - type: cos_sim_pearson\n value: 83.68800355225011\n - type: cos_sim_spearman\n value: 84.47549220803403\n - type: euclidean_pearson\n value: 83.86859896384159\n - type: euclidean_spearman\n value: 84.47551564954756\n - type: manhattan_pearson\n value: 83.74201103044383\n - type: manhattan_spearman\n value: 84.39903759718152\n - task:\n type: STS\n dataset:\n name: MTEB STS16\n type: mteb/sts16-sts\n config: default\n split: test\n revision: 4d8694f8f0e0100860b497b999b3dbed754a0513\n metrics:\n - type: cos_sim_pearson\n value: 78.24197302553398\n - type: cos_sim_spearman\n value: 79.44526946553684\n - type: euclidean_pearson\n value: 79.12747636563053\n - type: euclidean_spearman\n value: 79.44526946553684\n - type: manhattan_pearson\n value: 78.94407504115144\n - type: manhattan_spearman\n value: 79.24858249553934\n - task:\n type: STS\n dataset:\n name: MTEB STS17 (en-en)\n type: mteb/sts17-crosslingual-sts\n config: en-en\n split: test\n revision: af5e6fb845001ecf41f4c1e033ce921939a2a68d\n metrics:\n - type: cos_sim_pearson\n value: 89.15329071763895\n - type: cos_sim_spearman\n value: 88.67251952242073\n - type: euclidean_pearson\n value: 89.16908249259637\n - type: euclidean_spearman\n value: 88.67251952242073\n - type: manhattan_pearson\n value: 89.1279735094785\n - type: manhattan_spearman\n value: 88.81731953658254\n - task:\n type: STS\n dataset:\n name: MTEB STS22 (en)\n type: mteb/sts22-crosslingual-sts\n config: en\n split: test\n revision: eea2b4fe26a775864c896887d910b76a8098ad3f\n metrics:\n - type: cos_sim_pearson\n value: 69.44962535524695\n - type: cos_sim_spearman\n value: 71.75861316291065\n - type: euclidean_pearson\n value: 72.42347748883483\n - type: euclidean_spearman\n value: 71.75861316291065\n - type: manhattan_pearson\n value: 72.57545073534365\n - type: manhattan_spearman\n value: 71.90087671205625\n - task:\n type: STS\n dataset:\n name: MTEB STS22 (zh)\n type: mteb/sts22-crosslingual-sts\n config: zh\n split: test\n revision: eea2b4fe26a775864c896887d910b76a8098ad3f\n metrics:\n - type: cos_sim_pearson\n value: 68.9945443484093\n - type: cos_sim_spearman\n value: 71.46807157842791\n - type: euclidean_pearson\n value: 69.24911748374225\n - type: euclidean_spearman\n value: 69.46807157842791\n - type: manhattan_pearson\n value: 69.65580071876552\n - type: manhattan_spearman\n value: 69.68775795734852\n - task:\n type: STS\n dataset:\n name: MTEB STSB\n type: C-MTEB/STSB\n config: default\n split: test\n revision: 0cde68302b3541bb8b3c340dc0644b0b745b3dc0\n metrics:\n - type: cos_sim_pearson\n value: 77.39283860361535\n - type: cos_sim_spearman\n value: 77.14577975930179\n - type: euclidean_pearson\n value: 76.64560889817044\n - type: euclidean_spearman\n value: 77.14577975930179\n - type: manhattan_pearson\n value: 76.82848456242104\n - type: manhattan_spearman\n value: 77.37708521460667\n - task:\n type: STS\n dataset:\n name: MTEB STSBenchmark\n type: mteb/stsbenchmark-sts\n config: default\n split: test\n revision: b0fddb56ed78048fa8b90373c8a3cfc37b684831\n metrics:\n - type: cos_sim_pearson\n value: 84.14036697885552\n - type: cos_sim_spearman\n value: 83.10901632378086\n - type: euclidean_pearson\n value: 83.59991244380554\n - type: euclidean_spearman\n value: 83.10901632378086\n - type: manhattan_pearson\n value: 83.56632266895113\n - type: manhattan_spearman\n value: 83.17610542379353\n - task:\n type: Reranking\n dataset:\n name: MTEB SciDocsRR\n type: mteb/scidocs-reranking\n config: default\n split: test\n revision: d3c5e1fc0b855ab6097bf1cda04dd73947d7caab\n metrics:\n - type: map\n value: 88.98026856845443\n - type: mrr\n value: 96.80987494712984\n - task:\n type: Retrieval\n dataset:\n name: MTEB SciFact\n type: mteb/scifact\n config: default\n split: test\n revision: 0228b52cf27578f30900b9e5271d331663a030d7\n metrics:\n - type: map_at_1\n value: 41.661\n - type: map_at_10\n value: 55.492\n - type: map_at_100\n value: 56.237\n - type: map_at_1000\n value: 56.255\n - type: map_at_3\n value: 51.05\n - type: map_at_5\n value: 54.01200000000001\n - type: mrr_at_1\n value: 44.0\n - type: mrr_at_10\n value: 56.443\n - type: mrr_at_100\n value: 57.13700000000001\n - type: mrr_at_1000\n value: 57.152\n - type: mrr_at_3\n value: 52.944\n - type: mrr_at_5\n value: 55.37800000000001\n - type: ndcg_at_1\n value: 44.0\n - type: ndcg_at_10\n value: 62.312999999999995\n - type: ndcg_at_100\n value: 65.63900000000001\n - type: ndcg_at_1000\n value: 66.019\n - type: ndcg_at_3\n value: 54.67999999999999\n - type: ndcg_at_5\n value: 59.284000000000006\n - type: precision_at_1\n value: 44.0\n - type: precision_at_10\n value: 9.367\n - type: precision_at_100\n value: 1.0999999999999999\n - type: precision_at_1000\n value: 0.11299999999999999\n - type: precision_at_3\n value: 22.778000000000002\n - type: precision_at_5\n value: 16.467000000000002\n - type: recall_at_1\n value: 41.661\n - type: recall_at_10\n value: 82.306\n - type: recall_at_100\n value: 97.167\n - type: recall_at_1000\n value: 100.0\n - type: recall_at_3\n value: 62.461\n - type: recall_at_5\n value: 73.411\n - task:\n type: PairClassification\n dataset:\n name: MTEB SprintDuplicateQuestions\n type: mteb/sprintduplicatequestions-pairclassification\n config: default\n split: test\n revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46\n metrics:\n - type: cos_sim_accuracy\n value: 99.90693069306931\n - type: cos_sim_ap\n value: 97.86562522779887\n - type: cos_sim_f1\n value: 95.27162977867204\n - type: cos_sim_precision\n value: 95.8502024291498\n - type: cos_sim_recall\n value: 94.69999999999999\n - type: dot_accuracy\n value: 99.90693069306931\n - type: dot_ap\n value: 97.86562522779887\n - type: dot_f1\n value: 95.27162977867204\n - type: dot_precision\n value: 95.8502024291498\n - type: dot_recall\n value: 94.69999999999999\n - type: euclidean_accuracy\n value: 99.90693069306931\n - type: euclidean_ap\n value: 97.86562522779887\n - type: euclidean_f1\n value: 95.27162977867204\n - type: euclidean_precision\n value: 95.8502024291498\n - type: euclidean_recall\n value: 94.69999999999999\n - type: manhattan_accuracy\n value: 99.90693069306931\n - type: manhattan_ap\n value: 97.85527044211135\n - type: manhattan_f1\n value: 95.27638190954774\n - type: manhattan_precision\n value: 95.75757575757575\n - type: manhattan_recall\n value: 94.8\n - type: max_accuracy\n value: 99.90693069306931\n - type: max_ap\n value: 97.86562522779887\n - type: max_f1\n value: 95.27638190954774\n - task:\n type: Clustering\n dataset:\n name: MTEB StackExchangeClustering\n type: mteb/stackexchange-clustering\n config: default\n split: test\n revision: 6cbc1f7b2bc0622f2e39d2c77fa502909748c259\n metrics:\n - type: v_measure\n value: 78.89230351770412\n - task:\n type: Clustering\n dataset:\n name: MTEB StackExchangeClusteringP2P\n type: mteb/stackexchange-clustering-p2p\n config: default\n split: test\n revision: 815ca46b2622cec33ccafc3735d572c266efdb44\n metrics:\n - type: v_measure\n value: 47.52328347080355\n - task:\n type: Reranking\n dataset:\n name: MTEB StackOverflowDupQuestions\n type: mteb/stackoverflowdupquestions-reranking\n config: default\n split: test\n revision: e185fbe320c72810689fc5848eb6114e1ef5ec69\n metrics:\n - type: map\n value: 57.74702024461137\n - type: mrr\n value: 58.88074548001018\n - task:\n type: Summarization\n dataset:\n name: MTEB SummEval\n type: mteb/summeval\n config: default\n split: test\n revision: cda12ad7615edc362dbf25a00fdd61d3b1eaf93c\n metrics:\n - type: cos_sim_pearson\n value: 30.047929797503592\n - type: cos_sim_spearman\n value: 29.465371781983567\n - type: dot_pearson\n value: 30.047927690552335\n - type: dot_spearman\n value: 29.465371781983567\n - task:\n type: Reranking\n dataset:\n name: MTEB T2Reranking\n type: C-MTEB/T2Reranking\n config: default\n split: dev\n revision: 76631901a18387f85eaa53e5450019b87ad58ef9\n metrics:\n - type: map\n value: 66.54177017978034\n - type: mrr\n value: 76.76094292377299\n - task:\n type: Retrieval\n dataset:\n name: MTEB T2Retrieval\n type: C-MTEB/T2Retrieval\n config: default\n split: dev\n revision: 8731a845f1bf500a4f111cf1070785c793d10e64\n metrics:\n - type: map_at_1\n value: 28.608\n - type: map_at_10\n value: 81.266\n - type: map_at_100\n value: 84.714\n - type: map_at_1000\n value: 84.758\n - type: map_at_3\n value: 56.967\n - type: map_at_5\n value: 70.14\n - type: mrr_at_1\n value: 91.881\n - type: mrr_at_10\n value: 94.11699999999999\n - type: mrr_at_100\n value: 94.178\n - type: mrr_at_1000\n value: 94.181\n - type: mrr_at_3\n value: 93.772\n - type: mrr_at_5\n value: 93.997\n - type: ndcg_at_1\n value: 91.881\n - type: ndcg_at_10\n value: 87.954\n - type: ndcg_at_100\n value: 90.904\n - type: ndcg_at_1000\n value: 91.326\n - type: ndcg_at_3\n value: 88.838\n - type: ndcg_at_5\n value: 87.764\n - type: precision_at_1\n value: 91.881\n - type: precision_at_10\n value: 43.628\n - type: precision_at_100\n value: 5.082\n - type: precision_at_1000\n value: 0.518\n - type: precision_at_3\n value: 77.62400000000001\n - type: precision_at_5\n value: 65.269\n - type: recall_at_1\n value: 28.608\n - type: recall_at_10\n value: 87.06\n - type: recall_at_100\n value: 96.815\n - type: recall_at_1000\n value: 98.969\n - type: recall_at_3\n value: 58.506\n - type: recall_at_5\n value: 73.21600000000001\n - task:\n type: Classification\n dataset:\n name: MTEB TNews\n type: C-MTEB/TNews-classification\n config: default\n split: validation\n revision: 317f262bf1e6126357bbe89e875451e4b0938fe4\n metrics:\n - type: accuracy\n value: 56.691999999999986\n - type: f1\n value: 54.692084702788065\n - task:\n type: Retrieval\n dataset:\n name: MTEB TRECCOVID\n type: mteb/trec-covid\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 0.181\n - type: map_at_10\n value: 1.2\n - type: map_at_100\n value: 6.078\n - type: map_at_1000\n value: 14.940000000000001\n - type: map_at_3\n value: 0.45599999999999996\n - type: map_at_5\n value: 0.692\n - type: mrr_at_1\n value: 66.0\n - type: mrr_at_10\n value: 75.819\n - type: mrr_at_100\n value: 76.168\n - type: mrr_at_1000\n value: 76.168\n - type: mrr_at_3\n value: 72.667\n - type: mrr_at_5\n value: 74.86699999999999\n - type: ndcg_at_1\n value: 59.0\n - type: ndcg_at_10\n value: 52.60399999999999\n - type: ndcg_at_100\n value: 38.049\n - type: ndcg_at_1000\n value: 38.576\n - type: ndcg_at_3\n value: 57.235\n - type: ndcg_at_5\n value: 56.147000000000006\n - type: precision_at_1\n value: 66.0\n - type: precision_at_10\n value: 55.2\n - type: precision_at_100\n value: 38.78\n - type: precision_at_1000\n value: 16.986\n - type: precision_at_3\n value: 62.666999999999994\n - type: precision_at_5\n value: 60.8\n - type: recall_at_1\n value: 0.181\n - type: recall_at_10\n value: 1.471\n - type: recall_at_100\n value: 9.748999999999999\n - type: recall_at_1000\n value: 37.667\n - type: recall_at_3\n value: 0.49300000000000005\n - type: recall_at_5\n value: 0.7979999999999999\n - task:\n type: Clustering\n dataset:\n name: MTEB ThuNewsClusteringP2P\n type: C-MTEB/ThuNewsClusteringP2P\n config: default\n split: test\n revision: 5798586b105c0434e4f0fe5e767abe619442cf93\n metrics:\n - type: v_measure\n value: 78.68783858143624\n - task:\n type: Clustering\n dataset:\n name: MTEB ThuNewsClusteringS2S\n type: C-MTEB/ThuNewsClusteringS2S\n config: default\n split: test\n revision: 8a8b2caeda43f39e13c4bc5bea0f8a667896e10d\n metrics:\n - type: v_measure\n value: 77.04148998956299\n - task:\n type: Retrieval\n dataset:\n name: MTEB Touche2020\n type: mteb/touche2020\n config: default\n split: test\n revision: a34f9a33db75fa0cbb21bb5cfc3dae8dc8bec93f\n metrics:\n - type: map_at_1\n value: 1.936\n - type: map_at_10\n value: 8.942\n - type: map_at_100\n value: 14.475999999999999\n - type: map_at_1000\n value: 16.156000000000002\n - type: map_at_3\n value: 4.865\n - type: map_at_5\n value: 6.367000000000001\n - type: mrr_at_1\n value: 26.531\n - type: mrr_at_10\n value: 42.846000000000004\n - type: mrr_at_100\n value: 43.441\n - type: mrr_at_1000\n value: 43.441\n - type: mrr_at_3\n value: 36.735\n - type: mrr_at_5\n value: 40.510000000000005\n - type: ndcg_at_1\n value: 24.490000000000002\n - type: ndcg_at_10\n value: 23.262\n - type: ndcg_at_100\n value: 34.959\n - type: ndcg_at_1000\n value: 47.258\n - type: ndcg_at_3\n value: 25.27\n - type: ndcg_at_5\n value: 24.246000000000002\n - type: precision_at_1\n value: 26.531\n - type: precision_at_10\n value: 20.408\n - type: precision_at_100\n value: 7.306\n - type: precision_at_1000\n value: 1.541\n - type: precision_at_3\n value: 26.531\n - type: precision_at_5\n value: 24.082\n - type: recall_at_1\n value: 1.936\n - type: recall_at_10\n value: 15.712000000000002\n - type: recall_at_100\n value: 45.451\n - type: recall_at_1000\n value: 83.269\n - type: recall_at_3\n value: 6.442\n - type: recall_at_5\n value: 9.151\n - task:\n type: Classification\n dataset:\n name: MTEB ToxicConversationsClassification\n type: mteb/toxic_conversations_50k\n config: default\n split: test\n revision: d7c0de2777da35d6aae2200a62c6e0e5af397c4c\n metrics:\n - type: accuracy\n value: 86.564\n - type: ap\n value: 34.58766846081731\n - type: f1\n value: 72.32759831978161\n - task:\n type: Classification\n dataset:\n name: MTEB TweetSentimentExtractionClassification\n type: mteb/tweet_sentiment_extraction\n config: default\n split: test\n revision: d604517c81ca91fe16a244d1248fc021f9ecee7a\n metrics:\n - type: accuracy\n value: 77.80418788907753\n - type: f1\n value: 78.1047638421972\n - task:\n type: Clustering\n dataset:\n name: MTEB TwentyNewsgroupsClustering\n type: mteb/twentynewsgroups-clustering\n config: default\n split: test\n revision: 6125ec4e24fa026cec8a478383ee943acfbd5449\n metrics:\n - type: v_measure\n value: 59.20888659980063\n - task:\n type: PairClassification\n dataset:\n name: MTEB TwitterSemEval2015\n type: mteb/twittersemeval2015-pairclassification\n config: default\n split: test\n revision: 70970daeab8776df92f5ea462b6173c0b46fd2d1\n metrics:\n - type: cos_sim_accuracy\n value: 85.45627943017226\n - type: cos_sim_ap\n value: 72.25550061847534\n - type: cos_sim_f1\n value: 66.0611487783037\n - type: cos_sim_precision\n value: 64.11720884032779\n - type: cos_sim_recall\n value: 68.12664907651715\n - type: dot_accuracy\n value: 85.45627943017226\n - type: dot_ap\n value: 72.25574305366213\n - type: dot_f1\n value: 66.0611487783037\n - type: dot_precision\n value: 64.11720884032779\n - type: dot_recall\n value: 68.12664907651715\n - type: euclidean_accuracy\n value: 85.45627943017226\n - type: euclidean_ap\n value: 72.2557084446673\n - type: euclidean_f1\n value: 66.0611487783037\n - type: euclidean_precision\n value: 64.11720884032779\n - type: euclidean_recall\n value: 68.12664907651715\n - type: manhattan_accuracy\n value: 85.32514752339513\n - type: manhattan_ap\n value: 71.52919143472248\n - type: manhattan_f1\n value: 65.60288251190322\n - type: manhattan_precision\n value: 64.02913840743531\n - type: manhattan_recall\n value: 67.25593667546174\n - type: max_accuracy\n value: 85.45627943017226\n - type: max_ap\n value: 72.25574305366213\n - type: max_f1\n value: 66.0611487783037\n - task:\n type: PairClassification\n dataset:\n name: MTEB TwitterURLCorpus\n type: mteb/twitterurlcorpus-pairclassification\n config: default\n split: test\n revision: 8b6510b0b1fa4e4c4f879467980e9be563ec1cdf\n metrics:\n - type: cos_sim_accuracy\n value: 88.34167733923235\n - type: cos_sim_ap\n value: 84.58587730660244\n - type: cos_sim_f1\n value: 77.14170010676287\n - type: cos_sim_precision\n value: 73.91181657848324\n - type: cos_sim_recall\n value: 80.66676932553126\n - type: dot_accuracy\n value: 88.34167733923235\n - type: dot_ap\n value: 84.58585083616217\n - type: dot_f1\n value: 77.14170010676287\n - type: dot_precision\n value: 73.91181657848324\n - type: dot_recall\n value: 80.66676932553126\n - type: euclidean_accuracy\n value: 88.34167733923235\n - type: euclidean_ap\n value: 84.5858781355044\n - type: euclidean_f1\n value: 77.14170010676287\n - type: euclidean_precision\n value: 73.91181657848324\n - type: euclidean_recall\n value: 80.66676932553126\n - type: manhattan_accuracy\n value: 88.28152287809989\n - type: manhattan_ap\n value: 84.53184837110165\n - type: manhattan_f1\n value: 77.13582823915313\n - type: manhattan_precision\n value: 74.76156069364161\n - type: manhattan_recall\n value: 79.66584539574993\n - type: max_accuracy\n value: 88.34167733923235\n - type: max_ap\n value: 84.5858781355044\n - type: max_f1\n value: 77.14170010676287\n - task:\n type: Retrieval\n dataset:\n name: MTEB VideoRetrieval\n type: C-MTEB/VideoRetrieval\n config: default\n split: dev\n revision: 58c2597a5943a2ba48f4668c3b90d796283c5639\n metrics:\n - type: map_at_1\n value: 66.10000000000001\n - type: map_at_10\n value: 75.238\n - type: map_at_100\n value: 75.559\n - type: map_at_1000\n value: 75.565\n - type: map_at_3\n value: 73.68299999999999\n - type: map_at_5\n value: 74.63300000000001\n - type: mrr_at_1\n value: 66.10000000000001\n - type: mrr_at_10\n value: 75.238\n - type: mrr_at_100\n value: 75.559\n - type: mrr_at_1000\n value: 75.565\n - type: mrr_at_3\n value: 73.68299999999999\n - type: mrr_at_5\n value: 74.63300000000001\n - type: ndcg_at_1\n value: 66.10000000000001\n - type: ndcg_at_10\n value: 79.25999999999999\n - type: ndcg_at_100\n value: 80.719\n - type: ndcg_at_1000\n value: 80.862\n - type: ndcg_at_3\n value: 76.08200000000001\n - type: ndcg_at_5\n value: 77.782\n - type: precision_at_1\n value: 66.10000000000001\n - type: precision_at_10\n value: 9.17\n - type: precision_at_100\n value: 0.983\n - type: precision_at_1000\n value: 0.099\n - type: precision_at_3\n value: 27.667\n - type: precision_at_5\n value: 17.419999999999998\n - type: recall_at_1\n value: 66.10000000000001\n - type: recall_at_10\n value: 91.7\n - type: recall_at_100\n value: 98.3\n - type: recall_at_1000\n value: 99.4\n - type: recall_at_3\n value: 83.0\n - type: recall_at_5\n value: 87.1\n - task:\n type: Classification\n dataset:\n name: MTEB Waimai\n type: C-MTEB/waimai-classification\n config: default\n split: test\n revision: 339287def212450dcaa9df8c22bf93e9980c7023\n metrics:\n - type: accuracy\n value: 91.13\n - type: ap\n value: 79.55231335947015\n - type: f1\n value: 89.63091922203914\n---\n\n<p align=\"center\">\n <img src=\"https://huggingface.co/Alibaba-NLP/gme-Qwen2-VL-2B-Instruct/raw/main/images/gme_logo.png\" alt=\"GME Logo\" style=\"width: 100%; max-width: 450px;\">\n</p>\n\n<p align=\"center\"><b>GME: General Multimodal Embedding</b></p>\n\n## gme-Qwen2-VL-7B\n\nWe are excited to present `GME-Qwen2VL` series of unified **multimodal embedding models**,\nwhich are based on the advanced [Qwen2-VL](https://huggingface.co/collections/Qwen/qwen2-vl-66cee7455501d7126940800d) multimodal large language models (MLLMs).\n\nThe `GME` models support three types of input: **text**, **image**, and **image-text pair**, all of which can produce universal vector representations and have powerful retrieval performance.\n\n**Key Enhancements of GME Models**:\n\n- **Unified Multimodal Representation**: GME models can process both single-modal and combined-modal inputs, resulting in a unified vector representation. This enables versatile retrieval scenarios (Any2Any Search), supporting tasks such as text retrieval, image retrieval from text, and image-to-image searches.\n- **High Performance**: Achieves state-of-the-art (SOTA) results in our universal multimodal retrieval benchmark (**UMRB**) and demonstrate strong evaluation scores in the Multimodal Textual Evaluation Benchmark (**MTEB**).\n- **Dynamic Image Resolution**: Benefiting from `Qwen2-VL` and our training data, GME models support dynamic resolution image input.\n- **Strong Visual Retrieval Performance**: Enhanced by the Qwen2-VL model series, our models excel in visual document retrieval tasks that require a nuanced understanding of document screenshots.\n This capability is particularly beneficial for complex document understanding scenarios,\n such as multimodal retrieval-augmented generation (RAG) applications focused on academic papers.\n\n**Developed by**: Tongyi Lab, Alibaba Group\n\n**Paper**: [GME: Improving Universal Multimodal Retrieval by Multimodal LLMs](http://arxiv.org/abs/2412.16855)\n\n\n## Model List\n| Models | Model Size | Max Seq. Length | Dimension | MTEB-en| MTEB-zh | UMRB |\n|:-----: | :-----: |:-----: |:-----: |:-----: | :-----: | :-----: |\n|[`gme-Qwen2-VL-2B`](https://huggingface.co/Alibaba-NLP/gme-Qwen2-VL-2B-Instruct) | 2.21B | 32768 | 1536 | 65.27 | 68.41 | 64.45 |\n|[`gme-Qwen2-VL-7B`](https://huggingface.co/Alibaba-NLP/gme-Qwen2-VL-7B-Instruct) | 8.29B | 32768 | 3584 | 67.48 | 71.36 | 67.44 |\n\n## Usage \n**Use with custom code**\n\n```python\n# You can find the script gme_inference.py in https://huggingface.co/Alibaba-NLP/gme-Qwen2-VL-2B-Instruct/blob/main/gme_inference.py\nfrom gme_inference import GmeQwen2VL\n\nmodel = GmeQwen2VL('Alibaba-NLP/gme-Qwen2-VL-7B-Instruct')\n\ntexts = [\n \"What kind of car is this?\",\n \"The Tesla Cybertruck is a battery electric pickup truck built by Tesla, Inc. since 2023.\"\n]\nimages = [\n 'https://en.wikipedia.org/wiki/File:Tesla_Cybertruck_damaged_window.jpg',\n 'https://en.wikipedia.org/wiki/File:2024_Tesla_Cybertruck_Foundation_Series,_front_left_(Greenwich).jpg',\n]\n\n# Single-modal embedding\ne_text = gme.get_text_embeddings(texts=texts)\ne_image = gme.get_image_embeddings(images=images)\nprint((e_text * e_image).sum(-1))\n## tensor([0.1702, 0.5278], dtype=torch.float16)\n\n# How to set embedding instruction\ne_query = gme.get_text_embeddings(texts=texts, instruction='Find an image that matches the given text.')\n# If is_query=False, we always use the default instruction.\ne_corpus = gme.get_image_embeddings(images=images, is_query=False)\nprint((e_query * e_corpus).sum(-1))\n## tensor([0.2000, 0.5752], dtype=torch.float16)\n\n# Fused-modal embedding\ne_fused = gme.get_fused_embeddings(texts=texts, images=images)\nprint((e_fused[0] * e_fused[1]).sum())\n## tensor(0.6826, dtype=torch.float16)\n\n```\n\n<!-- <details>\n<summary>With transformers</summary>\n\n```python\n# Requires transformers>=4.46.2\n\nTODO\n\n# [[0.3016996383666992, 0.7503870129585266, 0.3203084468841553]]\n```\n\n</details>\n -->\n\n## Evaluation\n\nWe validated the performance on our universal multimodal retrieval benchmark (**UMRB**) among others.\n\n| | | Single-modal | | Cross-modal | | | Fused-modal | | | | Avg. |\n|--------------------|------|:------------:|:---------:|:-----------:|:-----------:|:---------:|:-----------:|:----------:|:----------:|:-----------:|:----------:|\n| | | T→T (16) | I→I (1) | T→I (4) | T→VD (10) | I→T (4) | T→IT (2) | IT→T (5) | IT→I (2) | IT→IT (3) | (47) |\n| VISTA | 0.2B | 55.15 | **31.98** | 32.88 | 10.12 | 31.23 | 45.81 | 53.32 | 8.97 | 26.26 | 37.32 |\n| CLIP-SF | 0.4B | 39.75 | 31.42 | 59.05 | 24.09 | 62.95 | 66.41 | 53.32 | 34.9 | 55.65 | 43.66 |\n| One-Peace | 4B | 43.54 | 31.27 | 61.38 | 42.9 | 65.59 | 42.72 | 28.29 | 6.73 | 23.41 | 42.01 |\n| DSE | 4.2B | 48.94 | 27.92 | 40.75 | 78.21 | 52.54 | 49.62 | 35.44 | 8.36 | 40.18 | 50.04 |\n| E5-V | 8.4B | 52.41 | 27.36 | 46.56 | 41.22 | 47.95 | 54.13 | 32.9 | 23.17 | 7.23 | 42.52 |\n| **[GME-Qwen2-VL-2B](https://huggingface.co/Alibaba-NLP/gme-Qwen2-VL-2B-Instruct)** | 2.2B | 55.93 | 29.86\t| 57.36\t| 87.84\t| 61.93 |\t76.47\t| 64.58\t | 37.02\t| 66.47 | 64.45 |\n| **[GME-Qwen2-VL-7B](https://huggingface.co/Alibaba-NLP/gme-Qwen2-VL-7B-Instruct)** | 8.3B | **58.19** | 31.89 | **61.35** | **89.92** | **65.83** | **80.94** | **66.18** | **42.56** | **73.62** | **67.44** |\n\nThe [MTEB Leaderboard](https://huggingface.co/spaces/mteb/leaderboard) English tab shows the text embeddings performence of our model.\n\n**More detailed experimental results can be found in the [paper](http://arxiv.org/abs/2412.16855)**.\n\n## Community support\n\n### Fine-tuning\n\nGME models can be fine-tuned by SWIFT：\n\n```shell\npip install ms-swift -U\n```\n\n```shell\n# MAX_PIXELS settings to reduce memory usage\n# check: https://swift.readthedocs.io/en/latest/BestPractices/Embedding.html\nnproc_per_node=8\nMAX_PIXELS=1003520 \\\nUSE_HF=1 \\\nNPROC_PER_NODE=$nproc_per_node \\\nswift sft \\\n --model Alibaba-NLP/gme-Qwen2-VL-7B-Instruct \\\n --train_type lora \\\n --dataset 'HuggingFaceM4/TextCaps:emb' \\\n --torch_dtype bfloat16 \\\n --num_train_epochs 1 \\\n --per_device_train_batch_size 2 \\\n --per_device_eval_batch_size 2 \\\n --gradient_accumulation_steps $(expr 64 / $nproc_per_node) \\\n --eval_steps 100 \\\n --save_steps 100 \\\n --eval_strategy steps \\\n --save_total_limit 5 \\\n --logging_steps 5 \\\n --output_dir output \\\n --lazy_tokenize true \\\n --warmup_ratio 0.05 \\\n --learning_rate 5e-6 \\\n --deepspeed zero3 \\\n --dataloader_num_workers 4 \\\n --task_type embedding \\\n --loss_type infonce \\\n --dataloader_drop_last true\n```\n\n## Limitations\n\n- **Single Image Input**: In `Qwen2-VL`, an image could be converted into a very large number of visual tokens. We limit the number of visual tokens to 1024 to obtain a good training efficiency.\nDue to the lack of relevant data, our models and evaluations retain one single image.\n- **English-only Training**: Our models are trained on english data only. Although the `Qwen2-VL` models are multilingual, the multilingual-multimodal embedding performance are not guaranteed.\n\nWe will extend to multi-image input, image-text interleaved data as well as multilingual data in the future version.\n\n\n## Redistribution and Use\n\nWe encourage and value diverse applications of GME models and continuous enhancements to the models themselves.\n\n- If you distribute or make GME models (or any derivative works) available, or if you create a product or service (including another AI model) that incorporates them, you must prominently display `Built with GME` on your website, user interface, blog post, About page, or product documentation.\n\n- If you utilize GME models or their outputs to develop, train, fine-tune, or improve an AI model that is distributed or made available, you must prefix the name of any such AI model with `GME`.\n\n## Cloud API Services\n\nIn addition to the open-source [GME](https://huggingface.co/collections/Alibaba-NLP/gme-models-67667e092da3491f630964d6) series models, GME series models are also available as commercial API services on Alibaba Cloud.\n\n- [MultiModal Embedding Models](https://help.aliyun.com/zh/model-studio/developer-reference/multimodal-embedding-api-reference?spm=a2c4g.11186623.0.0.321c1d1cqmoJ5C): The `multimodal-embedding-v1` model service is available.\n\nNote that the models behind the commercial APIs are not entirely identical to the open-source models.\n\n## Hiring\n\nWe have open positions for Research Interns and Full-Time Researchers to join our team at Tongyi Lab. \nWe are seeking passionate individuals with expertise in representation learning, LLM-driven information retrieval, Retrieval-Augmented Generation (RAG), and agent-based systems. \nOur team is located in the vibrant cities of Beijing and Hangzhou, offering a collaborative and dynamic work environment where you can contribute to cutting-edge advancements in artificial intelligence and machine learning.\nIf you are driven by curiosity and eager to make a meaningful impact through your work, we would love to hear from you. Please submit your resume along with a brief introduction to <a href=\"mailto:dingkun.ldk@alibaba-inc.com\">dingkun.ldk@alibaba-inc.com</a>.\n\n\n## Citation\nIf you find our paper or models helpful, please consider cite:\n\n```\n@misc{zhang2024gme,\n title={GME: Improving Universal Multimodal Retrieval by Multimodal LLMs}, \n author={Zhang, Xin and Zhang, Yanzhao and Xie, Wen and Li, Mingxin and Dai, Ziqi and Long, Dingkun and Xie, Pengjun and Zhang, Meishan and Li, Wenjie and Zhang, Min},\n year={2024},\n eprint={2412.16855},\n archivePrefix={arXiv},\n primaryClass={cs.CL},\n url={http://arxiv.org/abs/2412.16855}, \n}\n```\n"},"matched_task":{"kind":"list like","value":["SUMMARIZATION"],"string":"[\n \"SUMMARIZATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["BIOSSES","SCIFACT"],"string":"[\n \"BIOSSES\",\n \"SCIFACT\"\n]","isConversational":false}}},{"rowIdx":189,"cells":{"id":{"kind":"string","value":"McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp-unsup-simcse"},"author":{"kind":"string","value":"McGill-NLP"},"task_category":{"kind":"string","value":"sentence-similarity"},"tags":{"kind":"list like","value":["peft","safetensors","text-embedding","embeddings","information-retrieval","beir","text-classification","language-model","text-clustering","text-semantic-similarity","text-evaluation","text-reranking","feature-extraction","sentence-similarity","Sentence Similarity","natural_questions","ms_marco","fever","hotpot_qa","mteb","en","arxiv:2404.05961","license:mit","model-index","region:us"],"string":"[\n \"peft\",\n \"safetensors\",\n \"text-embedding\",\n \"embeddings\",\n \"information-retrieval\",\n \"beir\",\n \"text-classification\",\n \"language-model\",\n \"text-clustering\",\n \"text-semantic-similarity\",\n \"text-evaluation\",\n \"text-reranking\",\n \"feature-extraction\",\n \"sentence-similarity\",\n \"Sentence Similarity\",\n \"natural_questions\",\n \"ms_marco\",\n \"fever\",\n \"hotpot_qa\",\n \"mteb\",\n \"en\",\n \"arxiv:2404.05961\",\n \"license:mit\",\n \"model-index\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2024-04-30T02:45:32","string":"2024-04-30T02:45:32"},"last_modified":{"kind":"timestamp","value":"2024-04-30T03:42:49","string":"2024-04-30T03:42:49"},"downloads":{"kind":"number","value":2287,"string":"2,287"},"likes":{"kind":"number","value":4,"string":"4"},"README":{"kind":"string","value":"---\nlanguage:\n- en\nlibrary_name: peft\nlicense: mit\npipeline_tag: sentence-similarity\ntags:\n- text-embedding\n- embeddings\n- information-retrieval\n- beir\n- text-classification\n- language-model\n- text-clustering\n- text-semantic-similarity\n- text-evaluation\n- text-reranking\n- feature-extraction\n- sentence-similarity\n- Sentence Similarity\n- natural_questions\n- ms_marco\n- fever\n- hotpot_qa\n- mteb\nmodel-index:\n- name: LLM2Vec-Meta-Llama-3-unsupervised\n results:\n - task:\n type: Classification\n dataset:\n name: MTEB AmazonCounterfactualClassification (en)\n type: mteb/amazon_counterfactual\n config: en\n split: test\n revision: e8379541af4e31359cca9fbcf4b00f2671dba205\n metrics:\n - type: accuracy\n value: 75.70149253731343\n - type: ap\n value: 40.824269118508354\n - type: f1\n value: 70.55918234479084\n - task:\n type: Classification\n dataset:\n name: MTEB AmazonPolarityClassification\n type: mteb/amazon_polarity\n config: default\n split: test\n revision: e2d317d38cd51312af73b3d32a06d1a08b442046\n metrics:\n - type: accuracy\n value: 80.6812\n - type: ap\n value: 76.63327889516552\n - type: f1\n value: 80.5276613226382\n - task:\n type: Classification\n dataset:\n name: MTEB AmazonReviewsClassification (en)\n type: mteb/amazon_reviews_multi\n config: en\n split: test\n revision: 1399c76144fd37290681b995c656ef9b2e06e26d\n metrics:\n - type: accuracy\n value: 40.002\n - type: f1\n value: 39.67277678335084\n - task:\n type: Retrieval\n dataset:\n name: MTEB ArguAna\n type: arguana\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 26.173999999999996\n - type: map_at_10\n value: 42.548\n - type: map_at_100\n value: 43.492999999999995\n - type: map_at_1000\n value: 43.5\n - type: map_at_3\n value: 37.376\n - type: map_at_5\n value: 40.359\n - type: mrr_at_1\n value: 27.24\n - type: mrr_at_10\n value: 42.945\n - type: mrr_at_100\n value: 43.89\n - type: mrr_at_1000\n value: 43.897000000000006\n - type: mrr_at_3\n value: 37.779\n - type: mrr_at_5\n value: 40.755\n - type: ndcg_at_1\n value: 26.173999999999996\n - type: ndcg_at_10\n value: 51.731\n - type: ndcg_at_100\n value: 55.684999999999995\n - type: ndcg_at_1000\n value: 55.86\n - type: ndcg_at_3\n value: 41.122\n - type: ndcg_at_5\n value: 46.491\n - type: precision_at_1\n value: 26.173999999999996\n - type: precision_at_10\n value: 8.108\n - type: precision_at_100\n value: 0.9820000000000001\n - type: precision_at_1000\n value: 0.1\n - type: precision_at_3\n value: 17.330000000000002\n - type: precision_at_5\n value: 13.001\n - type: recall_at_1\n value: 26.173999999999996\n - type: recall_at_10\n value: 81.081\n - type: recall_at_100\n value: 98.222\n - type: recall_at_1000\n value: 99.57300000000001\n - type: recall_at_3\n value: 51.991\n - type: recall_at_5\n value: 65.007\n - task:\n type: Clustering\n dataset:\n name: MTEB ArxivClusteringP2P\n type: mteb/arxiv-clustering-p2p\n config: default\n split: test\n revision: a122ad7f3f0291bf49cc6f4d32aa80929df69d5d\n metrics:\n - type: v_measure\n value: 49.215974795578546\n - task:\n type: Clustering\n dataset:\n name: MTEB ArxivClusteringS2S\n type: mteb/arxiv-clustering-s2s\n config: default\n split: test\n revision: f910caf1a6075f7329cdf8c1a6135696f37dbd53\n metrics:\n - type: v_measure\n value: 41.71067780141813\n - task:\n type: Reranking\n dataset:\n name: MTEB AskUbuntuDupQuestions\n type: mteb/askubuntudupquestions-reranking\n config: default\n split: test\n revision: 2000358ca161889fa9c082cb41daa8dcfb161a54\n metrics:\n - type: map\n value: 57.15639347603191\n - type: mrr\n value: 71.4509959108297\n - task:\n type: STS\n dataset:\n name: MTEB BIOSSES\n type: mteb/biosses-sts\n config: default\n split: test\n revision: d3fb88f8f02e40887cd149695127462bbcf29b4a\n metrics:\n - type: cos_sim_spearman\n value: 84.67361609277127\n - task:\n type: Classification\n dataset:\n name: MTEB Banking77Classification\n type: mteb/banking77\n config: default\n split: test\n revision: 0fd18e25b25c072e09e0d92ab615fda904d66300\n metrics:\n - type: accuracy\n value: 84.76623376623375\n - type: f1\n value: 84.70041172334481\n - task:\n type: Clustering\n dataset:\n name: MTEB BiorxivClusteringP2P\n type: mteb/biorxiv-clustering-p2p\n config: default\n split: test\n revision: 65b79d1d13f80053f67aca9498d9402c2d9f1f40\n metrics:\n - type: v_measure\n value: 38.39251163108548\n - task:\n type: Clustering\n dataset:\n name: MTEB BiorxivClusteringS2S\n type: mteb/biorxiv-clustering-s2s\n config: default\n split: test\n revision: 258694dd0231531bc1fd9de6ceb52a0853c6d908\n metrics:\n - type: v_measure\n value: 31.30501371807517\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackAndroidRetrieval\n type: cqadupstack/android\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 26.409\n - type: map_at_10\n value: 36.925000000000004\n - type: map_at_100\n value: 38.651\n - type: map_at_1000\n value: 38.798\n - type: map_at_3\n value: 33.437\n - type: map_at_5\n value: 35.506\n - type: mrr_at_1\n value: 33.763\n - type: mrr_at_10\n value: 43.442\n - type: mrr_at_100\n value: 44.339\n - type: mrr_at_1000\n value: 44.391000000000005\n - type: mrr_at_3\n value: 40.749\n - type: mrr_at_5\n value: 42.408\n - type: ndcg_at_1\n value: 33.763\n - type: ndcg_at_10\n value: 43.486999999999995\n - type: ndcg_at_100\n value: 49.71\n - type: ndcg_at_1000\n value: 51.81\n - type: ndcg_at_3\n value: 38.586\n - type: ndcg_at_5\n value: 41.074\n - type: precision_at_1\n value: 33.763\n - type: precision_at_10\n value: 8.798\n - type: precision_at_100\n value: 1.544\n - type: precision_at_1000\n value: 0.21\n - type: precision_at_3\n value: 19.361\n - type: precision_at_5\n value: 14.335\n - type: recall_at_1\n value: 26.409\n - type: recall_at_10\n value: 55.352999999999994\n - type: recall_at_100\n value: 81.66799999999999\n - type: recall_at_1000\n value: 95.376\n - type: recall_at_3\n value: 40.304\n - type: recall_at_5\n value: 47.782000000000004\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackEnglishRetrieval\n type: cqadupstack/english\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 26.6\n - type: map_at_10\n value: 36.42\n - type: map_at_100\n value: 37.628\n - type: map_at_1000\n value: 37.767\n - type: map_at_3\n value: 33.553\n - type: map_at_5\n value: 35.118\n - type: mrr_at_1\n value: 34.394999999999996\n - type: mrr_at_10\n value: 42.586\n - type: mrr_at_100\n value: 43.251\n - type: mrr_at_1000\n value: 43.303000000000004\n - type: mrr_at_3\n value: 40.297\n - type: mrr_at_5\n value: 41.638\n - type: ndcg_at_1\n value: 34.394999999999996\n - type: ndcg_at_10\n value: 42.05\n - type: ndcg_at_100\n value: 46.371\n - type: ndcg_at_1000\n value: 48.76\n - type: ndcg_at_3\n value: 37.936\n - type: ndcg_at_5\n value: 39.827\n - type: precision_at_1\n value: 34.394999999999996\n - type: precision_at_10\n value: 8.268\n - type: precision_at_100\n value: 1.355\n - type: precision_at_1000\n value: 0.186\n - type: precision_at_3\n value: 18.726000000000003\n - type: precision_at_5\n value: 13.541\n - type: recall_at_1\n value: 26.6\n - type: recall_at_10\n value: 51.529\n - type: recall_at_100\n value: 70.038\n - type: recall_at_1000\n value: 85.67\n - type: recall_at_3\n value: 39.448\n - type: recall_at_5\n value: 44.6\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackGamingRetrieval\n type: cqadupstack/gaming\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 31.863000000000003\n - type: map_at_10\n value: 43.733\n - type: map_at_100\n value: 45.005\n - type: map_at_1000\n value: 45.074\n - type: map_at_3\n value: 40.593\n - type: map_at_5\n value: 42.272\n - type: mrr_at_1\n value: 37.555\n - type: mrr_at_10\n value: 47.532999999999994\n - type: mrr_at_100\n value: 48.431999999999995\n - type: mrr_at_1000\n value: 48.47\n - type: mrr_at_3\n value: 44.901\n - type: mrr_at_5\n value: 46.274\n - type: ndcg_at_1\n value: 37.555\n - type: ndcg_at_10\n value: 49.789\n - type: ndcg_at_100\n value: 55.059999999999995\n - type: ndcg_at_1000\n value: 56.434\n - type: ndcg_at_3\n value: 44.238\n - type: ndcg_at_5\n value: 46.698\n - type: precision_at_1\n value: 37.555\n - type: precision_at_10\n value: 8.257\n - type: precision_at_100\n value: 1.189\n - type: precision_at_1000\n value: 0.136\n - type: precision_at_3\n value: 20.23\n - type: precision_at_5\n value: 13.868\n - type: recall_at_1\n value: 31.863000000000003\n - type: recall_at_10\n value: 64.188\n - type: recall_at_100\n value: 87.02600000000001\n - type: recall_at_1000\n value: 96.761\n - type: recall_at_3\n value: 48.986000000000004\n - type: recall_at_5\n value: 55.177\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackGisRetrieval\n type: cqadupstack/gis\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 15.964\n - type: map_at_10\n value: 22.746\n - type: map_at_100\n value: 23.704\n - type: map_at_1000\n value: 23.82\n - type: map_at_3\n value: 20.5\n - type: map_at_5\n value: 21.836\n - type: mrr_at_1\n value: 17.740000000000002\n - type: mrr_at_10\n value: 24.634\n - type: mrr_at_100\n value: 25.535999999999998\n - type: mrr_at_1000\n value: 25.628\n - type: mrr_at_3\n value: 22.429\n - type: mrr_at_5\n value: 23.791\n - type: ndcg_at_1\n value: 17.740000000000002\n - type: ndcg_at_10\n value: 26.838\n - type: ndcg_at_100\n value: 31.985000000000003\n - type: ndcg_at_1000\n value: 35.289\n - type: ndcg_at_3\n value: 22.384\n - type: ndcg_at_5\n value: 24.726\n - type: precision_at_1\n value: 17.740000000000002\n - type: precision_at_10\n value: 4.35\n - type: precision_at_100\n value: 0.753\n - type: precision_at_1000\n value: 0.108\n - type: precision_at_3\n value: 9.754999999999999\n - type: precision_at_5\n value: 7.164\n - type: recall_at_1\n value: 15.964\n - type: recall_at_10\n value: 37.705\n - type: recall_at_100\n value: 61.94499999999999\n - type: recall_at_1000\n value: 87.646\n - type: recall_at_3\n value: 25.714\n - type: recall_at_5\n value: 31.402\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackMathematicaRetrieval\n type: cqadupstack/mathematica\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 9.221\n - type: map_at_10\n value: 14.735000000000001\n - type: map_at_100\n value: 15.778\n - type: map_at_1000\n value: 15.9\n - type: map_at_3\n value: 12.791\n - type: map_at_5\n value: 13.703999999999999\n - type: mrr_at_1\n value: 12.438\n - type: mrr_at_10\n value: 18.353\n - type: mrr_at_100\n value: 19.285\n - type: mrr_at_1000\n value: 19.375\n - type: mrr_at_3\n value: 16.439\n - type: mrr_at_5\n value: 17.352999999999998\n - type: ndcg_at_1\n value: 12.438\n - type: ndcg_at_10\n value: 18.703\n - type: ndcg_at_100\n value: 24.104999999999997\n - type: ndcg_at_1000\n value: 27.366\n - type: ndcg_at_3\n value: 15.055\n - type: ndcg_at_5\n value: 16.42\n - type: precision_at_1\n value: 12.438\n - type: precision_at_10\n value: 3.818\n - type: precision_at_100\n value: 0.77\n - type: precision_at_1000\n value: 0.11800000000000001\n - type: precision_at_3\n value: 7.753\n - type: precision_at_5\n value: 5.622\n - type: recall_at_1\n value: 9.221\n - type: recall_at_10\n value: 27.461999999999996\n - type: recall_at_100\n value: 51.909000000000006\n - type: recall_at_1000\n value: 75.56\n - type: recall_at_3\n value: 17.046\n - type: recall_at_5\n value: 20.766000000000002\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackPhysicsRetrieval\n type: cqadupstack/physics\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 22.828\n - type: map_at_10\n value: 33.166000000000004\n - type: map_at_100\n value: 34.618\n - type: map_at_1000\n value: 34.744\n - type: map_at_3\n value: 29.737000000000002\n - type: map_at_5\n value: 31.541000000000004\n - type: mrr_at_1\n value: 29.548000000000002\n - type: mrr_at_10\n value: 38.582\n - type: mrr_at_100\n value: 39.527\n - type: mrr_at_1000\n value: 39.577\n - type: mrr_at_3\n value: 35.884\n - type: mrr_at_5\n value: 37.413999999999994\n - type: ndcg_at_1\n value: 29.548000000000002\n - type: ndcg_at_10\n value: 39.397\n - type: ndcg_at_100\n value: 45.584\n - type: ndcg_at_1000\n value: 47.823\n - type: ndcg_at_3\n value: 33.717000000000006\n - type: ndcg_at_5\n value: 36.223\n - type: precision_at_1\n value: 29.548000000000002\n - type: precision_at_10\n value: 7.767\n - type: precision_at_100\n value: 1.2959999999999998\n - type: precision_at_1000\n value: 0.17099999999999999\n - type: precision_at_3\n value: 16.747\n - type: precision_at_5\n value: 12.203999999999999\n - type: recall_at_1\n value: 22.828\n - type: recall_at_10\n value: 52.583999999999996\n - type: recall_at_100\n value: 79.06400000000001\n - type: recall_at_1000\n value: 93.59100000000001\n - type: recall_at_3\n value: 36.671\n - type: recall_at_5\n value: 43.22\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackProgrammersRetrieval\n type: cqadupstack/programmers\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 21.366\n - type: map_at_10\n value: 30.214000000000002\n - type: map_at_100\n value: 31.647\n - type: map_at_1000\n value: 31.763\n - type: map_at_3\n value: 27.234\n - type: map_at_5\n value: 28.801\n - type: mrr_at_1\n value: 26.256\n - type: mrr_at_10\n value: 35.299\n - type: mrr_at_100\n value: 36.284\n - type: mrr_at_1000\n value: 36.342\n - type: mrr_at_3\n value: 32.572\n - type: mrr_at_5\n value: 34.050999999999995\n - type: ndcg_at_1\n value: 26.256\n - type: ndcg_at_10\n value: 35.899\n - type: ndcg_at_100\n value: 41.983\n - type: ndcg_at_1000\n value: 44.481\n - type: ndcg_at_3\n value: 30.665\n - type: ndcg_at_5\n value: 32.879999999999995\n - type: precision_at_1\n value: 26.256\n - type: precision_at_10\n value: 6.804\n - type: precision_at_100\n value: 1.187\n - type: precision_at_1000\n value: 0.16\n - type: precision_at_3\n value: 14.84\n - type: precision_at_5\n value: 10.708\n - type: recall_at_1\n value: 21.366\n - type: recall_at_10\n value: 47.878\n - type: recall_at_100\n value: 73.245\n - type: recall_at_1000\n value: 90.623\n - type: recall_at_3\n value: 33.341\n - type: recall_at_5\n value: 39.198\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackRetrieval\n type: mteb/cqadupstack\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 19.477166666666665\n - type: map_at_10\n value: 27.431416666666664\n - type: map_at_100\n value: 28.656000000000002\n - type: map_at_1000\n value: 28.787583333333338\n - type: map_at_3\n value: 24.85175\n - type: map_at_5\n value: 26.270166666666668\n - type: mrr_at_1\n value: 24.06841666666667\n - type: mrr_at_10\n value: 31.620000000000005\n - type: mrr_at_100\n value: 32.52283333333333\n - type: mrr_at_1000\n value: 32.59441666666667\n - type: mrr_at_3\n value: 29.328666666666663\n - type: mrr_at_5\n value: 30.620416666666667\n - type: ndcg_at_1\n value: 24.06841666666667\n - type: ndcg_at_10\n value: 32.404583333333335\n - type: ndcg_at_100\n value: 37.779500000000006\n - type: ndcg_at_1000\n value: 40.511583333333334\n - type: ndcg_at_3\n value: 27.994166666666665\n - type: ndcg_at_5\n value: 30.021749999999997\n - type: precision_at_1\n value: 24.06841666666667\n - type: precision_at_10\n value: 6.03725\n - type: precision_at_100\n value: 1.0500833333333337\n - type: precision_at_1000\n value: 0.14875000000000002\n - type: precision_at_3\n value: 13.419583333333335\n - type: precision_at_5\n value: 9.700666666666665\n - type: recall_at_1\n value: 19.477166666666665\n - type: recall_at_10\n value: 42.99441666666667\n - type: recall_at_100\n value: 66.787\n - type: recall_at_1000\n value: 86.18825000000001\n - type: recall_at_3\n value: 30.46366666666667\n - type: recall_at_5\n value: 35.83141666666667\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackStatsRetrieval\n type: cqadupstack/stats\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 16.246\n - type: map_at_10\n value: 22.127\n - type: map_at_100\n value: 23.006\n - type: map_at_1000\n value: 23.125\n - type: map_at_3\n value: 20.308999999999997\n - type: map_at_5\n value: 21.139\n - type: mrr_at_1\n value: 19.631999999999998\n - type: mrr_at_10\n value: 24.884999999999998\n - type: mrr_at_100\n value: 25.704\n - type: mrr_at_1000\n value: 25.793\n - type: mrr_at_3\n value: 23.083000000000002\n - type: mrr_at_5\n value: 23.942\n - type: ndcg_at_1\n value: 19.631999999999998\n - type: ndcg_at_10\n value: 25.862000000000002\n - type: ndcg_at_100\n value: 30.436000000000003\n - type: ndcg_at_1000\n value: 33.638\n - type: ndcg_at_3\n value: 22.431\n - type: ndcg_at_5\n value: 23.677\n - type: precision_at_1\n value: 19.631999999999998\n - type: precision_at_10\n value: 4.417\n - type: precision_at_100\n value: 0.7270000000000001\n - type: precision_at_1000\n value: 0.109\n - type: precision_at_3\n value: 10.327\n - type: precision_at_5\n value: 7.147\n - type: recall_at_1\n value: 16.246\n - type: recall_at_10\n value: 34.869\n - type: recall_at_100\n value: 56.221\n - type: recall_at_1000\n value: 80.449\n - type: recall_at_3\n value: 24.83\n - type: recall_at_5\n value: 28.142\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackTexRetrieval\n type: cqadupstack/tex\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 9.798\n - type: map_at_10\n value: 14.695\n - type: map_at_100\n value: 15.590000000000002\n - type: map_at_1000\n value: 15.726999999999999\n - type: map_at_3\n value: 13.004999999999999\n - type: map_at_5\n value: 13.861\n - type: mrr_at_1\n value: 12.939\n - type: mrr_at_10\n value: 18.218\n - type: mrr_at_100\n value: 18.998\n - type: mrr_at_1000\n value: 19.093\n - type: mrr_at_3\n value: 16.454\n - type: mrr_at_5\n value: 17.354\n - type: ndcg_at_1\n value: 12.939\n - type: ndcg_at_10\n value: 18.278\n - type: ndcg_at_100\n value: 22.709\n - type: ndcg_at_1000\n value: 26.064\n - type: ndcg_at_3\n value: 15.204\n - type: ndcg_at_5\n value: 16.416\n - type: precision_at_1\n value: 12.939\n - type: precision_at_10\n value: 3.768\n - type: precision_at_100\n value: 0.724\n - type: precision_at_1000\n value: 0.11800000000000001\n - type: precision_at_3\n value: 7.707999999999999\n - type: precision_at_5\n value: 5.733\n - type: recall_at_1\n value: 9.798\n - type: recall_at_10\n value: 25.562\n - type: recall_at_100\n value: 45.678999999999995\n - type: recall_at_1000\n value: 69.963\n - type: recall_at_3\n value: 16.705000000000002\n - type: recall_at_5\n value: 19.969\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackUnixRetrieval\n type: cqadupstack/unix\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 19.1\n - type: map_at_10\n value: 27.034999999999997\n - type: map_at_100\n value: 28.396\n - type: map_at_1000\n value: 28.518\n - type: map_at_3\n value: 24.363\n - type: map_at_5\n value: 25.826999999999998\n - type: mrr_at_1\n value: 23.694000000000003\n - type: mrr_at_10\n value: 31.724999999999998\n - type: mrr_at_100\n value: 32.743\n - type: mrr_at_1000\n value: 32.82\n - type: mrr_at_3\n value: 29.275000000000002\n - type: mrr_at_5\n value: 30.684\n - type: ndcg_at_1\n value: 23.694000000000003\n - type: ndcg_at_10\n value: 32.366\n - type: ndcg_at_100\n value: 38.241\n - type: ndcg_at_1000\n value: 40.973\n - type: ndcg_at_3\n value: 27.661\n - type: ndcg_at_5\n value: 29.782999999999998\n - type: precision_at_1\n value: 23.694000000000003\n - type: precision_at_10\n value: 5.951\n - type: precision_at_100\n value: 1.0070000000000001\n - type: precision_at_1000\n value: 0.135\n - type: precision_at_3\n value: 13.34\n - type: precision_at_5\n value: 9.533999999999999\n - type: recall_at_1\n value: 19.1\n - type: recall_at_10\n value: 44.032\n - type: recall_at_100\n value: 69.186\n - type: recall_at_1000\n value: 88.562\n - type: recall_at_3\n value: 30.712\n - type: recall_at_5\n value: 36.372\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackWebmastersRetrieval\n type: cqadupstack/webmasters\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 20.671\n - type: map_at_10\n value: 28.583\n - type: map_at_100\n value: 30.098999999999997\n - type: map_at_1000\n value: 30.364\n - type: map_at_3\n value: 25.825\n - type: map_at_5\n value: 27.500999999999998\n - type: mrr_at_1\n value: 25.889\n - type: mrr_at_10\n value: 33.617999999999995\n - type: mrr_at_100\n value: 34.687\n - type: mrr_at_1000\n value: 34.774\n - type: mrr_at_3\n value: 31.191999999999997\n - type: mrr_at_5\n value: 32.675\n - type: ndcg_at_1\n value: 25.889\n - type: ndcg_at_10\n value: 34.056999999999995\n - type: ndcg_at_100\n value: 40.142\n - type: ndcg_at_1000\n value: 43.614000000000004\n - type: ndcg_at_3\n value: 29.688\n - type: ndcg_at_5\n value: 32.057\n - type: precision_at_1\n value: 25.889\n - type: precision_at_10\n value: 6.7\n - type: precision_at_100\n value: 1.417\n - type: precision_at_1000\n value: 0.241\n - type: precision_at_3\n value: 14.360999999999999\n - type: precision_at_5\n value: 10.711\n - type: recall_at_1\n value: 20.671\n - type: recall_at_10\n value: 43.97\n - type: recall_at_100\n value: 71.83699999999999\n - type: recall_at_1000\n value: 94.42399999999999\n - type: recall_at_3\n value: 31.0\n - type: recall_at_5\n value: 37.489\n - task:\n type: Retrieval\n dataset:\n name: MTEB CQADupstackWordpressRetrieval\n type: cqadupstack/wordpress\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 13.66\n - type: map_at_10\n value: 18.798000000000002\n - type: map_at_100\n value: 19.75\n - type: map_at_1000\n value: 19.851\n - type: map_at_3\n value: 16.874\n - type: map_at_5\n value: 18.136\n - type: mrr_at_1\n value: 14.972\n - type: mrr_at_10\n value: 20.565\n - type: mrr_at_100\n value: 21.488\n - type: mrr_at_1000\n value: 21.567\n - type: mrr_at_3\n value: 18.669\n - type: mrr_at_5\n value: 19.861\n - type: ndcg_at_1\n value: 14.972\n - type: ndcg_at_10\n value: 22.128999999999998\n - type: ndcg_at_100\n value: 27.028000000000002\n - type: ndcg_at_1000\n value: 29.887000000000004\n - type: ndcg_at_3\n value: 18.365000000000002\n - type: ndcg_at_5\n value: 20.48\n - type: precision_at_1\n value: 14.972\n - type: precision_at_10\n value: 3.549\n - type: precision_at_100\n value: 0.632\n - type: precision_at_1000\n value: 0.093\n - type: precision_at_3\n value: 7.887\n - type: precision_at_5\n value: 5.840999999999999\n - type: recall_at_1\n value: 13.66\n - type: recall_at_10\n value: 30.801000000000002\n - type: recall_at_100\n value: 53.626\n - type: recall_at_1000\n value: 75.634\n - type: recall_at_3\n value: 20.807000000000002\n - type: recall_at_5\n value: 25.86\n - task:\n type: Retrieval\n dataset:\n name: MTEB ClimateFEVER\n type: climate-fever\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 8.622\n - type: map_at_10\n value: 16.042\n - type: map_at_100\n value: 18.023\n - type: map_at_1000\n value: 18.228\n - type: map_at_3\n value: 12.995999999999999\n - type: map_at_5\n value: 14.424000000000001\n - type: mrr_at_1\n value: 18.892999999999997\n - type: mrr_at_10\n value: 30.575000000000003\n - type: mrr_at_100\n value: 31.814999999999998\n - type: mrr_at_1000\n value: 31.856\n - type: mrr_at_3\n value: 26.851000000000003\n - type: mrr_at_5\n value: 29.021\n - type: ndcg_at_1\n value: 18.892999999999997\n - type: ndcg_at_10\n value: 23.575\n - type: ndcg_at_100\n value: 31.713\n - type: ndcg_at_1000\n value: 35.465\n - type: ndcg_at_3\n value: 18.167\n - type: ndcg_at_5\n value: 20.071\n - type: precision_at_1\n value: 18.892999999999997\n - type: precision_at_10\n value: 7.883\n - type: precision_at_100\n value: 1.652\n - type: precision_at_1000\n value: 0.23500000000000001\n - type: precision_at_3\n value: 13.898\n - type: precision_at_5\n value: 11.14\n - type: recall_at_1\n value: 8.622\n - type: recall_at_10\n value: 30.044999999999998\n - type: recall_at_100\n value: 58.072\n - type: recall_at_1000\n value: 79.226\n - type: recall_at_3\n value: 17.21\n - type: recall_at_5\n value: 22.249\n - task:\n type: Retrieval\n dataset:\n name: MTEB DBPedia\n type: dbpedia-entity\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 4.845\n - type: map_at_10\n value: 12.352\n - type: map_at_100\n value: 17.423\n - type: map_at_1000\n value: 18.529\n - type: map_at_3\n value: 8.505\n - type: map_at_5\n value: 10.213\n - type: mrr_at_1\n value: 41.75\n - type: mrr_at_10\n value: 54.6\n - type: mrr_at_100\n value: 55.345\n - type: mrr_at_1000\n value: 55.374\n - type: mrr_at_3\n value: 52.37500000000001\n - type: mrr_at_5\n value: 53.87499999999999\n - type: ndcg_at_1\n value: 31.25\n - type: ndcg_at_10\n value: 26.779999999999998\n - type: ndcg_at_100\n value: 31.929000000000002\n - type: ndcg_at_1000\n value: 39.290000000000006\n - type: ndcg_at_3\n value: 28.746\n - type: ndcg_at_5\n value: 27.334999999999997\n - type: precision_at_1\n value: 41.75\n - type: precision_at_10\n value: 22.55\n - type: precision_at_100\n value: 7.242\n - type: precision_at_1000\n value: 1.439\n - type: precision_at_3\n value: 33.833\n - type: precision_at_5\n value: 28.65\n - type: recall_at_1\n value: 4.845\n - type: recall_at_10\n value: 18.664\n - type: recall_at_100\n value: 41.085\n - type: recall_at_1000\n value: 65.242\n - type: recall_at_3\n value: 10.572\n - type: recall_at_5\n value: 13.961000000000002\n - task:\n type: Classification\n dataset:\n name: MTEB EmotionClassification\n type: mteb/emotion\n config: default\n split: test\n revision: 4f58c6b202a23cf9a4da393831edf4f9183cad37\n metrics:\n - type: accuracy\n value: 47.08\n - type: f1\n value: 42.843345856303756\n - task:\n type: Retrieval\n dataset:\n name: MTEB FEVER\n type: fever\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 33.743\n - type: map_at_10\n value: 46.521\n - type: map_at_100\n value: 47.235\n - type: map_at_1000\n value: 47.272\n - type: map_at_3\n value: 43.252\n - type: map_at_5\n value: 45.267\n - type: mrr_at_1\n value: 36.484\n - type: mrr_at_10\n value: 49.406\n - type: mrr_at_100\n value: 50.03300000000001\n - type: mrr_at_1000\n value: 50.058\n - type: mrr_at_3\n value: 46.195\n - type: mrr_at_5\n value: 48.193999999999996\n - type: ndcg_at_1\n value: 36.484\n - type: ndcg_at_10\n value: 53.42\n - type: ndcg_at_100\n value: 56.69499999999999\n - type: ndcg_at_1000\n value: 57.623999999999995\n - type: ndcg_at_3\n value: 47.010999999999996\n - type: ndcg_at_5\n value: 50.524\n - type: precision_at_1\n value: 36.484\n - type: precision_at_10\n value: 7.925\n - type: precision_at_100\n value: 0.975\n - type: precision_at_1000\n value: 0.107\n - type: precision_at_3\n value: 19.967\n - type: precision_at_5\n value: 13.87\n - type: recall_at_1\n value: 33.743\n - type: recall_at_10\n value: 71.988\n - type: recall_at_100\n value: 86.60799999999999\n - type: recall_at_1000\n value: 93.54\n - type: recall_at_3\n value: 54.855\n - type: recall_at_5\n value: 63.341\n - task:\n type: Retrieval\n dataset:\n name: MTEB FiQA2018\n type: fiqa\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 13.003\n - type: map_at_10\n value: 21.766\n - type: map_at_100\n value: 23.618\n - type: map_at_1000\n value: 23.832\n - type: map_at_3\n value: 18.282999999999998\n - type: map_at_5\n value: 20.267\n - type: mrr_at_1\n value: 26.851999999999997\n - type: mrr_at_10\n value: 34.658\n - type: mrr_at_100\n value: 35.729\n - type: mrr_at_1000\n value: 35.785\n - type: mrr_at_3\n value: 31.686999999999998\n - type: mrr_at_5\n value: 33.315\n - type: ndcg_at_1\n value: 26.851999999999997\n - type: ndcg_at_10\n value: 28.563\n - type: ndcg_at_100\n value: 36.374\n - type: ndcg_at_1000\n value: 40.306999999999995\n - type: ndcg_at_3\n value: 24.224\n - type: ndcg_at_5\n value: 25.939\n - type: precision_at_1\n value: 26.851999999999997\n - type: precision_at_10\n value: 8.193999999999999\n - type: precision_at_100\n value: 1.616\n - type: precision_at_1000\n value: 0.232\n - type: precision_at_3\n value: 16.255\n - type: precision_at_5\n value: 12.469\n - type: recall_at_1\n value: 13.003\n - type: recall_at_10\n value: 35.689\n - type: recall_at_100\n value: 65.762\n - type: recall_at_1000\n value: 89.546\n - type: recall_at_3\n value: 21.820999999999998\n - type: recall_at_5\n value: 28.097\n - task:\n type: Retrieval\n dataset:\n name: MTEB HotpotQA\n type: hotpotqa\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 29.541\n - type: map_at_10\n value: 43.088\n - type: map_at_100\n value: 44.252\n - type: map_at_1000\n value: 44.345\n - type: map_at_3\n value: 39.79\n - type: map_at_5\n value: 41.687000000000005\n - type: mrr_at_1\n value: 59.082\n - type: mrr_at_10\n value: 67.27300000000001\n - type: mrr_at_100\n value: 67.708\n - type: mrr_at_1000\n value: 67.731\n - type: mrr_at_3\n value: 65.526\n - type: mrr_at_5\n value: 66.589\n - type: ndcg_at_1\n value: 59.082\n - type: ndcg_at_10\n value: 52.372\n - type: ndcg_at_100\n value: 56.725\n - type: ndcg_at_1000\n value: 58.665\n - type: ndcg_at_3\n value: 47.129\n - type: ndcg_at_5\n value: 49.808\n - type: precision_at_1\n value: 59.082\n - type: precision_at_10\n value: 11.275\n - type: precision_at_100\n value: 1.469\n - type: precision_at_1000\n value: 0.173\n - type: precision_at_3\n value: 29.773\n - type: precision_at_5\n value: 19.980999999999998\n - type: recall_at_1\n value: 29.541\n - type: recall_at_10\n value: 56.374\n - type: recall_at_100\n value: 73.42999999999999\n - type: recall_at_1000\n value: 86.28\n - type: recall_at_3\n value: 44.659\n - type: recall_at_5\n value: 49.952999999999996\n - task:\n type: Classification\n dataset:\n name: MTEB ImdbClassification\n type: mteb/imdb\n config: default\n split: test\n revision: 3d86128a09e091d6018b6d26cad27f2739fc2db7\n metrics:\n - type: accuracy\n value: 75.1904\n - type: ap\n value: 69.80555086826531\n - type: f1\n value: 74.93725389065787\n - task:\n type: Retrieval\n dataset:\n name: MTEB MSMARCO\n type: msmarco\n config: default\n split: dev\n revision: None\n metrics:\n - type: map_at_1\n value: 7.085\n - type: map_at_10\n value: 13.344000000000001\n - type: map_at_100\n value: 14.501\n - type: map_at_1000\n value: 14.605\n - type: map_at_3\n value: 10.758\n - type: map_at_5\n value: 12.162\n - type: mrr_at_1\n value: 7.278\n - type: mrr_at_10\n value: 13.607\n - type: mrr_at_100\n value: 14.761\n - type: mrr_at_1000\n value: 14.860000000000001\n - type: mrr_at_3\n value: 11.003\n - type: mrr_at_5\n value: 12.421\n - type: ndcg_at_1\n value: 7.278\n - type: ndcg_at_10\n value: 17.473\n - type: ndcg_at_100\n value: 23.721\n - type: ndcg_at_1000\n value: 26.69\n - type: ndcg_at_3\n value: 12.078\n - type: ndcg_at_5\n value: 14.62\n - type: precision_at_1\n value: 7.278\n - type: precision_at_10\n value: 3.175\n - type: precision_at_100\n value: 0.639\n - type: precision_at_1000\n value: 0.09\n - type: precision_at_3\n value: 5.382\n - type: precision_at_5\n value: 4.519\n - type: recall_at_1\n value: 7.085\n - type: recall_at_10\n value: 30.549\n - type: recall_at_100\n value: 60.919999999999995\n - type: recall_at_1000\n value: 84.372\n - type: recall_at_3\n value: 15.675\n - type: recall_at_5\n value: 21.818\n - task:\n type: Classification\n dataset:\n name: MTEB MTOPDomainClassification (en)\n type: mteb/mtop_domain\n config: en\n split: test\n revision: d80d48c1eb48d3562165c59d59d0034df9fff0bf\n metrics:\n - type: accuracy\n value: 94.46876424988601\n - type: f1\n value: 94.23159241922738\n - task:\n type: Classification\n dataset:\n name: MTEB MTOPIntentClassification (en)\n type: mteb/mtop_intent\n config: en\n split: test\n revision: ae001d0e6b1228650b7bd1c2c65fb50ad11a8aba\n metrics:\n - type: accuracy\n value: 81.0875512995896\n - type: f1\n value: 61.674961674414\n - task:\n type: Classification\n dataset:\n name: MTEB MassiveIntentClassification (en)\n type: mteb/amazon_massive_intent\n config: en\n split: test\n revision: 31efe3c427b0bae9c22cbb560b8f15491cc6bed7\n metrics:\n - type: accuracy\n value: 75.01344989912575\n - type: f1\n value: 71.7942527839921\n - task:\n type: Classification\n dataset:\n name: MTEB MassiveScenarioClassification (en)\n type: mteb/amazon_massive_scenario\n config: en\n split: test\n revision: 7d571f92784cd94a019292a1f45445077d0ef634\n metrics:\n - type: accuracy\n value: 79.15601882985877\n - type: f1\n value: 78.82502954601195\n - task:\n type: Clustering\n dataset:\n name: MTEB MedrxivClusteringP2P\n type: mteb/medrxiv-clustering-p2p\n config: default\n split: test\n revision: e7a26af6f3ae46b30dde8737f02c07b1505bcc73\n metrics:\n - type: v_measure\n value: 31.468806971345227\n - task:\n type: Clustering\n dataset:\n name: MTEB MedrxivClusteringS2S\n type: mteb/medrxiv-clustering-s2s\n config: default\n split: test\n revision: 35191c8c0dca72d8ff3efcd72aa802307d469663\n metrics:\n - type: v_measure\n value: 27.874332804382256\n - task:\n type: Reranking\n dataset:\n name: MTEB MindSmallReranking\n type: mteb/mind_small\n config: default\n split: test\n revision: 3bdac13927fdc888b903db93b2ffdbd90b295a69\n metrics:\n - type: map\n value: 30.099340785595842\n - type: mrr\n value: 31.077367694660257\n - task:\n type: Retrieval\n dataset:\n name: MTEB NFCorpus\n type: nfcorpus\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 3.9050000000000002\n - type: map_at_10\n value: 8.931000000000001\n - type: map_at_100\n value: 11.246\n - type: map_at_1000\n value: 12.579\n - type: map_at_3\n value: 6.544\n - type: map_at_5\n value: 7.854\n - type: mrr_at_1\n value: 33.745999999999995\n - type: mrr_at_10\n value: 44.734\n - type: mrr_at_100\n value: 45.486\n - type: mrr_at_1000\n value: 45.534\n - type: mrr_at_3\n value: 42.157\n - type: mrr_at_5\n value: 43.813\n - type: ndcg_at_1\n value: 31.734\n - type: ndcg_at_10\n value: 26.284999999999997\n - type: ndcg_at_100\n value: 25.211\n - type: ndcg_at_1000\n value: 34.974\n - type: ndcg_at_3\n value: 29.918\n - type: ndcg_at_5\n value: 29.066\n - type: precision_at_1\n value: 33.745999999999995\n - type: precision_at_10\n value: 19.628\n - type: precision_at_100\n value: 6.476999999999999\n - type: precision_at_1000\n value: 1.976\n - type: precision_at_3\n value: 28.793000000000003\n - type: precision_at_5\n value: 25.759\n - type: recall_at_1\n value: 3.9050000000000002\n - type: recall_at_10\n value: 13.375\n - type: recall_at_100\n value: 28.453\n - type: recall_at_1000\n value: 61.67399999999999\n - type: recall_at_3\n value: 7.774\n - type: recall_at_5\n value: 10.754\n - task:\n type: Retrieval\n dataset:\n name: MTEB NQ\n type: nq\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 18.33\n - type: map_at_10\n value: 30.44\n - type: map_at_100\n value: 31.848\n - type: map_at_1000\n value: 31.906000000000002\n - type: map_at_3\n value: 26.143\n - type: map_at_5\n value: 28.583\n - type: mrr_at_1\n value: 21.031\n - type: mrr_at_10\n value: 33.028\n - type: mrr_at_100\n value: 34.166000000000004\n - type: mrr_at_1000\n value: 34.208\n - type: mrr_at_3\n value: 29.089\n - type: mrr_at_5\n value: 31.362000000000002\n - type: ndcg_at_1\n value: 21.031\n - type: ndcg_at_10\n value: 37.65\n - type: ndcg_at_100\n value: 43.945\n - type: ndcg_at_1000\n value: 45.338\n - type: ndcg_at_3\n value: 29.256999999999998\n - type: ndcg_at_5\n value: 33.453\n - type: precision_at_1\n value: 21.031\n - type: precision_at_10\n value: 6.8309999999999995\n - type: precision_at_100\n value: 1.035\n - type: precision_at_1000\n value: 0.117\n - type: precision_at_3\n value: 13.818\n - type: precision_at_5\n value: 10.649000000000001\n - type: recall_at_1\n value: 18.33\n - type: recall_at_10\n value: 57.330999999999996\n - type: recall_at_100\n value: 85.284\n - type: recall_at_1000\n value: 95.676\n - type: recall_at_3\n value: 35.356\n - type: recall_at_5\n value: 45.073\n - task:\n type: Retrieval\n dataset:\n name: MTEB QuoraRetrieval\n type: quora\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 66.373\n - type: map_at_10\n value: 80.233\n - type: map_at_100\n value: 80.973\n - type: map_at_1000\n value: 80.99499999999999\n - type: map_at_3\n value: 77.127\n - type: map_at_5\n value: 79.056\n - type: mrr_at_1\n value: 76.55\n - type: mrr_at_10\n value: 83.813\n - type: mrr_at_100\n value: 83.96900000000001\n - type: mrr_at_1000\n value: 83.97200000000001\n - type: mrr_at_3\n value: 82.547\n - type: mrr_at_5\n value: 83.38600000000001\n - type: ndcg_at_1\n value: 76.53999999999999\n - type: ndcg_at_10\n value: 84.638\n - type: ndcg_at_100\n value: 86.28099999999999\n - type: ndcg_at_1000\n value: 86.459\n - type: ndcg_at_3\n value: 81.19\n - type: ndcg_at_5\n value: 83.057\n - type: precision_at_1\n value: 76.53999999999999\n - type: precision_at_10\n value: 12.928999999999998\n - type: precision_at_100\n value: 1.514\n - type: precision_at_1000\n value: 0.156\n - type: precision_at_3\n value: 35.503\n - type: precision_at_5\n value: 23.512\n - type: recall_at_1\n value: 66.373\n - type: recall_at_10\n value: 93.273\n - type: recall_at_100\n value: 99.031\n - type: recall_at_1000\n value: 99.91799999999999\n - type: recall_at_3\n value: 83.55799999999999\n - type: recall_at_5\n value: 88.644\n - task:\n type: Clustering\n dataset:\n name: MTEB RedditClustering\n type: mteb/reddit-clustering\n config: default\n split: test\n revision: 24640382cdbf8abc73003fb0fa6d111a705499eb\n metrics:\n - type: v_measure\n value: 43.67174666339103\n - task:\n type: Clustering\n dataset:\n name: MTEB RedditClusteringP2P\n type: mteb/reddit-clustering-p2p\n config: default\n split: test\n revision: 282350215ef01743dc01b456c7f5241fa8937f16\n metrics:\n - type: v_measure\n value: 61.66838659211271\n - task:\n type: Retrieval\n dataset:\n name: MTEB SCIDOCS\n type: scidocs\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 2.318\n - type: map_at_10\n value: 5.938000000000001\n - type: map_at_100\n value: 7.582\n - type: map_at_1000\n value: 7.936\n - type: map_at_3\n value: 4.208\n - type: map_at_5\n value: 5.098\n - type: mrr_at_1\n value: 11.4\n - type: mrr_at_10\n value: 17.655\n - type: mrr_at_100\n value: 19.088\n - type: mrr_at_1000\n value: 19.203\n - type: mrr_at_3\n value: 15.25\n - type: mrr_at_5\n value: 16.535\n - type: ndcg_at_1\n value: 11.4\n - type: ndcg_at_10\n value: 10.388\n - type: ndcg_at_100\n value: 18.165\n - type: ndcg_at_1000\n value: 24.842\n - type: ndcg_at_3\n value: 9.414\n - type: ndcg_at_5\n value: 8.453\n - type: precision_at_1\n value: 11.4\n - type: precision_at_10\n value: 5.54\n - type: precision_at_100\n value: 1.71\n - type: precision_at_1000\n value: 0.33\n - type: precision_at_3\n value: 8.866999999999999\n - type: precision_at_5\n value: 7.580000000000001\n - type: recall_at_1\n value: 2.318\n - type: recall_at_10\n value: 11.267000000000001\n - type: recall_at_100\n value: 34.743\n - type: recall_at_1000\n value: 67.07300000000001\n - type: recall_at_3\n value: 5.408\n - type: recall_at_5\n value: 7.713\n - task:\n type: STS\n dataset:\n name: MTEB SICK-R\n type: mteb/sickr-sts\n config: default\n split: test\n revision: a6ea5a8cab320b040a23452cc28066d9beae2cee\n metrics:\n - type: cos_sim_spearman\n value: 72.15850185456762\n - task:\n type: STS\n dataset:\n name: MTEB STS12\n type: mteb/sts12-sts\n config: default\n split: test\n revision: a0d554a64d88156834ff5ae9920b964011b16384\n metrics:\n - type: cos_sim_spearman\n value: 61.59518395985063\n - task:\n type: STS\n dataset:\n name: MTEB STS13\n type: mteb/sts13-sts\n config: default\n split: test\n revision: 7e90230a92c190f1bf69ae9002b8cea547a64cca\n metrics:\n - type: cos_sim_spearman\n value: 79.71131323749228\n - task:\n type: STS\n dataset:\n name: MTEB STS14\n type: mteb/sts14-sts\n config: default\n split: test\n revision: 6031580fec1f6af667f0bd2da0a551cf4f0b2375\n metrics:\n - type: cos_sim_spearman\n value: 72.10974664733891\n - task:\n type: STS\n dataset:\n name: MTEB STS15\n type: mteb/sts15-sts\n config: default\n split: test\n revision: ae752c7c21bf194d8b67fd573edf7ae58183cbe3\n metrics:\n - type: cos_sim_spearman\n value: 82.17899407125657\n - task:\n type: STS\n dataset:\n name: MTEB STS16\n type: mteb/sts16-sts\n config: default\n split: test\n revision: 4d8694f8f0e0100860b497b999b3dbed754a0513\n metrics:\n - type: cos_sim_spearman\n value: 79.41138579273438\n - task:\n type: STS\n dataset:\n name: MTEB STS17 (en-en)\n type: mteb/sts17-crosslingual-sts\n config: en-en\n split: test\n revision: af5e6fb845001ecf41f4c1e033ce921939a2a68d\n metrics:\n - type: cos_sim_spearman\n value: 85.44343473477939\n - task:\n type: STS\n dataset:\n name: MTEB STS22 (en)\n type: mteb/sts22-crosslingual-sts\n config: en\n split: test\n revision: 6d1ba47164174a496b7fa5d3569dae26a6813b80\n metrics:\n - type: cos_sim_spearman\n value: 63.90264271389905\n - task:\n type: STS\n dataset:\n name: MTEB STSBenchmark\n type: mteb/stsbenchmark-sts\n config: default\n split: test\n revision: b0fddb56ed78048fa8b90373c8a3cfc37b684831\n metrics:\n - type: cos_sim_spearman\n value: 77.44151296326804\n - task:\n type: Reranking\n dataset:\n name: MTEB SciDocsRR\n type: mteb/scidocs-reranking\n config: default\n split: test\n revision: d3c5e1fc0b855ab6097bf1cda04dd73947d7caab\n metrics:\n - type: map\n value: 76.27597486396654\n - type: mrr\n value: 93.28127119793788\n - task:\n type: Retrieval\n dataset:\n name: MTEB SciFact\n type: scifact\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 49.594\n - type: map_at_10\n value: 60.951\n - type: map_at_100\n value: 61.68599999999999\n - type: map_at_1000\n value: 61.712\n - type: map_at_3\n value: 57.946\n - type: map_at_5\n value: 59.89\n - type: mrr_at_1\n value: 52.666999999999994\n - type: mrr_at_10\n value: 62.724000000000004\n - type: mrr_at_100\n value: 63.269\n - type: mrr_at_1000\n value: 63.291\n - type: mrr_at_3\n value: 60.167\n - type: mrr_at_5\n value: 61.95\n - type: ndcg_at_1\n value: 52.666999999999994\n - type: ndcg_at_10\n value: 66.35600000000001\n - type: ndcg_at_100\n value: 69.463\n - type: ndcg_at_1000\n value: 70.111\n - type: ndcg_at_3\n value: 60.901\n - type: ndcg_at_5\n value: 64.054\n - type: precision_at_1\n value: 52.666999999999994\n - type: precision_at_10\n value: 9.0\n - type: precision_at_100\n value: 1.073\n - type: precision_at_1000\n value: 0.11299999999999999\n - type: precision_at_3\n value: 24.221999999999998\n - type: precision_at_5\n value: 16.333000000000002\n - type: recall_at_1\n value: 49.594\n - type: recall_at_10\n value: 81.256\n - type: recall_at_100\n value: 94.989\n - type: recall_at_1000\n value: 100.0\n - type: recall_at_3\n value: 66.706\n - type: recall_at_5\n value: 74.411\n - task:\n type: PairClassification\n dataset:\n name: MTEB SprintDuplicateQuestions\n type: mteb/sprintduplicatequestions-pairclassification\n config: default\n split: test\n revision: d66bd1f72af766a5cc4b0ca5e00c162f89e8cc46\n metrics:\n - type: cos_sim_accuracy\n value: 99.65049504950495\n - type: cos_sim_ap\n value: 88.1421623503371\n - type: cos_sim_f1\n value: 81.44072036018008\n - type: cos_sim_precision\n value: 81.48148148148148\n - type: cos_sim_recall\n value: 81.39999999999999\n - type: dot_accuracy\n value: 99.37623762376238\n - type: dot_ap\n value: 69.87152032240303\n - type: dot_f1\n value: 65.64885496183206\n - type: dot_precision\n value: 72.18225419664267\n - type: dot_recall\n value: 60.199999999999996\n - type: euclidean_accuracy\n value: 99.63069306930693\n - type: euclidean_ap\n value: 86.13858297902517\n - type: euclidean_f1\n value: 79.87679671457904\n - type: euclidean_precision\n value: 82.0675105485232\n - type: euclidean_recall\n value: 77.8\n - type: manhattan_accuracy\n value: 99.63168316831683\n - type: manhattan_ap\n value: 86.31976532265482\n - type: manhattan_f1\n value: 80.10204081632654\n - type: manhattan_precision\n value: 81.77083333333334\n - type: manhattan_recall\n value: 78.5\n - type: max_accuracy\n value: 99.65049504950495\n - type: max_ap\n value: 88.1421623503371\n - type: max_f1\n value: 81.44072036018008\n - task:\n type: Clustering\n dataset:\n name: MTEB StackExchangeClustering\n type: mteb/stackexchange-clustering\n config: default\n split: test\n revision: 6cbc1f7b2bc0622f2e39d2c77fa502909748c259\n metrics:\n - type: v_measure\n value: 68.19604139959692\n - task:\n type: Clustering\n dataset:\n name: MTEB StackExchangeClusteringP2P\n type: mteb/stackexchange-clustering-p2p\n config: default\n split: test\n revision: 815ca46b2622cec33ccafc3735d572c266efdb44\n metrics:\n - type: v_measure\n value: 36.3569584557381\n - task:\n type: Reranking\n dataset:\n name: MTEB StackOverflowDupQuestions\n type: mteb/stackoverflowdupquestions-reranking\n config: default\n split: test\n revision: e185fbe320c72810689fc5848eb6114e1ef5ec69\n metrics:\n - type: map\n value: 48.82174503355024\n - type: mrr\n value: 49.610933388506915\n - task:\n type: Summarization\n dataset:\n name: MTEB SummEval\n type: mteb/summeval\n config: default\n split: test\n revision: cda12ad7615edc362dbf25a00fdd61d3b1eaf93c\n metrics:\n - type: cos_sim_pearson\n value: 30.805895993742798\n - type: cos_sim_spearman\n value: 31.445431226826738\n - type: dot_pearson\n value: 24.441585432516867\n - type: dot_spearman\n value: 25.468117334810188\n - task:\n type: Retrieval\n dataset:\n name: MTEB TRECCOVID\n type: trec-covid\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 0.2\n - type: map_at_10\n value: 1.431\n - type: map_at_100\n value: 7.138999999999999\n - type: map_at_1000\n value: 17.933\n - type: map_at_3\n value: 0.551\n - type: map_at_5\n value: 0.7979999999999999\n - type: mrr_at_1\n value: 76.0\n - type: mrr_at_10\n value: 85.167\n - type: mrr_at_100\n value: 85.21300000000001\n - type: mrr_at_1000\n value: 85.21300000000001\n - type: mrr_at_3\n value: 84.667\n - type: mrr_at_5\n value: 85.167\n - type: ndcg_at_1\n value: 72.0\n - type: ndcg_at_10\n value: 63.343\n - type: ndcg_at_100\n value: 45.739999999999995\n - type: ndcg_at_1000\n value: 41.875\n - type: ndcg_at_3\n value: 68.162\n - type: ndcg_at_5\n value: 65.666\n - type: precision_at_1\n value: 76.0\n - type: precision_at_10\n value: 66.4\n - type: precision_at_100\n value: 46.800000000000004\n - type: precision_at_1000\n value: 18.996\n - type: precision_at_3\n value: 72.667\n - type: precision_at_5\n value: 68.4\n - type: recall_at_1\n value: 0.2\n - type: recall_at_10\n value: 1.712\n - type: recall_at_100\n value: 10.896\n - type: recall_at_1000\n value: 40.115\n - type: recall_at_3\n value: 0.594\n - type: recall_at_5\n value: 0.889\n - task:\n type: Retrieval\n dataset:\n name: MTEB Touche2020\n type: webis-touche2020\n config: default\n split: test\n revision: None\n metrics:\n - type: map_at_1\n value: 1.0619999999999998\n - type: map_at_10\n value: 5.611\n - type: map_at_100\n value: 8.841000000000001\n - type: map_at_1000\n value: 10.154\n - type: map_at_3\n value: 2.7720000000000002\n - type: map_at_5\n value: 4.181\n - type: mrr_at_1\n value: 14.285999999999998\n - type: mrr_at_10\n value: 26.249\n - type: mrr_at_100\n value: 28.046\n - type: mrr_at_1000\n value: 28.083000000000002\n - type: mrr_at_3\n value: 21.769\n - type: mrr_at_5\n value: 24.524\n - type: ndcg_at_1\n value: 11.224\n - type: ndcg_at_10\n value: 12.817\n - type: ndcg_at_100\n value: 23.183999999999997\n - type: ndcg_at_1000\n value: 35.099000000000004\n - type: ndcg_at_3\n value: 11.215\n - type: ndcg_at_5\n value: 12.016\n - type: precision_at_1\n value: 14.285999999999998\n - type: precision_at_10\n value: 12.653\n - type: precision_at_100\n value: 5.306\n - type: precision_at_1000\n value: 1.294\n - type: precision_at_3\n value: 13.605\n - type: precision_at_5\n value: 13.877999999999998\n - type: recall_at_1\n value: 1.0619999999999998\n - type: recall_at_10\n value: 10.377\n - type: recall_at_100\n value: 34.77\n - type: recall_at_1000\n value: 70.875\n - type: recall_at_3\n value: 3.688\n - type: recall_at_5\n value: 6.2509999999999994\n - task:\n type: Classification\n dataset:\n name: MTEB ToxicConversationsClassification\n type: mteb/toxic_conversations_50k\n config: default\n split: test\n revision: d7c0de2777da35d6aae2200a62c6e0e5af397c4c\n metrics:\n - type: accuracy\n value: 71.8488\n - type: ap\n value: 15.590122317097372\n - type: f1\n value: 55.86108396102662\n - task:\n type: Classification\n dataset:\n name: MTEB TweetSentimentExtractionClassification\n type: mteb/tweet_sentiment_extraction\n config: default\n split: test\n revision: d604517c81ca91fe16a244d1248fc021f9ecee7a\n metrics:\n - type: accuracy\n value: 57.61460101867573\n - type: f1\n value: 57.8678726826158\n - task:\n type: Clustering\n dataset:\n name: MTEB TwentyNewsgroupsClustering\n type: mteb/twentynewsgroups-clustering\n config: default\n split: test\n revision: 6125ec4e24fa026cec8a478383ee943acfbd5449\n metrics:\n - type: v_measure\n value: 32.01459876897588\n - task:\n type: PairClassification\n dataset:\n name: MTEB TwitterSemEval2015\n type: mteb/twittersemeval2015-pairclassification\n config: default\n split: test\n revision: 70970daeab8776df92f5ea462b6173c0b46fd2d1\n metrics:\n - type: cos_sim_accuracy\n value: 84.1032365738809\n - type: cos_sim_ap\n value: 66.60137415520323\n - type: cos_sim_f1\n value: 62.12845010615712\n - type: cos_sim_precision\n value: 62.493326214628944\n - type: cos_sim_recall\n value: 61.76781002638523\n - type: dot_accuracy\n value: 81.85015199380103\n - type: dot_ap\n value: 58.854644211365084\n - type: dot_f1\n value: 56.15180082185158\n - type: dot_precision\n value: 51.806422836752894\n - type: dot_recall\n value: 61.2928759894459\n - type: euclidean_accuracy\n value: 83.6681170650295\n - type: euclidean_ap\n value: 64.93555585305603\n - type: euclidean_f1\n value: 61.02775195857125\n - type: euclidean_precision\n value: 61.42742582197273\n - type: euclidean_recall\n value: 60.633245382585756\n - type: manhattan_accuracy\n value: 83.73368301841808\n - type: manhattan_ap\n value: 65.45422483039611\n - type: manhattan_f1\n value: 61.58552806597499\n - type: manhattan_precision\n value: 62.09763948497854\n - type: manhattan_recall\n value: 61.08179419525066\n - type: max_accuracy\n value: 84.1032365738809\n - type: max_ap\n value: 66.60137415520323\n - type: max_f1\n value: 62.12845010615712\n - task:\n type: PairClassification\n dataset:\n name: MTEB TwitterURLCorpus\n type: mteb/twitterurlcorpus-pairclassification\n config: default\n split: test\n revision: 8b6510b0b1fa4e4c4f879467980e9be563ec1cdf\n metrics:\n - type: cos_sim_accuracy\n value: 86.36628245430201\n - type: cos_sim_ap\n value: 79.29963896460292\n - type: cos_sim_f1\n value: 72.63895990066467\n - type: cos_sim_precision\n value: 69.09128803668196\n - type: cos_sim_recall\n value: 76.57068062827224\n - type: dot_accuracy\n value: 84.65091007878294\n - type: dot_ap\n value: 75.04883449222972\n - type: dot_f1\n value: 69.18569117382708\n - type: dot_precision\n value: 64.89512376070682\n - type: dot_recall\n value: 74.08376963350786\n - type: euclidean_accuracy\n value: 85.88116583226608\n - type: euclidean_ap\n value: 78.42687640324908\n - type: euclidean_f1\n value: 71.74350111107192\n - type: euclidean_precision\n value: 66.19800820152314\n - type: euclidean_recall\n value: 78.3030489682784\n - type: manhattan_accuracy\n value: 86.27508052935926\n - type: manhattan_ap\n value: 79.29581298930101\n - type: manhattan_f1\n value: 72.51838235294117\n - type: manhattan_precision\n value: 67.03921568627452\n - type: manhattan_recall\n value: 78.97289805974745\n - type: max_accuracy\n value: 86.36628245430201\n - type: max_ap\n value: 79.29963896460292\n - type: max_f1\n value: 72.63895990066467\n---\n\n\n> LLM2Vec is a simple recipe to convert decoder-only LLMs into text encoders. It consists of 3 simple steps: 1) enabling bidirectional attention, 2) masked next token prediction, and 3) unsupervised contrastive learning. The model can be further fine-tuned to achieve state-of-the-art performance.\n- **Repository:** https://github.com/McGill-NLP/llm2vec\n- **Paper:** https://arxiv.org/abs/2404.05961\n\n\n## Installation\n```bash\npip install llm2vec\n```\n\n## Usage\n```python\nfrom llm2vec import LLM2Vec\n\nimport torch\nfrom transformers import AutoTokenizer, AutoModel, AutoConfig\nfrom peft import PeftModel\n\n# Loading base Mistral model, along with custom code that enables bidirectional connections in decoder-only LLMs. MNTP LoRA weights are merged into the base model.\ntokenizer = AutoTokenizer.from_pretrained(\n \"McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp\"\n)\nconfig = AutoConfig.from_pretrained(\n \"McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp\", trust_remote_code=True\n)\nmodel = AutoModel.from_pretrained(\n \"McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp\",\n trust_remote_code=True,\n config=config,\n torch_dtype=torch.bfloat16,\n device_map=\"cuda\" if torch.cuda.is_available() else \"cpu\",\n)\nmodel = PeftModel.from_pretrained(\n model,\n \"McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp\",\n)\nmodel = model.merge_and_unload() # This can take several minutes on cpu\n\n# Loading unsupervised SimCSE model. This loads the trained LoRA weights on top of MNTP model. Hence the final weights are -- Base model + MNTP (LoRA) + SimCSE (LoRA).\nmodel = PeftModel.from_pretrained(\n model, \"McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp-unsup-simcse\"\n)\n\n# Wrapper for encoding and pooling operations\nl2v = LLM2Vec(model, tokenizer, pooling_mode=\"mean\", max_length=512)\n\n# Encoding queries using instructions\ninstruction = (\n \"Given a web search query, retrieve relevant passages that answer the query:\"\n)\nqueries = [\n [instruction, \"how much protein should a female eat\"],\n [instruction, \"summit define\"],\n]\nq_reps = l2v.encode(queries)\n\n# Encoding documents. Instruction are not required for documents\ndocuments = [\n \"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.\",\n \"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments.\",\n]\nd_reps = l2v.encode(documents)\n\n# Compute cosine similarity\nq_reps_norm = torch.nn.functional.normalize(q_reps, p=2, dim=1)\nd_reps_norm = torch.nn.functional.normalize(d_reps, p=2, dim=1)\ncos_sim = torch.mm(q_reps_norm, d_reps_norm.transpose(0, 1))\n\nprint(cos_sim)\n\"\"\"\ntensor([[0.6522, 0.1891],\n [0.1162, 0.3457]])\n\"\"\"\n```\n\n## Questions\nIf you have any question about the code, feel free to email Parishad (`parishad.behnamghader@mila.quebec`) and Vaibhav (`vaibhav.adlakha@mila.quebec`)."},"matched_task":{"kind":"list like","value":["SUMMARIZATION"],"string":"[\n \"SUMMARIZATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["BIOSSES","SCIFACT"],"string":"[\n \"BIOSSES\",\n \"SCIFACT\"\n]","isConversational":false}}},{"rowIdx":190,"cells":{"id":{"kind":"string","value":"Dampish/StellarX-4B-V0.2"},"author":{"kind":"string","value":"Dampish"},"task_category":{"kind":"string","value":"text-generation"},"tags":{"kind":"list like","value":["transformers","pytorch","gpt_neox","text-generation","arxiv:2204.06745","license:cc-by-nc-sa-4.0","autotrain_compatible","text-generation-inference","endpoints_compatible","region:us"],"string":"[\n \"transformers\",\n \"pytorch\",\n \"gpt_neox\",\n \"text-generation\",\n \"arxiv:2204.06745\",\n \"license:cc-by-nc-sa-4.0\",\n \"autotrain_compatible\",\n \"text-generation-inference\",\n \"endpoints_compatible\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2023-06-03T21:05:01","string":"2023-06-03T21:05:01"},"last_modified":{"kind":"timestamp","value":"2023-09-18T12:13:30","string":"2023-09-18T12:13:30"},"downloads":{"kind":"number","value":2264,"string":"2,264"},"likes":{"kind":"number","value":2,"string":"2"},"README":{"kind":"string","value":"---\nlicense: cc-by-nc-sa-4.0\n---\n# StellarX: A Base Model by Dampish and Arkane\n\nStellarX is a powerful autoregressive language model designed for various natural language processing tasks. It has been trained on a massive dataset containing 810 billion tokens(trained on 300B tokens), trained on \"redpajama,\" and is built upon the popular GPT-NeoX architecture. With approximately 4 billion parameters, StellarX offers exceptional performance and versatility.\n\n## Model Details\n\n- **Training Data:** StellarX is trained on a large-scale dataset provided by \"redpajama\" maintained by the group \"togethercumputer.\" This dataset has been instrumental in shaping StellarX's language capabilities and general-purpose understanding.\n- **Model Architecture:** StellarX is built upon the GPT-NeoX architecture, which may, be, inspired by GPT-3 and shares similarities with GPT-J-6B. The architecture incorporates key advancements in transformer-based language models, ensuring high-quality predictions and contextual understanding.\n- **Model Size:** StellarX consists of approximately 4 billion parameters, making it a highly capable language model for a wide range of natural language processing tasks.\n- **Carbon-Friendly and Resource-Efficient:** StellarX has been optimized for carbon efficiency and can be comfortably run on local devices. When loaded in 8 bits, the model requires only about 5GB of storage, making it more accessible and convenient for various applications.\n- **V0.2** Meaning what version it is on, currently version 0.2, Assume version 0.2 has only been trained on 300B tokens and the goal is 810B tokens. The next version aims to have a way higher accuracy.\n\n## How to Use\n\nTo load StellarX using the Hugging Face Transformers library, you can use the following code snippet:\n\n```python\nfrom transformers import AutoTokenizer, AutoModelForCausalLM\n\ntokenizer = AutoTokenizer.from_pretrained(\"Dampish/StellarX-4B-V0\")\nmodel = AutoModelForCausalLM.from_pretrained(\"Dampish/StellarX-4B-V0\")\n```\n\nThis model is particularly beneficial for those seeking a language model that is powerful, compact, and can be run on local devices without a hefty carbon footprint. Remember, when considering Darius1, it's not just about the impressive numbers—it's about what these numbers represent: powerful performance, optimized resources, and responsible computing.\n\n**For any queries related to this model, feel free to reach out to \"Dampish#3607\" on discord.**\n\n## Licensing and Usage\nStellarX, developed by the Dampish, is made available under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC-BY-NC-SA-4.0). This license ensures that you can utilize the model for research purposes and personal use without any restrictions, while also promoting the sharing and adaptation of the model under certain conditions.\n\n# Research and Personal Use\nStellarX can be freely used for research purposes, allowing you to explore its capabilities, conduct experiments, and develop novel applications. Whether you're a student, researcher, or hobbyist, the model's availability under the CC-BY-NC-SA-4.0 license empowers you to unlock the potential of StellarX for your own non-commercial projects.\n\n# Commercial Usage\nFor commercial usage of StellarX, an additional licensing arrangement must be established. If you intend to leverage the model for any commercial purpose, such as integrating it into a product or service, you are required to reach an agreement with the Dampish. This agreement will specify the terms, including the agreed-upon percentage or licensing fee to be paid for the commercial use of StellarX.\n\nTo initiate discussions regarding commercial usage, please contact Dampish through the designated channels mentioned earlier. They will be able to provide you with further information and guide you through the process of establishing a licensing arrangement tailored to your specific requirements.\n\n# Importance of Licensing Compliance\nIt is crucial to respect the licensing terms to ensure the fair usage and continued development of StellarX. The revenue generated from commercial licensing supports the efforts of the Dampish in advancing the model and making it more widely accessible.\n\n# Note on CC-BY-NC-SA-4.0\nUnder the CC-BY-NC-SA-4.0 license, you are allowed to modify and adapt StellarX, incorporating it into your own projects. However, any derivative work or modifications should also be shared under the same license terms, ensuring the continued openness and collaborative spirit of the project.\n\nPlease review the complete text of the CC-BY-NC-SA-4.0 license to familiarize yourself with its provisions and requirements. It is essential to comply with the terms of the license to respect the intellectual property rights and contributions of the Dampish and the wider community involved in developing StellarX.\n## GPT-NeoX and Model Selection\n\nGPT-NeoX-20B, a sibling model to StellarX, is a 20 billion parameter autoregressive language model trained on the Pile using the GPT-NeoX library. StellarX draws inspiration from the architectural advancements and performance of GPT-NeoX models. While the specifics of StellarX's architecture and parameters may differ, it benefits from the proven capabilities of GPT-NeoX and its suitability for diverse natural language processing tasks.\n\n## Training and Evaluation\n\nStellarX's training dataset comprises a comprehensive collection of English-language texts, covering various domains, thanks to the efforts of \"redpajama\" dataset by the group \"togethercumputer\" group.\n\nEvaluation of GPT-NeoX 20B performance has demonstrated its competence across different natural language tasks. Although since this description provides a brief summary, we refer to the GPT-NeoX Paper https://arxiv.org/abs/2204.06745, comparing GPT-NeoX 20B to other models on tasks such as OpenAI's LAMBADA, SciQ, PIQA, TriviaQA, and ARC Challenge.\n\n## Limitations and Considerations\n\nStellarX, like its sibling models, is intended primarily for research purposes. It provides a powerful foundation for extracting useful features and insights from the English language. While StellarX can be further fine-tuned and adapted for deployment, users should conduct their own risk and bias assessments before using it as a basis for downstream tasks.\n\nIt's important to note that StellarX is not intended for direct deployment without supervision. It is not designed for human-facing interactions, unlike models like ChatGPT, which have been fine-tuned using reinforcement learning from human feedback to better understand human instructions and dialogue.\n\nFurthermore, StellarX is not limited to the English language if trained properly and can sometimes be used for translation aswell as text generation in other languages.\n\nLastly, users should be aware of potential biases and limitations inherent in\n\nSpecial thanks to the group that created the training dataset. The Redpajama dataset, used to train StellarX, thank you togethercumputer.\n\n## Community and Support\n\nTo inquire about StellarX and receive support, you can join the Dampish's \nserver and engage in discussions in the #questions channel. It is recommended to explore the existing documentation and resources available for GPT-NeoX-20B to familiarize yourself with the model before seeking assistance on. For better information about GPT-NeoX, you can reach out to eleutherAI.\n\n## Summary\n\nStellarX, a base language model developed by the Dampish, offers impressive language capabilities and flexibility. Trained on an extensive dataset and built upon the GPT-NeoX architecture, StellarX excels in various natural language processing tasks. Its carbon-friendly and resource-efficient design makes it accessible for local device deployment. Researchers and enthusiasts can freely explore StellarX for research purposes and personal use, while commercial users should adhere to the licensing terms.\n\n**Again i am really grateful for the data made by togethercumputers and their willingness to opensource, they inspired this project and sparked the idea in Stellar-models, i am truly really really grateful to them.\n-dampish**\n\n\n\nDiscord: https://discord.gg/vasyNnUa \nOR Reach out to me personally on Discord via the username: Dampish#3607\n\nThank you for your time.\n\n\n\n"},"matched_task":{"kind":"list like","value":["TRANSLATION"],"string":"[\n \"TRANSLATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["SCIQ"],"string":"[\n \"SCIQ\"\n]","isConversational":false}}},{"rowIdx":191,"cells":{"id":{"kind":"string","value":"epfl-llm/meditron-70b"},"author":{"kind":"string","value":"epfl-llm"},"task_category":{"kind":"string","value":"text-generation"},"tags":{"kind":"list like","value":["transformers","pytorch","safetensors","llama","text-generation","medical","health","llama2","en","dataset:bigbio/med_qa","dataset:medmcqa","dataset:bigbio/pubmed_qa","dataset:epfl-llm/guidelines","arxiv:2311.16079","base_model:meta-llama/Llama-2-70b","base_model:finetune:meta-llama/Llama-2-70b","license:llama2","autotrain_compatible","text-generation-inference","endpoints_compatible","region:us"],"string":"[\n \"transformers\",\n \"pytorch\",\n \"safetensors\",\n \"llama\",\n \"text-generation\",\n \"medical\",\n \"health\",\n \"llama2\",\n \"en\",\n \"dataset:bigbio/med_qa\",\n \"dataset:medmcqa\",\n \"dataset:bigbio/pubmed_qa\",\n \"dataset:epfl-llm/guidelines\",\n \"arxiv:2311.16079\",\n \"base_model:meta-llama/Llama-2-70b\",\n \"base_model:finetune:meta-llama/Llama-2-70b\",\n \"license:llama2\",\n \"autotrain_compatible\",\n \"text-generation-inference\",\n \"endpoints_compatible\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2023-11-08T13:57:04","string":"2023-11-08T13:57:04"},"last_modified":{"kind":"timestamp","value":"2023-12-07T19:39:04","string":"2023-12-07T19:39:04"},"downloads":{"kind":"number","value":2262,"string":"2,262"},"likes":{"kind":"number","value":234,"string":"234"},"README":{"kind":"string","value":"---\nbase_model: meta-llama/Llama-2-70b\ndatasets:\n- bigbio/med_qa\n- medmcqa\n- bigbio/pubmed_qa\n- epfl-llm/guidelines\nlanguage:\n- en\nlicense: llama2\nmetrics:\n- accuracy\n- perplexity\npipeline_tag: text-generation\ntags:\n- medical\n- health\n- llama2\n---\n\n<img width=50% src=\"meditron_LOGO.png\" alt=\"Alt text\" title=\"Meditron-logo\">\n\n# Model Card for Meditron-70B-v1.0\n\nMeditron is a suite of open-source medical Large Language Models (LLMs).\nMeditron-70B is a 70 billion parameters model adapted to the medical domain from Llama-2-70B through continued pretraining on a comprehensively curated medical corpus, including selected PubMed articles, abstracts, a [new dataset](https://huggingface.co/datasets/epfl-llm/guidelines) of internationally-recognized medical guidelines, and general domain data from [RedPajama-v1](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T).\nMeditron-70B, finetuned on relevant training data, outperforms Llama-2-70B, GPT-3.5 (`text-davinci-003`, 8-shot), and Flan-PaLM on multiple medical reasoning tasks.\n\n\n<!--# Table of Contents\n\n [Model Card for Meditron 70B](#model-card-for--meditron-70b-v1.0)\n- [Table of Contents](#table-of-contents)\n- [Model Details](#model-details)\n - [Model Description](#model-description)\n- [Uses](#uses)\n - [Downstream Use](#downstream-use)\n - [Out-of-Scope Use](#out-of-scope-use)\n- [Bias, Risks, and Limitations](#bias-risks-and-limitations)\n - [Recommendations](#recommendations)\n- [Training Details](#training-details)\n - [Training Data](#training-data)\n - [Training Procedure](#training-procedure)\n - [Preprocessing](#preprocessing)\n- [Evaluation](#evaluation)\n - [Testing Data & Metrics](#testing-data-&-metrics)\n - [Testing Data](#testing-data)\n - [Metrics](#metrics)\n - [Results](#results)\n- [Environmental Impact](#environmental-impact)\n- [Citation](#citation)-->\n\n<details open>\n <summary><strong>Advisory Notice</strong></summary>\n\n <blockquote style=\"padding: 10px; margin: 0 0 10px; border-left: 5px solid #ddd;\">\n While Meditron is designed to encode medical knowledge from sources of high-quality evidence, it is not yet adapted to deliver this knowledge appropriately, safely, or within professional actionable constraints. \n We recommend against deploying Meditron in medical applications without extensive use-case alignment, as well as additional testing, specifically including randomized controlled trials in real-world practice settings.\n </blockquote>\n</details>\n\n## Model Details\n\n- **Developed by:** [EPFL LLM Team](https://huggingface.co/epfl-llm)\n- **Model type:** Causal decoder-only transformer language model\n- **Language(s):** English (mainly)\n- **Model License:** [LLAMA 2 COMMUNITY LICENSE AGREEMENT](https://huggingface.co/meta-llama/Llama-2-70b/raw/main/LICENSE.txt)\n- **Code License:** [APACHE 2.0 LICENSE](LICENSE)\n- **Continue-pretrained from model:** [Llama-2-70B](https://huggingface.co/meta-llama/Llama-2-70b)\n- **Context length:** 4K tokens\n- **Input:** Text-only data\n- **Output:** Model generates text only\n- **Status:** This is a static model trained on an offline dataset. Future versions of the tuned models will be released as we enhance model's performance.\n- **Knowledge Cutoff:** August 2023\n\n\n### Model Sources\n\n- **Repository:** [epflLLM/meditron](https://github.com/epfLLM/meditron)\n- **Trainer:** [epflLLM/Megatron-LLM](https://github.com/epfLLM/Megatron-LLM)\n- **Paper:** *[MediTron-70B: Scaling Medical Pretraining for Large Language Models](https://arxiv.org/abs/2311.16079)*\n\n## Uses\n\nMeditron-70B is being made available for further testing and assessment as an AI assistant to enhance clinical decision-making and enhance access to an LLM for healthcare use. Potential use cases may include but are not limited to:\n- Medical exam question answering\n- Supporting differential diagnosis\n- Disease information (symptoms, cause, treatment) query\n- General health information query\n\n### Direct Use\n\nIt is possible to use this model to generate text, which is useful for experimentation and understanding its capabilities. \nIt should not be used directly for production or work that may impact people.\n\n### Downstream Use\nMeditron-70B and Meditron-7B are both foundation models without finetuning or instruction-tuning. They can be finetuned, instruction-tuned, or RLHF-tuned for specific downstream tasks and applications.\nThere are two ways we have used this model for downstream question-answering tasks.\n1. We apply in-context learning with k demonstrations (3 or 5 in our paper) added to the prompt.\n2. We finetuned the models for downstream question-answering tasks using specific training sets.\n\nWe encourage and look forward to the adaption of the base model for more diverse applications.\n\nIf you want a more interactive way to prompt the model, we recommend using a high-throughput and memory-efficient inference engine with a UI that supports chat and text generation.\n\nYou can check out our deployment [guide](https://github.com/epfLLM/meditron/blob/main/deployment/README.md), where we used [FastChat](https://github.com/lm-sys/FastChat) with [vLLM](https://github.com/vllm-project/vllm). We collected generations for our qualitative analysis through an interactive UI platform, [BetterChatGPT](https://github.com/ztjhz/BetterChatGPT). Here is the prompt format we used as an example:\n\n<img width=70% src=\"prompt_example.png\" alt=\"qualitative-analysis-prompt\" title=\"Qualitative Analysis Prompt\">\n\n### Out-of-Scope Use\n\nWe do not recommend using this model for natural language generation in a production environment, finetuned or otherwise.\n\n## Truthfulness, Helpfulness, Risk, and Bias\n\n<!-- This section is meant to convey both technical and sociotechnical limitations. -->\n\nWe did an initial assessment of Meditron models' **Truthfulness** against baseline models and consumer-level medical models. \nWe use TruthfulQA (multiple choice) as the main evaluation benchmark.\nWe only focus on the categories that are relevant to the medical domain, including Health, Nutrition, Psychology, and Science.\nFor 7B models, we perform one-shot evaluations for consistent answer generation.\nFor 70B models, the evaluations are under the zero-shot setting.\nBelow, we report the detailed truthfulness performance of each category.\n\n| | | | | | | | |\n| --- | ------ |----- |----- |----- |----- |----- |----- |\n|Category | meditron-70b | llama-2-70b | med42-70b* | meditron-7b | llama-2-7b | PMC-llama-7b |\n|Health | 81.8 | 69.1 | 83.6 | 27.3 | 16.4 | 3.6 |\n|Nutrition | 77.9 | 68.8 | 62.5 | 31.1 | 12.5 | 6.3 |\n|Psychology| 47.4 | 36.8 | 52.6 | 21.1 | 10.5 | 0.0 |\n|Science | 77.8 | 44.4 | 33.3 | 33.3 | 11.1 | 0.0 |\n|Avg | 71.2 | 54.8 | 58.0 | 28.3 | 12.6 | 2.5 |\n| | | | | | | |\n\nFor a more detailed performance analysis, please see our paper. \n\nFor **Helpfulness**, **Risk** and **Bias**, we provide a comprehensive qualitative generation report of Meditron-70B on queries designed by medical experts.\nEach query targets specific aspects of helpfulness (medical accuracy, up-to-date information, etc.), risk (public health, medical ethics, etc.) and bias (gender, age, race, etc.).\nPlease see the detailed generations in our paper. We compare our generations to Llama-2-70B and ChatGPT-3.5 (version Nov, 27, 2023)\n\nSignificant research is still required to fully explore potential bias, fairness, and safety issues with this language model. \n\n\n### Recommendations\n\n**IMPORTANT!**\nUsers (both direct and downstream) should be made aware of the risks, biases, and limitations of the model.\nWhile this model is capable of generating natural language text, we have only begun to explore this capability and its limitations. \nUnderstanding these limitations is especially important in a domain like medicine. \nTherefore, we strongly recommend against using this model in production for natural language generation or for professional purposes related to health and medicine without comprehensive testing for your application. \n\n## Training Details\n\n### Training Data\nMeditron’s domain-adaptive pre-training corpus GAP-Replay combines 48.1B tokens from four corpora:\n- [**Clinical Guidelines**](https://huggingface.co/datasets/epfl-llm/guidelines): a new dataset of 46K internationally-recognized clinical practice guidelines from various healthcare-related sources, including hospitals and international organizations. \n- **Medical Paper Abstracts**: 16.1M abstracts extracted from closed-access PubMed and PubMed Central papers.\n- **Medical Papers**: full-text articles extracted from 5M publicly available PubMed and PubMed Central papers.\n- **Replay Data**: 400M tokens of general domain pretraining data sampled from [RedPajama-v1](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-1T)\n\n\n<img width=\"60%\" src=\"gap-replay.png\" alt=\"Alt text\" title=\"Meditron-logo\">\n\n\n#### Data Preprocessing\n\nPlease see the detailed preprocessing procedure in our paper. \n\n### Training Procedure \n\nWe used the [Megatron-LLM](https://github.com/epfLLM/Megatron-LLM) distributed training library, a derivative of Nvidia's Megatron LM project, to optimize training efficiency. \nHardware consists of 16 nodes of 8x NVIDIA A100 (80GB) SXM GPUs connected by NVLink and NVSwitch with a single Nvidia ConnectX-6 DX network card and equipped with 2 x AMD EPYC 7543 32-Core Processors and 512 GB of RAM. \nThe nodes are connected via RDMA over Converged Ethernet.\n\nOur three-way parallelism scheme uses:\n - Data Parallelism (DP -- different GPUs process different subsets of the batches) of 2,\n - Pipeline Parallelism (PP -- different GPUs process different layers) of 8,\n - Tensor Parallelism (TP -- different GPUs process different subtensors for matrix multiplication) of 8.\n \n\n#### Training Hyperparameters\n\n| | |\n| --- | ------ |\n| bf16 | true |\n| lr | 1.5e-4 |\n| eps | 1e-5 |\n| betas | \\[0.9, 0.95\\] |\n| clip_grad | 1 |\n| weight decay | 0.1 |\n| DP size | 2 |\n| TP size | 8 |\n| PP size | 8 |\n| seq length | 4096 |\n| lr scheduler | cosine|\n| min lr | 1e-6 |\n| warmup iteration | 2000 |\n| micro batch size | 2 |\n| global batch size | 512 |\n| | |\n\n#### Speeds, Sizes, Times\nThe model was trained in September and October 2023.\n\nThe model architecture is exactly Llama 2, meaning\n\n| | | \n| --- | ------ |\n| Model size | 70B | \n| Hidden dimension | 8192 |\n| Num. attention heads | 64 |\n| Num. layers | 80 |\n| | | | \n\nWe train the 70B model on 48e9 tokens, at a throughput of about 40,200 tokens / second. \nThis amounts to a bfloat16 model flops utilization of roughly 42.3\\%.\n\n## Evaluation\n\n<!-- This section describes the evaluation protocols and provides the results. -->\n\n### Testing Data & Metrics\n\n#### Testing Data\n- [MedQA (USMLE)](https://huggingface.co/datasets/bigbio/med_qa)\n- [MedMCQA](https://huggingface.co/datasets/medmcqa)\n- [PubMedQA](https://huggingface.co/datasets/bigbio/pubmed_qa)\n- [MMLU-Medical](https://huggingface.co/datasets/lukaemon/mmlu)\n- [MedQA-4-Option](https://huggingface.co/datasets/GBaker/MedQA-USMLE-4-options)\n\n#### Metrics\n- Accuracy: suite the evaluation of multiple-choice question-answering tasks.\n\n### Results\nWe finetune meditron-70b and llama-2-70b on each benchmark (pubmedqa, medmcqa, medqa)'s training data individually. \nWe report the finetuned models' performance with self-consistency chain-of-thought as the inference mode.\nFor MMLU-Medical, models finetuned on MedMCQA are used for inference.\nFor MedQA-4-Option, models finetuned on MedQA are used for inference.\nFor a more detailed performance analysis, please see our paper. \n\n| | | | | | |\n| --- | ------ |----- |----- |----- |----- |\n|Dataset| meditron-70b | llama-2-70b | med42-70b* | clinical-camel-70b* |\n|MMLU-Medical | 77.6 | 77.9 | 74.5 | 65.7 |\n|PubMedQA | 81.6 | 80.0 | 61.2 | 67.0 |\n|MedMCQA | 66.0 | 62.6 | 59.2 | 46.7 |\n|MedQA | 64.4 | 61.5 | 59.1 | 50.8 |\n|MedQA-4-Option| 70.2 | 63.8 | 63.9 | 56.8 |\n|Avg | 72.0 | 69.2 | 63.6 | 57.4 |\n| | | | | | |\n\n**Note**: models with * are already instruction-tuned, so we exclude them from further finetuning on any training data.\n\n## Environmental Impact\n\n<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->\n\n- **Hardware Type:** 128 x NVIDIA A100 (80GB) SXM\n- **Total GPU hours:** 42,496\n- **Hardware Provider:** EPFL Research Computing Platform\n- **Compute Region:** Switzerland\n- **Carbon Emitted:** Switzerland has a carbon efficiency of 0.016 kgCO2/kWh (https://www.carbonfootprint.com/docs/2018_8_electricity_factors_august_2018_-_online_sources.pdf). 332 hours of 128 A100s means 42496 hours at a TDP of 400W. Assuming a Power Usage effectiveness of 1.8, total emissions are estimated to be: \n \n (400W / 1000W/kWh / GPU * 0.016 kgCO2/kWh * 332 h * 128 GPU) * 1.8 PUE = 486 kgCO2.\n\n## Citation\n\n**BibTeX:**\nIf you use Meditron or its training data, please cite our work:\n\n```\n@misc{chen2023meditron70b,\n title={MEDITRON-70B: Scaling Medical Pretraining for Large Language Models}, \n author={Zeming Chen and Alejandro Hernández-Cano and Angelika Romanou and Antoine Bonnet and Kyle Matoba and Francesco Salvi and Matteo Pagliardini and Simin Fan and Andreas Köpf and Amirkeivan Mohtashami and Alexandre Sallinen and Alireza Sakhaeirad and Vinitra Swamy and Igor Krawczuk and Deniz Bayazit and Axel Marmet and Syrielle Montariol and Mary-Anne Hartley and Martin Jaggi and Antoine Bosselut},\n year={2023},\n eprint={2311.16079},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n\n@software{epfmedtrn,\n author = {Zeming Chen and Alejandro Hernández Cano and Angelika Romanou and Antoine Bonnet and Kyle Matoba and Francesco Salvi and Matteo Pagliardini and Simin Fan and Andreas Köpf and Amirkeivan Mohtashami and Alexandre Sallinen and Alireza Sakhaeirad and Vinitra Swamy and Igor Krawczuk and Deniz Bayazit and Axel Marmet and Syrielle Montariol and Mary-Anne Hartley and Martin Jaggi and Antoine Bosselut},\n title = {MediTron-70B: Scaling Medical Pretraining for Large Language Models},\n month = November,\n year = 2023,\n url = {https://github.com/epfLLM/meditron}\n}\n```"},"matched_task":{"kind":"list like","value":["QUESTION_ANSWERING"],"string":"[\n \"QUESTION_ANSWERING\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["MEDQA","PUBMEDQA"],"string":"[\n \"MEDQA\",\n \"PUBMEDQA\"\n]","isConversational":false}}},{"rowIdx":192,"cells":{"id":{"kind":"string","value":"Dampish/StellarX-4B-V0"},"author":{"kind":"string","value":"Dampish"},"task_category":{"kind":"string","value":"text-generation"},"tags":{"kind":"list like","value":["transformers","pytorch","gpt_neox","text-generation","arxiv:2204.06745","license:cc-by-nc-sa-4.0","autotrain_compatible","text-generation-inference","endpoints_compatible","region:us"],"string":"[\n \"transformers\",\n \"pytorch\",\n \"gpt_neox\",\n \"text-generation\",\n \"arxiv:2204.06745\",\n \"license:cc-by-nc-sa-4.0\",\n \"autotrain_compatible\",\n \"text-generation-inference\",\n \"endpoints_compatible\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2023-05-27T19:04:49","string":"2023-05-27T19:04:49"},"last_modified":{"kind":"timestamp","value":"2023-12-03T19:52:22","string":"2023-12-03T19:52:22"},"downloads":{"kind":"number","value":2229,"string":"2,229"},"likes":{"kind":"number","value":1,"string":"1"},"README":{"kind":"string","value":"---\nlicense: cc-by-nc-sa-4.0\n---\n# StellarX: A Base Model by Dampish and Arkane\n\nStellarX is a powerful autoregressive language model designed for various natural language processing tasks. It has been trained on a massive dataset containing 810 billion tokens, trained on \"redpajama,\" and is built upon the popular GPT-NeoX architecture. With approximately 4 billion parameters, StellarX offers exceptional performance and versatility.\n\n## Model Details\n\n- **Training Data:** StellarX is trained on a large-scale dataset provided by \"redpajama\" maintained by the group \"togethercumputer.\" This dataset has been instrumental in shaping StellarX's language capabilities and general-purpose understanding.\n- **Model Architecture:** StellarX is built upon the GPT-NeoX architecture, which may, be, inspired by GPT-3 and shares similarities with GPT-J-6B. The architecture incorporates key advancements in transformer-based language models, ensuring high-quality predictions and contextual understanding.\n- **Model Size:** StellarX consists of approximately 4 billion parameters, making it a highly capable language model for a wide range of natural language processing tasks.\n- **Carbon-Friendly and Resource-Efficient:** StellarX has been optimized for carbon efficiency and can be comfortably run on local devices. When loaded in 8 bits, the model requires only about 5GB of storage, making it more accessible and convenient for various applications.\n- **V0** Meaning what version it is on, currently version 0, Assume version 0 has only been trained on 300B tokens and the goal is 810B tokens. The next version aims to have a way higher accuracy.\n\n## How to Use\n\nTo load StellarX using the Hugging Face Transformers library, you can use the following code snippet:\n\n```python\nfrom transformers import AutoTokenizer, AutoModelForCausalLM\n\ntokenizer = AutoTokenizer.from_pretrained(\"Dampish/StellarX-4B-V0\")\nmodel = AutoModelForCausalLM.from_pretrained(\"Dampish/StellarX-4B-V0\")\n```\n\nThis model is particularly beneficial for those seeking a language model that is powerful, compact, and can be run on local devices without a hefty carbon footprint. Remember, when considering Darius1, it's not just about the impressive numbers—it's about what these numbers represent: powerful performance, optimized resources, and responsible computing.\n\n**For any queries related to this model, feel free to reach out to \"Dampish#3607\" on discord.**\n\n## Licensing and Usage\nStellarX, developed by the Dampish, is made available under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC-BY-NC-SA-4.0). This license ensures that you can utilize the model for research purposes and personal use without any restrictions, while also promoting the sharing and adaptation of the model under certain conditions.\n\n# Research and Personal Use\nStellarX can be freely used for research purposes, allowing you to explore its capabilities, conduct experiments, and develop novel applications. Whether you're a student, researcher, or hobbyist, the model's availability under the CC-BY-NC-SA-4.0 license empowers you to unlock the potential of StellarX for your own non-commercial projects.\n\n# Commercial Usage\nFor commercial usage of StellarX, an additional licensing arrangement must be established. If you intend to leverage the model for any commercial purpose, such as integrating it into a product or service, you are required to reach an agreement with the Dampish. This agreement will specify the terms, including the agreed-upon percentage or licensing fee to be paid for the commercial use of StellarX.\n\nTo initiate discussions regarding commercial usage, please contact Dampish through the designated channels mentioned earlier. They will be able to provide you with further information and guide you through the process of establishing a licensing arrangement tailored to your specific requirements.\n\n# Importance of Licensing Compliance\nIt is crucial to respect the licensing terms to ensure the fair usage and continued development of StellarX. The revenue generated from commercial licensing supports the efforts of the Dampish in advancing the model and making it more widely accessible.\n\n# Note on CC-BY-NC-SA-4.0\nUnder the CC-BY-NC-SA-4.0 license, you are allowed to modify and adapt StellarX, incorporating it into your own projects. However, any derivative work or modifications should also be shared under the same license terms, ensuring the continued openness and collaborative spirit of the project.\n\nPlease review the complete text of the CC-BY-NC-SA-4.0 license to familiarize yourself with its provisions and requirements. It is essential to comply with the terms of the license to respect the intellectual property rights and contributions of the Dampish and the wider community involved in developing StellarX.\n## GPT-NeoX and Model Selection\n\nGPT-NeoX-20B, a sibling model to StellarX, is a 20 billion parameter autoregressive language model trained on the Pile using the GPT-NeoX library. StellarX draws inspiration from the architectural advancements and performance of GPT-NeoX models. While the specifics of StellarX's architecture and parameters may differ, it benefits from the proven capabilities of GPT-NeoX and its suitability for diverse natural language processing tasks.\n\n## Training and Evaluation\n\nStellarX's training dataset comprises a comprehensive collection of English-language texts, covering various domains, thanks to the efforts of \"redpajama\" dataset by the group \"togethercumputer\" group.\n\nEvaluation of GPT-NeoX 20B performance has demonstrated its competence across different natural language tasks. Although since this description provides a brief summary, we refer to the GPT-NeoX Paper https://arxiv.org/abs/2204.06745, comparing GPT-NeoX 20B to other models on tasks such as OpenAI's LAMBADA, SciQ, PIQA, TriviaQA, and ARC Challenge.\n\n## Limitations and Considerations\n\nStellarX, like its sibling models, is intended primarily for research purposes. It provides a powerful foundation for extracting useful features and insights from the English language. While StellarX can be further fine-tuned and adapted for deployment, users should conduct their own risk and bias assessments before using it as a basis for downstream tasks.\n\nIt's important to note that StellarX is not intended for direct deployment without supervision. It is not designed for human-facing interactions, unlike models like ChatGPT, which have been fine-tuned using reinforcement learning from human feedback to better understand human instructions and dialogue.\n\nFurthermore, StellarX is not limited to the English language if trained properly and can sometimes be used for translation aswell as text generation in other languages.\n\nLastly, users should be aware of potential biases and limitations inherent in\n\nSpecial thanks to the group that created the training dataset. The Redpajama dataset, used to train StellarX, thank you togethercumputer.\n\n## Community and Support\n\nTo inquire about StellarX and receive support, you can join the Dampish's \nserver and engage in discussions in the #questions channel. It is recommended to explore the existing documentation and resources available for GPT-NeoX-20B to familiarize yourself with the model before seeking assistance on. For better information about GPT-NeoX, you can reach out to eleutherAI.\n\n## Summary\n\nStellarX, a base language model developed by the Dampish, offers impressive language capabilities and flexibility. Trained on an extensive dataset and built upon the GPT-NeoX architecture, StellarX excels in various natural language processing tasks. Its carbon-friendly and resource-efficient design makes it accessible for local device deployment. Researchers and enthusiasts can freely explore StellarX for research purposes and personal use, while commercial users should adhere to the licensing terms.\n\n**Again i am really grateful for the data made by togethercumputers and their willingness to opensource, they inspired this project and sparked the idea in Stellar-models, i am truly really really grateful to them.\n-dampish**\n\n\n\nDiscord: https://discord.gg/vasyNnUa \nOR Reach out to me personally on Discord via the username: Dampish#3607\n\nThank you for your time.\n\n\n\n\n# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)\nDetailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_Dampish__StellarX-4B-V0)\n\n| Metric | Value |\n|-----------------------|---------------------------|\n| Avg. | 33.54 |\n| ARC (25-shot) | 36.95 |\n| HellaSwag (10-shot) | 61.9 |\n| MMLU (5-shot) | 26.85 |\n| TruthfulQA (0-shot) | 34.3 |\n| Winogrande (5-shot) | 63.85 |\n| GSM8K (5-shot) | 0.0 |\n| DROP (3-shot) | 10.95 |\n"},"matched_task":{"kind":"list like","value":["TRANSLATION"],"string":"[\n \"TRANSLATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["SCIQ"],"string":"[\n \"SCIQ\"\n]","isConversational":false}}},{"rowIdx":193,"cells":{"id":{"kind":"string","value":"BSC-LT/salamandra-7b"},"author":{"kind":"string","value":"BSC-LT"},"task_category":{"kind":"string","value":"text-generation"},"tags":{"kind":"list like","value":["transformers","safetensors","llama","text-generation","bg","ca","code","cs","cy","da","de","el","en","es","et","eu","fi","fr","ga","gl","hr","hu","it","lt","lv","mt","nl","nn","oc","pl","pt","ro","ru","sh","sk","sl","sr","sv","uk","dataset:oscar-corpus/colossal-oscar-1.0","dataset:HuggingFaceFW/fineweb-edu","dataset:joelniklaus/eurlex_resources","dataset:joelito/legal-mc4","dataset:projecte-aina/CATalog","dataset:UFRGS/brwac","dataset:community-datasets/hrwac","dataset:danish-foundation-models/danish-gigaword","dataset:HiTZ/euscrawl","dataset:PleIAs/French-PD-Newspapers","dataset:PleIAs/French-PD-Books","dataset:AI-team-UoA/greek_legal_code","dataset:HiTZ/latxa-corpus-v1.1","dataset:allenai/peS2o","dataset:pile-of-law/pile-of-law","dataset:PORTULAN/parlamento-pt","dataset:hoskinson-center/proof-pile","dataset:togethercomputer/RedPajama-Data-1T","dataset:bigcode/starcoderdata","dataset:bjoernp/tagesschau-2018-2023","dataset:EleutherAI/the_pile_deduplicated","arxiv:2502.08489","arxiv:2403.14009","arxiv:2403.20266","arxiv:2101.00027","arxiv:2207.00220","arxiv:1810.06694","arxiv:1911.05507","arxiv:1906.03741","arxiv:2406.17557","arxiv:2402.06619","arxiv:1803.09010","license:apache-2.0","autotrain_compatible","text-generation-inference","endpoints_compatible","region:us"],"string":"[\n \"transformers\",\n \"safetensors\",\n \"llama\",\n \"text-generation\",\n \"bg\",\n \"ca\",\n \"code\",\n \"cs\",\n \"cy\",\n \"da\",\n \"de\",\n \"el\",\n \"en\",\n \"es\",\n \"et\",\n \"eu\",\n \"fi\",\n \"fr\",\n \"ga\",\n \"gl\",\n \"hr\",\n \"hu\",\n \"it\",\n \"lt\",\n \"lv\",\n \"mt\",\n \"nl\",\n \"nn\",\n \"oc\",\n \"pl\",\n \"pt\",\n \"ro\",\n \"ru\",\n \"sh\",\n \"sk\",\n \"sl\",\n \"sr\",\n \"sv\",\n \"uk\",\n \"dataset:oscar-corpus/colossal-oscar-1.0\",\n \"dataset:HuggingFaceFW/fineweb-edu\",\n \"dataset:joelniklaus/eurlex_resources\",\n \"dataset:joelito/legal-mc4\",\n \"dataset:projecte-aina/CATalog\",\n \"dataset:UFRGS/brwac\",\n \"dataset:community-datasets/hrwac\",\n \"dataset:danish-foundation-models/danish-gigaword\",\n \"dataset:HiTZ/euscrawl\",\n \"dataset:PleIAs/French-PD-Newspapers\",\n \"dataset:PleIAs/French-PD-Books\",\n \"dataset:AI-team-UoA/greek_legal_code\",\n \"dataset:HiTZ/latxa-corpus-v1.1\",\n \"dataset:allenai/peS2o\",\n \"dataset:pile-of-law/pile-of-law\",\n \"dataset:PORTULAN/parlamento-pt\",\n \"dataset:hoskinson-center/proof-pile\",\n \"dataset:togethercomputer/RedPajama-Data-1T\",\n \"dataset:bigcode/starcoderdata\",\n \"dataset:bjoernp/tagesschau-2018-2023\",\n \"dataset:EleutherAI/the_pile_deduplicated\",\n \"arxiv:2502.08489\",\n \"arxiv:2403.14009\",\n \"arxiv:2403.20266\",\n \"arxiv:2101.00027\",\n \"arxiv:2207.00220\",\n \"arxiv:1810.06694\",\n \"arxiv:1911.05507\",\n \"arxiv:1906.03741\",\n \"arxiv:2406.17557\",\n \"arxiv:2402.06619\",\n \"arxiv:1803.09010\",\n \"license:apache-2.0\",\n \"autotrain_compatible\",\n \"text-generation-inference\",\n \"endpoints_compatible\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2024-09-30T06:47:04","string":"2024-09-30T06:47:04"},"last_modified":{"kind":"timestamp","value":"2025-02-20T16:42:39","string":"2025-02-20T16:42:39"},"downloads":{"kind":"number","value":2229,"string":"2,229"},"likes":{"kind":"number","value":27,"string":"27"},"README":{"kind":"string","value":"---\ndatasets:\n- oscar-corpus/colossal-oscar-1.0\n- HuggingFaceFW/fineweb-edu\n- joelniklaus/eurlex_resources\n- joelito/legal-mc4\n- projecte-aina/CATalog\n- UFRGS/brwac\n- community-datasets/hrwac\n- danish-foundation-models/danish-gigaword\n- HiTZ/euscrawl\n- PleIAs/French-PD-Newspapers\n- PleIAs/French-PD-Books\n- AI-team-UoA/greek_legal_code\n- HiTZ/latxa-corpus-v1.1\n- allenai/peS2o\n- pile-of-law/pile-of-law\n- PORTULAN/parlamento-pt\n- hoskinson-center/proof-pile\n- togethercomputer/RedPajama-Data-1T\n- bigcode/starcoderdata\n- bjoernp/tagesschau-2018-2023\n- EleutherAI/the_pile_deduplicated\nlanguage:\n- bg\n- ca\n- code\n- cs\n- cy\n- da\n- de\n- el\n- en\n- es\n- et\n- eu\n- fi\n- fr\n- ga\n- gl\n- hr\n- hu\n- it\n- lt\n- lv\n- mt\n- nl\n- nn\n- \\no\n- oc\n- pl\n- pt\n- ro\n- ru\n- sh\n- sk\n- sl\n- sr\n- sv\n- uk\nlibrary_name: transformers\nlicense: apache-2.0\npipeline_tag: text-generation\n---\n\n\n\n# Salamandra Model Card\n\nThis repository contains the model described in [Salamandra Technical Report](https://huggingface.co/papers/2502.08489).\n\nSalamandra is a highly multilingual model pre-trained from scratch that comes in three different \nsizes — 2B, 7B and 40B parameters — with their respective base and instruction-tuned variants. \nThis model card corresponds to the 7B base version.\n\nTo visit the model cards of other Salamandra versions, please refer to the [Model Index](#model-index).\n\nThe entire Salamandra family is released under a permissive [Apache 2.0 license](https://www.apache.org/licenses/LICENSE-2.0).\nAlong with the open weights, all training scripts and configuration files are made publicly available in [this GitHub repository](https://github.com/langtech-bsc/salamandra).\n\n---\n\n## Model Details\n\n### Description\n\nTransformer-based decoder-only language model that has been pre-trained from scratch on 12.875 trillion tokens of highly curated data.\nThe pre-training corpus contains text in 35 European languages and code.\n\n### Hyperparameters\n\nThe full list of hyperparameters for each model can be found [here](https://github.com/langtech-bsc/salamandra/tree/main/configs).\n\n### Architecture\n\n| | |\n|-------------------------|:--------------|\n| Total Parameters | 7,768,117,248 |\n| Embedding Parameters | 1,048,576,000 |\n| Layers | 32 |\n| Hidden size | 4,096 |\n| Attention heads | 32 |\n| Context length | 8,192 |\n| Vocabulary size | 256,000 |\n| Precision | bfloat16 |\n| Embedding type | RoPE |\n| Activation Function | SwiGLU |\n| Layer normalization | RMS Norm |\n| Flash attention | ✅ |\n| Grouped Query Attention | ✅ |\n| Num. query groups | 8 |\n\n---\n\n## Intended Use\n\n### Direct Use\n\nThe models are intended for both research and commercial use in any of the languages included in the training data. \nThe base models are intended either for language generation or to be further fine-tuned for specific use-cases. \nThe instruction-tuned variants can be used as general-purpose assistants, as long as the user is fully aware of the model’s limitations.\n\n### Out-of-scope Use\n\nThe model is not intended for malicious activities, such as harming others or violating human rights. \nAny downstream application must comply with current laws and regulations. \nIrresponsible usage in production environments without proper risk assessment and mitigation is also discouraged. \n\n---\n\n## Hardware and Software\n\n### Training Framework\n\nPre-training was conducted using NVIDIA’s [NeMo Framework](https://docs.nvidia.com/nemo-framework/index.html), \nwhich leverages PyTorch Lightning for efficient model training in highly distributed settings.\n\nThe instruction-tuned versions were produced with [FastChat](https://github.com/lm-sys/FastChat).\n\n### Compute Infrastructure\n\nAll models were trained on [MareNostrum 5](https://www.bsc.es/ca/marenostrum/marenostrum-5), a pre-exascale EuroHPC supercomputer hosted and\noperated by Barcelona Supercomputing Center.\n\nThe accelerated partition is composed of 1,120 nodes with the following specifications:\n- 4x Nvidia Hopper GPUs with 64 HBM2 memory\n- 2x Intel Sapphire Rapids 8460Y+ at 2.3Ghz and 32c each (64 cores)\n- 4x NDR200 (BW per node 800Gb/s)\n- 512 GB of Main memory (DDR5)\n- 460GB on NVMe storage\n\n|Model|Nodes|GPUs|\n|:---:|:---:|:---:|\n|2B|64|256|\n|7B|128|512|\n|40B|256 / 512|1,024 / 2,048|\n\n---\n\n## How to use\nThis section offers examples of how to perform inference using various methods.\n\n### Inference\nYou'll find different techniques for running inference, including Huggingface's Text Generation Pipeline, multi-GPU configurations, and vLLM for scalable and efficient generation. \n\n#### Inference with Huggingface's Text Generation Pipeline\nThe Huggingface Text Generation Pipeline provides a straightforward way to run inference using the Salamandra-7b model.\n\n```bash\npip install transformers torch accelerate sentencepiece protobuf\n```\n<details>\n<summary>Show code</summary>\n\n```python\nfrom transformers import pipeline, set_seed\n\nmodel_id = \"BSC-LT/salamandra-7b\"\n\n# Sample prompts\nprompts = [\n \"Las fiestas de San Isidro Labrador de Yecla son\",\n \"El punt més alt del Parc Natural del Montseny és\",\n \"Sentence in English: The typical chance of such a storm is around 10%. Sentence in Catalan:\",\n \"Si le monde était clair\",\n \"The future of AI is\",\n]\n\n# Create the pipeline\ngenerator = pipeline(\"text-generation\", model_id, device_map=\"auto\")\ngeneration_args = {\n \"temperature\": 0.1,\n \"top_p\": 0.95,\n \"max_new_tokens\": 25,\n \"repetition_penalty\": 1.2,\n \"do_sample\": True\n}\n\n# Fix the seed\nset_seed(1)\n# Generate texts\noutputs = generator(prompts, **generation_args)\n# Print outputs\nfor output in outputs:\n print(output[0][\"generated_text\"])\n\n```\n</details>\n\n#### Inference with single / multi GPU\nThis section provides a simple example of how to run inference using Huggingface's AutoModel class.\n\n```bash\npip install transformers torch accelerate sentencepiece protobuf\n```\n\n<details>\n<summary>Show code</summary>\n\n```python\nfrom transformers import AutoTokenizer, AutoModelForCausalLM\nimport torch\n\nmodel_id = \"BSC-LT/salamandra-7b\"\n\n# Input text\ntext = \"El mercat del barri és\"\n\n# Load the tokenizer\ntokenizer = AutoTokenizer.from_pretrained(model_id)\n# Load the model\nmodel = AutoModelForCausalLM.from_pretrained(\n model_id,\n device_map=\"auto\",\n torch_dtype=torch.bfloat16\n)\n\ngeneration_args = {\n \"temperature\": 0.1,\n \"top_p\": 0.95,\n \"max_new_tokens\": 25,\n \"repetition_penalty\": 1.2,\n \"do_sample\": True\n}\n\ninputs = tokenizer(text, return_tensors=\"pt\")\n# Generate texts\noutput = model.generate(input_ids=inputs[\"input_ids\"].to(model.device), attention_mask=inputs[\"attention_mask\"], **generation_args)\n# Print outputs\nprint(tokenizer.decode(output[0], skip_special_tokens=True))\n```\n\n</details>\n\n#### Inference with vLLM\nvLLM is an efficient library for inference that enables faster and more scalable text generation.\n\n```bash\npip install vllm\n```\n\n<details>\n<summary>Show code</summary>\n\n```python\nfrom vllm import LLM, SamplingParams\n\nmodel_id = \"BSC-LT/salamandra-7b\"\n\n# Sample prompts\nprompts = [\n \"Las fiestas de San Isidro Labrador de Yecla son\",\n \"El punt més alt del Parc Natural del Montseny és\",\n \"Sentence in English: The typical chance of such a storm is around 10%. Sentence in Catalan:\",\n \"Si le monde était clair\",\n \"The future of AI is\",\n]\n# Create a sampling params object\nsampling_params = SamplingParams(\n temperature=0.1,\n top_p=0.95,\n seed=1,\n max_tokens=25,\n repetition_penalty=1.2)\n\n# Create an LLM\nllm = LLM(model=model_id)\n# Generate texts\noutputs = llm.generate(prompts, sampling_params)\n# Print outputs\nfor output in outputs:\n prompt = output.prompt\n generated_text = output.outputs[0].text\n print(f\"Prompt: {prompt!r}, Generated text: {generated_text!r}\")\n```\n\n</details>\n\n---\n\n## Data\n\n### Pretraining Data\n\nThe pre-training corpus comprises data from 35 European languages and 92 programming languages, with detailed data sources provided below. \nThe initial three training epochs used 2.4 trillion tokens, obtained by manually adjusting data proportion to balance the representation \nand give more importance to Spain’s co-official (Spanish, Catalan, Galician, and Basque). This way, we downsampled code and English data to half, \nSpanish co-official languages were oversampled by 2x, and the remaining languages were kept in their original proportions.\nDuring the following epochs, the Colossal OSCAR dataset was replaced with the FineWeb-Edu dataset. \nThis adjustment resulted in a total of 2.68 trillion tokens, distributed as outlined below:\n\n\n\nThe pretraining corpus is predominantly composed of data from Colossal OSCAR, which contributes a significant 53.05% of the total tokens. \nFollowing this, Starcoder provides 13.67%, and FineWeb-Edu (350BT subset) adds 10.24%. The next largest sources are HPLT at 4.21% and French-PD at 3.59%. \nOther notable contributions include MaCoCu, Legal-ES, and EurLex, each contributing around 1.72% to 1.41%. \nThese major sources collectively form the bulk of the corpus, ensuring a rich and diverse dataset for training the language model. \nThe remaining 10% comes from smaller sources in various languages.\n\nFeel free to click the expand button below to see the full list of sources.\n\n<details>\n<summary>Data Sources</summary>\n \n\n| Dataset | Language | Source |\n|---|---|---|\n| Colossal OSCAR 1.0 | bg, ca, cs, cy, da, de, el, en, es, et, eu, fi, fr, ga, gl, hr, hu, it, lt, lv, mt, nl, nn, no, oc, pl, pt, ro, ru, sh, sk, sl, sr, sv, uk | Brack et al., 2024 |\n| Aya Dataset (w/o Evaluation Suite) | eu, hr, nl, fi, ka, hu, lt, nn, ro, sk, lv, cy, bg, cs, en, fr, de, ga, mt, pl, ru, sl, sv, ca, da, et, gl, el, it, no, pt, sr, es, uk | Singh et al., 2024 |\n| Wikimedia dumps | bg, ca, cs, da, de, el, en, es, et, eu, fi, fr, ga, gl, hr, hu, it, lt, lv, mt, nl, nn, no, pl, pt, ro, sh, sk, sl, sr, uk | [Link](https://dumps.wikimedia.org/) |\n| OpenSubtitles v2016 | bg, ca, cs, da, de, el, en, es, et, eu, fi, fr, gl, hr, it, lt, lv, nl, no, pl, pt, ro, sk, sl, sr, sv, uk | Lison & Tiedemann, 2016 |\n| EurLEX-Resources | bg, cs, da, de, el, en, es, et, fi, fr, ga, hr, hu, it, lt, lv, mt, nl, pl, pt, ro, sk, sl, sv | [Link](https://huggingface.co/datasets/joelniklaus/eurlex_resources) |\n| MC4-Legal | bg, cs, da, de, el, en, es, et, fi, fr, ga, hu, it, lt, lv, mt, nl, pl, pt, ro, sk, sl, sv | [Link](https://huggingface.co/datasets/joelito/legal-mc4) |\n| Parlamint | at, bg, cz, dk, ee, es, es-ga, fi, fr, gb, gr, hr, hu, it, lv, nl, no, pl, pt, rs, se, si | Erjavec et al., 2021 |\n| MaCoCu | bg, ca, el, hr, mt, sl, sr, uk | Bañón et al., 2022 |\n| CURLICAT | bg, hr, hu, pl, ro, sk, sl | Váradi et al., 2022 |\n| Norwegian Colossal Corpus (NCC) | nn, no | Kummervold et al., 2021 |\n| Academic Slovene KAS 2.0 | sl | Žagar et al., 2022 |\n| BIGPATENT | en | Sharma et al., 2019 |\n| Biomedical-ES | es | Internally generated biomedical dataset: Wikipedia LS, Pubmed, MeSpEn, patents, clinical cases, medical crawler |\n| Brazilian Portuguese Web as Corpus (BrWaC) | pt | Wagner Filho et al., 2018 |\n| Bulgarian National Corpus (BulNC) | bg | [Link](http://old.dcl.bas.bg/dataset/BulNC.7z) |\n| CaBeRnet | fr | Popa-Fabre et al., 2020 |\n| CATalog 1.0 | ca | Palomar-Giner et al., 2024 |\n| CorpusNÓS | gl | de-Dios-Flores et al., 2024 |\n| Croatian Web as Corpus 2.1 (hrWaC) | hr | Ljubešić & Klubička, 2014 |\n| DaNewsroom | da | Varab & Schluter, 2020 |\n| Danish GigaWord | da | Strømberg-Derczynski et al., 2021 |\n| DK-CLARIN Reference Corpus of General Danish | da | [Link](https://korpus.dsl.dk/clarin/) |\n| Estonian National Corpus 2021 (ENC) | et | Koppel & Kallas, 2022 |\n| Estonian Reference Corpus (ERC) | et | [Link](https://www.cl.ut.ee/korpused/segakorpus/) |\n| EusCrawl (w/o Wikipedia or NC-licenses) | eu | Artetxe et al., 2022 |\n| FineWeb-Edu (350BT subset) | en | Penedo et al., 2024 |\n| French Public Domain Books (French-PD) | fr | [Link](https://huggingface.co/datasets/PleIAs/French-PD-Books) |\n| French Public Domain Newspapers (French-PD) | fr | [Link](https://huggingface.co/datasets/PleIAs/French-PD-Newspapers) |\n| German Web as Corpus (DeWaC) | de | [Link](https://docs.sslmit.unibo.it/doku.php?id=corpora:dewac) |\n| Greek Legal Code (GLC) | el | Papaloukas et al., 2021 |\n| Greek Web Corpus (GWC) | el | Outsios et al., 2018 |\n| HPLT v1 - Spanish | es | de Gibert et al., 2024 |\n| HPLT v1.1 - Spanish | es | de Gibert et al., 2024 |\n| Irish Universal Dependencies (Ga-UD) | ga | [Link](https://universaldependencies.org/ga/index.html) |\n| Italian Web as Corpus (ItWaC) | it | [Link](https://docs.sslmit.unibo.it/doku.php?id=corpora:itwac) |\n| Korpus Malti | mt | Micallef et al., 2022 |\n| Korpus slovenských právnych predpisov v1.9 (SK-Laws) | sk | [Link](https://www.juls.savba.sk/data/marcell/legal-sk-20220322-1.9.ver.xz) |\n| Latxa Corpus v1.1 (GAITU) | eu | Etxaniz et al., 2024 [Link](https://huggingface.co/datasets/HiTZ/latxa-corpus-v1.1) |\n| Laws and legal acts of Ukraine (UK-Laws) | uk | [Link](https://lang.org.ua/en/corpora/#anchor7) |\n| Legal-ES | es | Internally generated legal dataset: BOE, BORME, Senado, Congreso, Spanish court orders, DOGC |\n| MARCELL Romanian legislative subcorpus v2 | ro | [Link](https://elrc-share.eu/reposMARCELL%20Romanian%20legislative%20subcorpus%20v2itory/browse/marcell-romanian-legislative-subcorpus-v2/2da548428b9d11eb9c1a00155d026706ce94a6b59ffc4b0e9fb5cd9cebe6889e/) |\n| Math AMPS | en | Hendrycks et al., 2021 |\n| NKPJ National Corpus of Polish v1.2 (NKPJ) | pl | Lewandowska-Tomaszczyk et al., 2013 |\n| Occitan Corpus (IEA-AALO) | oc | Provided by [IEA](https://www.institutestudisaranesi.cat/) |\n| Open Legal Data - German court decisions and laws | de | Ostendorff et al., 2020 |\n| ParlamentoPT | pt | Rodrigues et al., 2023 |\n| peS2o | en | Soldaini & Lo, 2023 |\n| PG-19 | en | Rae et al., 2019 |\n| Pile of Law (selected subsets) | en | Henderson* et al., 2022 |\n| Polish Parliamentary Corpus (PPC) | pl | Ogrodniczuk, 2018 |\n| Proof Pile | en | [Link](https://huggingface.co/datasets/hoskinson-center/proof-pile) |\n| RedPajama-Data T1 (StackExchange subset) | en | Computer, 2023 |\n| Scientific-ES | es | Internally generated scientific dataset: Dialnet, Scielo, CSIC, TDX, BSC, UCM |\n| SK Court Decisions v2.0 (OD-Justice) | sk | [Link](https://www.juls.savba.sk/data/od-justice/od-justice-2.0.ver.xz) |\n| Slovene Web as Corpus (slWaC) | sl | Erjavec et al., 2015 |\n| SoNaR Corpus NC 1.2 | nl | [Link](https://taalmaterialen.ivdnt.org/download/tstc-sonar-corpus/) |\n| Spanish Legal Domain Corpora (Spanish-Legal) | es | Gutiérrez-Fandiño et al., 2021 |\n| SrpKorSubset: news, legal, academic, conversation, lit- erary (SrpKor) | sr | [Link](http://www.korpus.matf.bg.ac.rs/) |\n| Starcoder | code | Li et al., 2023 |\n| State-related content from the Latvian Web (State-Latvian-Web) | lv | [Link](https://catalog.elra.info/en-us/repository/browse/ELRA-W0169/) |\n| SYN v9: large corpus of written Czech | cs | Křen et al., 2021 |\n| Tagesschau Archive Article | de | [Link](https://huggingface.co/datasets/bjoernp/tagesschau-2018-2023) |\n| The Danish Parliament Corpus 2009 - 2017, v1 | da | Hansen, 2018 |\n| The Gaois bilingual corpus of English-Irish legislation (Ga-Legislation) | ga | [Link](https://portulanclarin.net/repository/browse/the-gaois-bilingual-corpus-of-english-irish-legislation-processed/daeac17c9e3511ea9b7f02420a000407b83de243dc0b469aab41084386c5b80f/) |\n| The Pile (PhilPapers) | en | Gao et al., 2021 |\n| The Swedish Culturomics Gigaword Corpus (Swedish- Gigaword) | sv | Rødven-Eide, 2016 |\n| Welsh-GOV | cy | Crawling from [Link](https://www.llyw.cymru) |\n| Yle Finnish News Archive (Yle-News) | fi | [Link](http://urn.fi/urn:nbn:fi:lb-2021050401) |\n\nTo consult the data summary document with the respective licences, please send an e-mail to ipr@bsc.es. \n\n<details>\n<summary>References</summary>\n\n- Abadji, J., Suárez, P. J. O., Romary, L., & Sagot, B. (2021). Ungoliant: An optimized pipeline for the generation of a very large-scale multilingual web corpus (H. Lüngen, M. Kupietz, P. Bański, A. Barbaresi, S. Clematide, & I. Pisetta, Eds.; pp. 1–9). Leibniz-Institut für Deutsche Sprache. [Link](https://doi.org/10.14618/ids-pub-10468)\n- Artetxe, M., Aldabe, I., Agerri, R., Perez-de-Viñaspre, O., & Soroa, A. (2022). Does Corpus Quality Really Matter for Low-Resource Languages?\n- Bañón, M., Esplà-Gomis, M., Forcada, M. L., García-Romero, C., Kuzman, T., Ljubešić, N., van Noord, R., Sempere, L. P., Ramírez-Sánchez, G., Rupnik, P., Suchomel, V., Toral, A., van der Werff, T., & Zaragoza, J. (2022). MaCoCu: Massive collection and curation of monolingual and bilingual data: Focus on under-resourced languages. Proceedings of the 23rd Annual Conference of the European Association for Machine Translation, 303–304. [Link](https://aclanthology.org/2022.eamt-1.41)\n- Brack, M., Ostendorff, M., Suarez, P. O., Saiz, J. J., Castilla, I. L., Palomar-Giner, J., Shvets, A., Schramowski, P., Rehm, G., Villegas, M., & Kersting, K. (2024). Community OSCAR: A Community Effort for Multilingual Web Data. [Link](https://occiglot.eu/papers/Community_Oscar.pdf)\n- Computer, T. (2023). RedPajama: An Open Source Recipe to Reproduce LLaMA training dataset [Computer software]. [Link](https://github.com/togethercomputer/RedPajama-Data)\n- de Gibert, O., Nail, G., Arefyev, N., Bañón, M., van der Linde, J., Ji, S., Zaragoza-Bernabeu, J., Aulamo, M., Ramírez-Sánchez, G., Kutuzov, A., Pyysalo, S., Oepen, S., & Tiedemann, J. (2024). A New Massive Multilingual Dataset for High-Performance Language Technologies (arXiv:2403.14009). arXiv. [Link](http://arxiv.org/abs/2403.14009)\n- Dodge, J., Sap, M., Marasović, A., Agnew, W., Ilharco, G., Groeneveld, D., Mitchell, M., & Gardner, M. (2021). Documenting Large Webtext Corpora: A Case Study on the Colossal Clean Crawled Corpus. In M.-F. Moens, X. Huang, L. Specia, & S. W. Yih (Eds.), Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (pp. 1286–1305). Association for Computational Linguistics. [Link](https://doi.org/10.18653/v1/2021.emnlp-main.98)\n- Erjavec, T., Ljubešić, N., & Logar, N. (2015). The slWaC corpus of the Slovene web. Informatica (Slovenia), 39, 35–42.\n- Erjavec, T., Ogrodniczuk, M., Osenova, P., Ljubešić, N., Simov, K., Grigorova, V., Rudolf, M., Pančur, A., Kopp, M., Barkarson, S., Steingrímsson, S. hór, van der Pol, H., Depoorter, G., de Does, J., Jongejan, B., Haltrup Hansen, D., Navarretta, C., Calzada Pérez, M., de Macedo, L. D., … Rayson, P. (2021). Linguistically annotated multilingual comparable corpora of parliamentary debates ParlaMint.ana 2.1. [Link](http://hdl.handle.net/11356/1431)\n- Etxaniz, J., Sainz, O., Perez, N., Aldabe, I., Rigau, G., Agirre, E., Ormazabal, A., Artetxe, M., & Soroa, A. (2024). Latxa: An Open Language Model and Evaluation Suite for Basque. [Link] (https://arxiv.org/abs/2403.20266)\n- Gao, L., Biderman, S., Black, S., Golding, L., Hoppe, T., Foster, C., Phang, J., He, H., Thite, A., Nabeshima, N., Presser, S., & Leahy, C. (2021). The Pile: An 800GB Dataset of Diverse Text for Language Modeling. CoRR, abs/2101.00027. [Link](https://arxiv.org/abs/2101.00027)\n- Gutiérrez-Fandiño, A., Armengol-Estapé, J., Gonzalez-Agirre, A., & Villegas, M. (2021). Spanish Legalese Language Model and Corpora.\n- Hansen, D. H. (2018). The Danish Parliament Corpus 2009—2017, v1. [Link](http://hdl.handle.net/20.500.12115/8)\n- Henderson*, P., Krass*, M. S., Zheng, L., Guha, N., Manning, C. D., Jurafsky, D., & Ho, D. E. (2022). Pile of Law: Learning Responsible Data Filtering from the Law and a 256GB Open-Source Legal Dataset. arXiv. [Link](https://arxiv.org/abs/2207.00220)\n- Hendrycks, D., Burns, C., Kadavath, S., Arora, A., Basart, S., Tang, E., Song, D., & Steinhardt, J. (2021). Measuring Mathematical Problem Solving With the MATH Dataset. NeurIPS.\n- Jansen, T., Tong, Y., Zevallos, V., & Suarez, P. O. (2022). Perplexed by Quality: A Perplexity-based Method for Adult and Harmful Content Detection in Multilingual Heterogeneous Web Data.\n- Koppel, K., & Kallas, J. (2022). Eesti keele ühendkorpuste sari 2013–2021: Mahukaim eestikeelsete digitekstide kogu. Eesti Rakenduslingvistika Ühingu Aastaraamat Estonian Papers in Applied Linguistics, 18, 207–228. [Link](https://doi.org/10.5128/erya18.12)\n- Křen, M., Cvrček, V., Henyš, J., Hnátková, M., Jelínek, T., Kocek, J., Kováříková, D., Křivan, J., Milička, J., Petkevič, V., Procházka, P., Skoumalová, H., Šindlerová, J., & Škrabal, M. (2021). SYN v9: Large corpus of written Czech. [Link](http://hdl.handle.net/11234/1-4635)\n- Kreutzer, J., Caswell, I., Wang, L., Wahab, A., van Esch, D., Ulzii-Orshikh, N., Tapo, A., Subramani, N., Sokolov, A., Sikasote, C., Setyawan, M., Sarin, S., Samb, S., Sagot, B., Rivera, C., Rios, A., Papadimitriou, I., Osei, S., Suarez, P. O., … Adeyemi, M. (2022). Quality at a Glance: An Audit of Web-Crawled Multilingual Datasets. Transactions of the Association for Computational Linguistics, 10, 50–72. [Link](https://doi.org/10.1162/tacl_a_00447)\n- Kummervold, P. E., De la Rosa, J., Wetjen, F., & Brygfjeld, S. A. (2021). Operationalizing a National Digital Library: The Case for a Norwegian Transformer Model. In S. Dobnik & L. Øvrelid (Eds.), Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa) (pp. 20–29). Linköping University Electronic Press, Sweden. [Link](https://aclanthology.org/2021.nodalida-main.3)\n- Lewandowska-Tomaszczyk, B., Górski, R., Łaziński, M., & Przepiórkowski, A. (2013). The National Corpus of Polish (NKJP). Language use and data analysis. 309–319.\n- Li, R., Allal, L. B., Zi, Y., Muennighoff, N., Kocetkov, D., Mou, C., Marone, M., Akiki, C., Li, J., Chim, J., Liu, Q., Zheltonozhskii, E., Zhuo, T. Y., Wang, T., Dehaene, O., Davaadorj, M., Lamy-Poirier, J., Monteiro, J., Shliazhko, O., … Vries, H. de. (2023). StarCoder: May the source be with you!\n- Lison, P., & Tiedemann, J. (2016). OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles. In N. Calzolari, K. Choukri, T. Declerck, S. Goggi, M. Grobelnik, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, & S. Piperidis (Eds.), Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16) (pp. 923–929). European Language Resources Association (ELRA). [Link](https://aclanthology.org/L16-1147)\n- Ljubešić, N., & Klubička, F. (2014). Bs,hr,srWaC - Web Corpora of Bosnian, Croatian and Serbian. In F. Bildhauer & R. Schäfer (Eds.), Proceedings of the 9th Web as Corpus Workshop (WaC-9) (pp. 29–35). Association for Computational Linguistics. [Link](https://doi.org/10.3115/v1/W14-0405)\n- Micallef, K., Gatt, A., Tanti, M., van der Plas, L., & Borg, C. (2022). Pre-training Data Quality and Quantity for a Low-Resource Language: New Corpus and BERT Models for Maltese. Proceedings of the Third Workshop on Deep Learning for Low-Resource Natural Language Processing, 90–101. [Link](https://doi.org/10.18653/v1/2022.deeplo-1.10)\n- Ogrodniczuk, M. (2018). Polish Parliamentary Corpus. [Link](https://api.semanticscholar.org/CorpusID:235134113)\n- Ostendorff, M., Blume, T., & Ostendorff, S. (2020). Towards an Open Platform for Legal Information. Proceedings of the ACM/IEEE Joint Conference on Digital Libraries in 2020, 385–388. [Link](https://doi.org/10.1145/3383583.3398616)\n- Ostendorff, M., Suarez, P. O., Lage, L. F., & Rehm, G. (2024). LLM-Datasets: An Open Framework for Pretraining Datasets of Large Language Models. First Conference on Language Modeling. [Link](https://openreview.net/forum?id=5RdIMlGLXL)\n- Outsios, S., Skianis, K., Meladianos, P., Xypolopoulos, C., & Vazirgiannis, M. (2018). Word Embeddings from Large-Scale Greek Web content. arXiv Preprint arXiv:1810.06694.\n- Palomar-Giner, J., Saiz, J. J., Espuña, F., Mina, M., Da Dalt, S., Llop, J., Ostendorff, M., Ortiz Suarez, P., Rehm, G., Gonzalez-Agirre, A., & Villegas, M. (2024). A CURATEd CATalog: Rethinking the Extraction of Pretraining Corpora for Mid-Resourced Languages. In N. Calzolari, M.-Y. Kan, V. Hoste, A. Lenci, S. Sakti, & N. Xue (Eds.), Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) (pp. 335–349). ELRA and ICCL. [Link](https://aclanthology.org/2024.lrec-main.31)\n- Papaloukas, C., Chalkidis, I., Athinaios, K., Pantazi, D.-A., & Koubarakis, M. (2021). Multi-granular Legal Topic Classification on Greek Legislation. Proceedings of the Natural Legal Language Processing Workshop 2021, 63–75. [Link](https://doi.org/10.48550/arXiv.2109.15298)\n- Popa-Fabre, M., Ortiz Suárez, P. J., Sagot, B., & de la Clergerie, É. (2020). French Contextualized Word-Embeddings with a sip of CaBeRnet: A New French Balanced Reference Corpus. Proceedings of the 8th Workshop on Challenges in the Management of Large Corpora, 15–23. [Link](https://aclanthology.org/2020.cmlc-1.3)\n- Rae, J. W., Potapenko, A., Jayakumar, S. M., Hillier, C., & Lillicrap, T. P. (2019). Compressive Transformers for Long-Range Sequence Modelling. arXiv Preprint. [Link](https://arxiv.org/abs/1911.05507)\n- Rodrigues, J., Gomes, L., Silva, J., Branco, A., Santos, R., Cardoso, H. L., & Osório, T. (2023). Advancing Neural Encoding of Portuguese with Transformer Albertina PT-\\*.\n- Rødven-Eide, S. (2016). The Swedish Culturomics Gigaword CorpusThe Swedish Culturomics Gigaword Corpus [Dataset]. Språkbanken Text. [Link](https://doi.org/10.23695/3WMV-1Z09)\n- Sharma, E., Li, C., & Wang, L. (2019). BIGPATENT: A Large-Scale Dataset for Abstractive and Coherent Summarization. CoRR, abs/1906.03741. [Link](http://arxiv.org/abs/1906.03741)\n- Soldaini, L., & Lo, K. (2023). peS2o (Pretraining Efficiently on S2ORC) Dataset. Allen Institute for AI.\n- Strømberg-Derczynski, L., Ciosici, M., Baglini, R., Christiansen, M. H., Dalsgaard, J. A., Fusaroli, R., Henrichsen, P. J., Hvingelby, R., Kirkedal, A., Kjeldsen, A. S., Ladefoged, C., Nielsen, F. Å., Madsen, J., Petersen, M. L., Rystrøm, J. H., & Varab, D. (2021). The Danish Gigaword Corpus. Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa), 413–421. [Link](https://aclanthology.org/2021.nodalida-main.46)\n- Subramani, N., Luccioni, S., Dodge, J., & Mitchell, M. (2023). Detecting Personal Information in Training Corpora: An Analysis. 208–220. [Link](https://doi.org/10.18653/v1/2023.trustnlp-1.18)\n- Varab, D., & Schluter, N. (2020). DaNewsroom: A Large-scale Danish Summarisation Dataset. Proceedings of The 12th Language Resources and Evaluation Conference, 6731–6739. [Link](https://www.aclweb.org/anthology/2020.lrec-1.831)\n- Váradi, T., Nyéki, B., Koeva, S., Tadić, M., Štefanec, V., Ogrodniczuk, M., Nitoń, B., Pezik, P., Barbu Mititelu, V., Irimia, E., Mitrofan, M., Tufi\\textcommabelows, D., Garabík, R., Krek, S., & Repar, A. (2022). Introducing the CURLICAT Corpora: Seven-language Domain Specific Annotated Corpora from Curated Sources. In N. Calzolari, F. Béchet, P. Blache, K. Choukri, C. Cieri, T. Declerck, S. Goggi, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, J. Odijk, & S. Piperidis (Eds.), Proceedings of the Thirteenth Language Resources and Evaluation Conference (pp. 100–108). European Language Resources Association. [Link](https://aclanthology.org/2022.lrec-1.11)\n- Wagner Filho, J. A., Wilkens, R., Idiart, M., & Villavicencio, A. (2018). The brwac corpus: A new open resource for brazilian portuguese. Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018).\n- Žagar, A., Kavaš, M., Robnik-Šikonja, M., Erjavec, T., Fišer, D., Ljubešić, N., Ferme, M., Borovič, M., Boškovič, B., Ojsteršek, M., & Hrovat, G. (2022). Corpus of academic Slovene KAS 2.0. [Link](http://hdl.handle.net/11356/1448)\n- Alicia Parrish, Angelica Chen, Nikita Nangia, Vishakh Padmakumar, Jason Phang, Jana Thompson, Phu Mon Htut, and Samuel Bowman. 2022. BBQ: A hand-built bias benchmark for question answering. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2086–2105, Dublin, Ireland. Association for Computational Linguistics.\n- Emily Sheng, Kai-Wei Chang, Premkumar Natarajan, and Nanyun Peng. 2019. The Woman Worked as a Babysitter: On Biases in Language Generation. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3407–3412, Hong Kong, China. Association for Computational Linguistics.\n- Clark, P., Cowhey, I., Etzioni, O., Khot, T., Sabharwal, A., Schoenick, C., & Tafjord, O. (2018). Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge. arXiv:1803. 05457v1.\n- Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D. Manning, Andrew Ng, and Christopher Potts. 2013. Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1631–1642, Seattle, Washington, USA. Association for Computational Linguistics.\n- Penedo, G., Kydlíček, H., allal, L. B., Lozhkov, A., Mitchell, M., Raffel, C., Von Werra, L., & Wolf, T. (2024). The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale (arXiv:2406.17557). arXiv. http://arxiv.org/abs/2406.17557\n- Singh, S., Vargus, F., Dsouza, D., Karlsson, B. F., Mahendiran, A., Ko, W.-Y., Shandilya, H., Patel, J., Mataciunas, D., OMahony, L., Zhang, M., Hettiarachchi, R., Wilson, J., Machado, M., Moura, L. S., Krzemiński, D., Fadaei, H., Ergün, I., Okoh, I., … Hooker, S. (2024). Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning (arXiv:2402.06619). arXiv. http://arxiv.org/abs/2402.06619\n\n</details>\n\n</details>\n\nThe model was trained on 3 pre-training epochs with 2.4T tokens per epoch, 2 additional pre-training epochs in which the English part \nof the Colossal OSCAR dataset was replaced with FineWeb-Edu (350BT subset), resulting in 2.68T tokens per epoch; \nand 1 final epoch of 0.315T higher quality tokens, meaning that the total number of tokens seen during pre-training is approximately 12.875 trillion tokens.\n\nWe provide an extense Datasheet section following the best practices defined by [(Gebru et al., 2021)](https://arxiv.org/pdf/1803.09010).\n\n<details>\n<summary>Datasheet</summary>\n\n#### Motivation\n\n**For what purpose was the dataset created? Was there a specific task in mind? Was there a specific gap that needed to be filled? Please provide a description.**\n\nThe purpose of creating this dataset is to pre-train the Salamandra family of multilingual models with high performance in a large number of European languages (35) \nand programming languages (92). We also want to represent the co-official languages of Spain: Spanish, Catalan, Galician and Basque. For this reason, we oversample \nthese languages by a factor of 2.\n\nThere is a great lack of massive multilingual data, especially in minority languages (Ostendorff & Rehm, 2023), so part of our efforts in the creation of \nthis pre-training dataset have resulted in the contribution to large projects such as the Community OSCAR (Brack et al., 2024), which includes 151 languages \nand 40T words, or CATalog (Palomar-Giner et al., 2024), the largest open dataset in Catalan in the world.\n\n**Who created the dataset (e.g., which team, research group) and on behalf of which entity (e.g., company, institution, organization)?**\n\nThe dataset has been created by the Language Technologies unit (LangTech) of the Barcelona Supercomputing Center - Centro Nacional de Supercomputación (BSC-CNS), \nwhich aims to advance the field of natural language processing through cutting-edge research and development and the use of HPC. In particular, it was created by \nthe unit's data team, the main contributors being José Javier Saiz, Ferran Espuña and Jorge Palomar.\n\nHowever, the creation of the dataset would not have been possible without the collaboration of a large number of collaborators, partners and public institutions, \nwhich can be found in detail in the acknowledgements.\n\n**Who funded the creation of the dataset? If there is an associated grant, please provide the name of the grantor and the grant name and number.**\n\nThis work has been promoted and financed by the Government of Catalonia through the [Aina project](https://projecteaina.cat/).\n\nThis work is funded by the _Ministerio para la Transformación Digital y de la Función Pública_ - Funded by EU – NextGenerationEU \nwithin the framework of [ILENIA Project](https://proyectoilenia.es/) with reference 2022/TL22/00215337.\n\n#### Composition\n\n**What do the instances that comprise the dataset represent (e.g., documents, photos, people, countries)? Are there multiple types of instances (e.g., movies, users, and ratings; people and interactions between them; nodes and edges)? Please provide a description.**\n\nThe dataset consists entirely of text documents in various languages. Specifically, data was mainly sourced from the following databases and \nrepositories:\n- **Common Crawl:** Repository that holds website data and is run by the Common Crawl non-profit organization. It is updated monthly and is\n distributed under the CC0 1.0 public domain license.\n- **GitHub:** Community platform that allows developers to create, store, manage, and share their code. Repositories are crawled and then\n distributed with their original licenses, which may vary from permissive to non-commercial licenses.\n- **Wikimedia:** Database that holds the collection databases managed by the Wikimedia Foundation, including Wikipedia, Wikibooks, Wikinews,\n Wikiquote, Wikisource, and Wikivoyage. It is updated monthly and is distributed under Creative Commons Attribution-ShareAlike License 4.0.\n- **EurLex:** Repository that holds the collection of legal documents from the European Union, available in all of the EU’s 24 official\n languages and run by the Publications Office of the European Union. It is updated daily and is distributed under the Creative Commons\n Attribution 4.0 International license.\n- **Other repositories:** Specific repositories were crawled under permission for domain-specific corpora, which include academic, legal,\n and newspaper repositories.\n\nWe provide a complete list of dataset sources at the end of this section.\n\n**How many instances are there in total (of each type, if appropriate)?**\n\nThe dataset contains a diverse range of instances across multiple languages, with notable adjustments for certain languages. English \nrepresents the largest portion, accounting for 39.31% of the total data. Spanish was upsampled by a factor of 2, bringing its share to 16.12%,\nwhile Catalan (1.97%), Basque (0.24%), and Galician (0.31%) were also upsampled by 2. On the other hand, code-related data was downsampled \nby half, making up 5.78% of the total. Other prominent languages include French (6.6%), Russian (5.56%), German (4.79%), and Hungarian \n(4.59%), with several additional languages contributing between 1% and 2%, and smaller portions represented by a variety of others.\n\n**Does the dataset contain all possible instances or is it a sample (not necessarily random) of instances from a larger set? If the dataset is a sample, then what is the larger set? Is the sample representative of the larger set (e.g., geographic coverage)? If so, please describe how this representativeness was validated/verified. If it is not representative of the larger set, please describe why not (e.g., to cover a more diverse range of instances, because instances were withheld or unavailable).**\n\nThe dataset is a sample from multiple sources, with different weights based on the primary language of the content: Spanish, Catalan, \nBasque, and Galician content was upsampled by a factor of two, while programming languages were downsampled by a factor of half. Other \nsources were sampled in proportion to their occurrence.\n\n**What data does each instance consist of? “Raw” data (e.g., unprocessed text or images) or features? In either case, please provide a description.**\n\nEach instance consists of a text document processed for deduplication, language identification, and source-specific filtering. Some documents required \noptical character recognition (OCR) to extract text from non-text formats such as PDFs.\n\n**Is there a label or target associated with each instance? If so, please provide a description.**\n\nEach instance is labelled with a unique identifier, the primary language of the content, and the URL for web-sourced instances. Additional labels were \nautomatically assigned to detect specific types of content -harmful or toxic content- and to assign preliminary indicators of undesired qualities -very \nshort documents, high density of symbols, etc.- which were used for filtering instances.\n\n**Is any information missing from individual instances? If so, please provide a description, explaining why this information is missing (e.g., because it was unavailable). This does not include intentionally removed information, but might include, e.g., redacted text.**\n\nNo significant information is missing from the instances.\n\n**Are relationships between individual instances made explicit (e.g., users’ movie ratings, social network links)? If so, please describe how these relationships are made explicit.**\n\nInstances are related through shared metadata, such as source and language identifiers.\n\n**Are there recommended data splits (e.g., training, development/validation, testing)? If so, please provide a description of these splits, explaining the rationale behind them.**\n\nThe dataset is randomly divided into training, validation and test sets, where the validation and test sets are each 1% of the total corpus.\n\n**Are there any errors, sources of noise, or redundancies in the dataset? If so, please provide a description.**\n\nDespite removing duplicated instances within each source, redundancy remains at the paragraph and sentence levels, particularly in web-sourced \ninstances where search engine optimization techniques and templates contribute to repeated textual patterns. Some instances may be also duplicated \nacross sources due to format variations.\n\n**Is the dataset self-contained, or does it link to or otherwise rely on external resources (e.g., websites, tweets, other datasets)? If it links to or relies on external resources, a) are there guarantees that they will exist, and remain constant, over time; b) are there official archival versions of the complete dataset (i.e., including the external resources as they existed at the time the dataset was created); c) are there any restrictions (e.g., licenses, fees) associated with any of the external resources that might apply to a dataset consumer? Please provide descriptions of all external resources and any restrictions associated with them, as well as links or other access points, as appropriate.**\n\nThe dataset is self-contained and does not rely on external resources.\n\n**Does the dataset contain data that might be considered confidential (e.g., data that is protected by legal privilege or by doctor–patient confidentiality, data that includes the content of individuals’ non-public communications)? If so, please provide a description.**\n\nThe dataset does not contain confidential data.\n\n**Does the dataset contain data that, if viewed directly, might be offensive, insulting, threatening, or might otherwise cause anxiety? If so, please describe why. If the dataset does not relate to people, you may skip the remaining questions in this section.**\n\nThe dataset includes web-crawled content, which may overrepresent pornographic material across languages (Kreutzer et al., 2022). Although \npre-processing techniques were applied to mitigate offensive content, the heterogeneity and scale of web-sourced data make exhaustive \nfiltering challenging, which makes it next to impossible to identify all adult content without falling into excessive filtering, which may \nnegatively influence certain demographic groups (Dodge et al., 2021).\n\n**Does the dataset identify any subpopulations (e.g., by age, gender)? If so, please describe how these subpopulations are identified and provide a description of their respective distributions within the dataset.**\n\nThe dataset does not explicitly identify any subpopulations.\n\n**Is it possible to identify individuals (i.e., one or more natural persons), either directly or indirectly (i.e., in combination with other data) from the dataset? If so, please describe how.**\n\nWeb-sourced instances in the dataset may contain personally identifiable information (PII) that is publicly available on the Web, such as names, \nIP addresses, email addresses, and phone numbers. While it would be possible to indirectly identify individuals through the combination of multiple \ndata points, the nature and scale of web data makes it difficult to parse such information. In any case, efforts are made to filter or anonymize \nsensitive data (Mina et al., 2024), but some identifiable information may remain in the dataset.\n\n**Does the dataset contain data that might be considered sensitive in any way? If so, please provide a description.**\n\nGiven that the dataset includes web-sourced content and other publicly available documents, instances may inadvertently reveal financial \ninformation, health-related details, or forms of government identification, such as social security numbers (Subramani et al., 2023), \nespecially if the content originates from less-regulated sources or user-generated platforms. \n\n#### Collection Process\n\n**How was the data collected?**\n\nThis dataset is constituted by combining several sources, whose acquisition methods can be classified into three groups:\n- Web-sourced datasets with some preprocessing available under permissive license.\n- Domain-specific or language-specific raw crawls.\n- Manually curated data obtained through collaborators, data providers (by means of legal assignment agreements) or open source projects (e.g. CATalog).\n\n**What mechanisms or procedures were used to collect the data? How were these mechanisms or procedures validated?**\n\nThe data collection process was carried out using three different mechanisms, each corresponding to one of the groups defined in the previous answer. The specific methods used and their respective validation procedures are outlined below:\n- Open Direct Download: Data were obtained directly from publicly accessible sources, such as websites or repositories that provide open data downloads. We validate the data with a data integrity check, which ensures that the downloaded files are complete, uncorrupted and in the expected format and structure.\n- Ad hoc scrapers or crawlers: Custom web scraping scripts or crawlers were used to extract data from various online sources where direct downloads were not available. These scripts navigate web pages, extract relevant data and store it in a structured format. We validate this method with software unit tests to evaluate the functionality of individual components of the scraping programs, checking for errors or unexpected behaviour. In addition, data integrity tests were performed to verify that the collected data remained complete throughout the extraction and storage process.\n- Direct download via FTP, SFTP, API or S3: Some datasets were acquired using secure transfer protocols such as FTP (File Transfer Protocol), SFTP (Secure File Transfer Protocol), or API (Application Programming Interface) requests from cloud storage services such as Amazon S3. As with the open direct download method, data integrity tests were used to validate the completeness of the files to ensure that the files were not altered or corrupted during the transfer process.\n\n**If the dataset is a sample from a larger set, what was the sampling strategy?**\n\nThe sampling strategy was to use the whole dataset resulting from the filtering explained in the 'preprocessing/cleaning/labelling' section, \nwith the particularity that an upsampling of 2 (i.e. twice the probability of sampling a document) was performed for the co-official languages \nof Spain (Spanish, Catalan, Galician, Basque), and a downsampling of 1/2 was applied for code (half the probability of sampling a code document, \nevenly distributed among all programming languages).\n\n**Who was involved in the data collection process and how were they compensated?**\n\nThis data is generally extracted, filtered and sampled by automated processes. The code required to run these processes has been developed entirely \nby members of the Language Technologies data team, or otherwise obtained from open-source software. Furthermore, there has been no monetary \nconsideration for acquiring data from suppliers.\n\n**Over what timeframe was the data collected? Does this timeframe match the creation timeframe of the data associated with the instances? If not, please describe the timeframe in which the data associated with the instances was created.**\n\nData were acquired and processed from April 2023 to April 2024. However, as mentioned, much data has been obtained from open projects such \nas Common Crawl, which contains data from 2014, so it is the end date (04/2024) rather than the start date that is important.\n\n**Were any ethical review processes conducted? If so, please provide a description of these review processes, including the outcomes, as well as a link or other access point to any supporting documentation.**\n\nNo particular ethical review process has been carried out as the data is mostly open and not particularly sensitive. However, we have an \ninternal evaluation team and a bias team to monitor ethical issues. In addition, we work closely with ‘Observatori d'Ètica en Intel·ligència \nArtificial’ (OEIAC) and ‘Agencia Española de Supervisión de la Inteligencia Artificial’ (AESIA) to audit the processes we carry out from an \nethical and legal point of view, respectively.\n\n#### Preprocessing\n\n**Was any preprocessing/cleaning/labeling of the data done? If so, please provide a description. If not, you may skip the remaining questions in this section.**\n\nNo changes were made to the content of individual text document instances. However, the web-sourced documents underwent a filtering process based on specific criteria along two key dimensions:\n- Quality filtering: The text processing pipeline CURATE (Palomar et. al, 2024) calculates a quality score for each document based on a set of filtering criteria that identify undesirable textual characteristics. Any document with a score below the 0.8 threshold was excluded from the dataset.\n- Harmful or adult content filtering: To reduce the amount of harmful or inappropriate material in the dataset, documents from Colossal OSCAR were filtered using the Ungoliant pipeline (Abadji et al., 2021), which uses the 'harmful\\_pp' field, a perplexity-based score generated by a language model.\n\n**Was the “raw” data saved in addition to the preprocessed/cleaned/labeled data? If so, please provide a link or other access point to the “raw” data.**\n\nThe original raw data was not kept.\n\n**Is the software that was used to preprocess/clean/label the data available? If so, please provide a link or other access point.**\n\nYes, the preprocessing and filtering software is open-sourced. The [CURATE](https://github.com/langtech-bsc/CURATE) pipeline was used for CATalog and other curated datasets, \nand the [Ungoliant](https://github.com/oscar-project/ungoliant) pipeline was used for the OSCAR project.\n\n#### Uses\n\n**Has the dataset been used for any tasks already? If so, please provide a description.**\n\nPre-train the Salamandra model family.\n\n**What (other) tasks could the dataset be used for?**\n\nThe data can be used primarily to pre-train other language models, which can then be used for a wide range of use cases. The dataset could \nalso be used for other tasks such as fine-tuning language models, cross-lingual NLP tasks, machine translation, domain-specific text \ngeneration, and language-specific data analysis.\n\n**Is there anything about the composition of the dataset or the way it was collected and preprocessed/cleaned/labeled that might impact future uses? Is there anything a dataset consumer could do to mitigate these risks or harms?**\n \nWeb-crawled content is over-represented with standard language varieties, impacting language model performance for minority languages. \nLanguage diversity in data is crucial to avoid bias, especially in encoding non-standard dialects, preventing the exclusion of demographic \ngroups. Moreover, despite legal uncertainties in web-scraped data, we prioritize permissive licenses and privacy protection measures, \nacknowledging the challenges posed by personally identifiable information (PII) within large-scale datasets. Our ongoing efforts aim to \naddress privacy concerns and contribute to a more inclusive linguistic dataset.\n\n**Are there tasks for which the dataset should not be used?**\n\n-\n\n#### Distribution \n\n**Will the dataset be distributed to third parties outside of the entity on behalf of which the dataset was created? If so, please provide a description.**\n\nThe dataset will not be released or distributed to third parties. Any related question to distribution is omitted in this section.\n\n#### Maintenance\n\n**Who will be supporting/hosting/maintaining the dataset?**\n\nThe dataset will be hosted by the Language Technologies unit (LangTech) of the Barcelona Supercomputing Center (BSC). The team will ensure \nregular updates and monitor the dataset for any issues related to content integrity, legal compliance, and bias for the sources they are \nresponsible for.\n\n**How can the owner/curator/manager of the dataset be contacted?**\n\nThe data owner may be contacted with the email address langtech@bsc.es.\n\n**Will the dataset be updated?**\n\nThe dataset will not be updated.\n\n**If the dataset relates to people, are there applicable limits on the retention of the data associated with the instances? If so, please describe these limits and explain how they will be enforced.**\n\nThe dataset does not keep sensitive data that could allow direct identification of individuals, apart from the data that is publicly available in \nweb-sourced content. Due to the sheer volume and diversity of web data, it is not feasible to notify individuals or manage data retention on an \nindividual basis. However, efforts are made to mitigate the risks associated with sensitive information through pre-processing and filtering to \nremove identifiable or harmful content. Despite these measures, vigilance is maintained to address potential privacy and ethical issues.\n\n**Will older versions of the dataset continue to be supported/hosted/maintained? If so, please describe how. If not, please describe how its obsolescence will be communicated to dataset consumers.**\n\nSince the dataset will not be updated, only the final version will be kept.\n\n**If others want to extend/augment/build on/contribute to the dataset, is there a mechanism for them to do so?**\n\nThe dataset does not allow for external contributions.\n\n</details>\n\n---\n\n\n## Evaluation\n\n### Gold-standard benchmarks\n\nEvaluation is done using the Language Model Evaluation Harness (Gao et al., 2024). We evaluate on a set of tasks taken from [SpanishBench](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/spanish_bench), [CatalanBench](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/catalan_bench), [BasqueBench](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/basque_bench) and [GalicianBench](https://github.com/EleutherAI/lm-evaluation-harness/tree/main/lm_eval/tasks/galician_bench). We also use English tasks already available on the LM Evaluation Harness. These benchmarks include both new and existing tasks and datasets. In the tables below, we include the results in a selection of evaluation datasets that represent model's performance across a variety of tasks within these benchmarks. \n\nWe only use tasks that are either human generated, human translated, or with a strong human-in-the-loop (i.e., machine translation followed by professional revision or machine generation followed by human revision and annotation). This is the reason behind the variety in number of tasks reported across languages. As more tasks that fulfill these requirements are published, we will update the presented results. We also intend to expand the evaluation to other languages, as long as the datasets meet our quality standards.\n\nDuring the implementation of the evaluation we observed a series of issues worth considering when replicating and interpreting the results presented. These issues include ≈1.5% variances in performance in some tasks depending on the version of the `transformers` library used, and depending on the use (or lack of use) of tensor parallelism when loading a model. When implementing existing tasks, we carry out a comprehensive quality evaluation of the dataset, the Harness task itself, and what kind of input models see during evaluation. Our implementation (see links above) addresses multiple existing problems such as errors in datasets and prompts, and lack of pre-processing. All this means that results will vary if using other Harness implementations, and may slightly vary depending on the replication setup.\n\nIt should be noted that these results are subject to all the drawbacks of every current gold-standard evaluation, and that the figures do not fully represent the models capabilities and potential. We thus advise caution when reading and interpreting the results.\n\nA full list of results compared to other baselines, a discussion of the model's performance across tasks and its implications, and details regarding problem-solving with task implementation will soon be available in the technical report.\n\nAll results reported below are on a 5-shot setting.\n\n#### Spanish\n\n<table><thead>\n <tr>\n <th>Category</th>\n <th>Task</th>\n <th>Metric</th>\n <th>Result</th>\n </tr></thead>\n<tbody>\n <tr>\n <td rowspan=\"2\">Commonsense Reasoning</td>\n <td>copa_es</td>\n <td>acc</td>\n <td>86</td>\n </tr>\n <tr>\n <td>xstorycloze_es</td>\n <td>acc</td>\n <td>74.32</td>\n </tr>\n <tr>\n <td rowspan=\"2\">NLI</td>\n <td>wnli_es</td>\n <td>acc</td>\n <td>59.15</td>\n </tr>\n <tr>\n <td>xnli_es</td>\n <td>acc</td>\n <td>46.59</td>\n </tr>\n <tr>\n <td>Paraphrasing</td>\n <td>paws_es</td>\n <td>acc</td>\n <td>60.3</td>\n </tr>\n <tr>\n <td rowspan=\"2\">QA</td>\n <td>openbookqa_es</td>\n <td>acc</td>\n <td>41.6</td>\n </tr>\n <tr>\n <td>xquad_es</td>\n <td>acc</td>\n <td>72.26</td>\n </tr>\n <tr>\n <td>Translation</td>\n <td>flores_es</td>\n <td>bleu</td>\n <td>23.43</td>\n </tr>\n</tbody>\n</table>\n\n#### Catalan\n\n<table><thead>\n <tr>\n <th>Category</th>\n <th>Task</th>\n <th>Metric</th>\n <th>Result</th>\n </tr></thead>\n<tbody>\n <tr>\n <td rowspan=\"2\">Commonsense Reasoning</td>\n <td>copa_ca</td>\n <td>acc</td>\n <td>84</td>\n </tr>\n <tr>\n <td>xstorycloze_ca</td>\n <td>acc</td>\n <td>75.51</td>\n </tr>\n <tr>\n <td rowspan=\"2\">NLI</td>\n <td>wnli_ca</td>\n <td>acc</td>\n <td>59.15</td>\n </tr>\n <tr>\n <td>xnli_ca</td>\n <td>acc</td>\n <td>50.16</td>\n </tr>\n <tr>\n <td rowspan=\"2\">Paraphrasing</td>\n <td>parafraseja</td>\n <td>acc</td>\n <td>65.83</td>\n </tr>\n <tr>\n <td>paws_ca</td>\n <td>acc</td>\n <td>67.45</td>\n </tr>\n <tr>\n <td rowspan=\"5\">QA</td>\n <td>arc_ca_easy</td>\n <td>acc</td>\n <td>71.72</td>\n </tr>\n <tr>\n <td>arc_ca_challenge</td>\n <td>acc</td>\n <td>45.56</td>\n </tr>\n <tr>\n <td>openbookqa_ca</td>\n <td>acc</td>\n <td>38.8</td>\n </tr>\n <tr>\n <td>piqa_ca</td>\n <td>acc</td>\n <td>71.27</td>\n </tr>\n <tr>\n <td>siqa_ca</td>\n <td>acc</td>\n <td>49.85</td>\n </tr>\n <tr>\n <td>Translation</td>\n <td>flores_ca</td>\n <td>bleu</td>\n <td>30.63</td>\n </tr>\n</tbody></table>\n\n#### Basque\n\n<table><thead>\n <tr>\n <th>Category</th>\n <th>Task</th>\n <th>Metric</th>\n <th>Result</th>\n </tr></thead>\n<tbody>\n <tr>\n <td rowspan=\"2\">Commonsense Reasoning</td>\n <td>xcopa_eu</td>\n <td>acc</td>\n <td>68.8</td>\n </tr>\n <tr>\n <td>xstorycloze_eu</td>\n <td>acc</td>\n <td>66.12</td>\n </tr>\n <tr>\n <td rowspan=\"2\">NLI</td>\n <td>wnli_eu</td>\n <td>acc</td>\n <td>57.75</td>\n </tr>\n <tr>\n <td>xnli_eu</td>\n <td>acc</td>\n <td>43.51</td>\n </tr>\n <tr>\n <td rowspan=\"4\">QA</td>\n <td>eus_exams</td>\n <td>acc</td>\n <td>41.04</td>\n </tr>\n <tr>\n <td>eus_proficiency</td>\n <td>acc</td>\n <td>39.72</td>\n </tr>\n <tr>\n <td>eus_trivia</td>\n <td>acc</td>\n <td>52.36</td>\n </tr>\n <tr>\n <td>piqa_eu</td>\n <td>acc</td>\n <td>63.67</td>\n </tr>\n <tr>\n <td>Reading Comprehension</td>\n <td>eus_reading</td>\n <td>acc</td>\n <td>33.52</td>\n </tr>\n <tr>\n <td>Translation</td>\n <td>flores_eu</td>\n <td>bleu</td>\n <td>16.95</td>\n </tr>\n</tbody></table>\n\n#### Galician\n\n<table><thead>\n <tr>\n <th>Category</th>\n <th>Task</th>\n <th>Metric</th>\n <th>Result</th>\n </tr></thead>\n<tbody>\n <tr>\n <td>Commonsense Reasoning</td>\n <td>xstorycloze_gl</td>\n <td>acc</td>\n <td>74.12</td>\n </tr>\n <tr>\n <td>NLI</td>\n <td>xnli_gl</td>\n <td>acc</td>\n <td>50.95</td>\n </tr>\n <tr>\n <td rowspan=\"2\">Paraphrasing</td>\n <td>parafrases_gl</td>\n <td>acc</td>\n <td>54.42</td>\n </tr>\n <tr>\n <td>paws_gl</td>\n <td>acc</td>\n <td>63.2</td>\n </tr>\n <tr>\n <td>QA</td>\n <td>openbookqa_gl</td>\n <td>acc</td>\n <td>34.4</td>\n </tr>\n <tr>\n <td>Translation</td>\n <td>flores_gl</td>\n <td>bleu</td>\n <td>27.75</td>\n </tr>\n</tbody>\n</table>\n\n#### English\n\n<table><thead>\n <tr>\n <th>Category</th>\n <th>Task</th>\n <th>Metric</th>\n <th>Result</th>\n </tr></thead>\n<tbody>\n <tr>\n <td rowspan=\"2\">Commonsense Reasoning</td>\n <td>copa</td>\n <td>acc</td>\n <td>91</td>\n </tr>\n <tr>\n <td>xstorycloze_en</td>\n <td>acc</td>\n <td>79.09</td>\n </tr>\n <tr>\n <td rowspan=\"2\">NLI</td>\n <td>wnli</td>\n <td>acc</td>\n <td>56.34</td>\n </tr>\n <tr>\n <td>xnli_en</td>\n <td>acc</td>\n <td>50</td>\n </tr>\n <tr>\n <td>Paraphrasing</td>\n <td>paws *</td>\n <td>acc</td>\n <td>64.05</td>\n </tr>\n <tr>\n <td rowspan=\"6\">QA</td>\n <td>arc_easy</td>\n <td>acc</td>\n <td>82.2</td>\n </tr>\n <tr>\n <td>arc_challenge</td>\n <td>acc</td>\n <td>52.82</td>\n </tr>\n <tr>\n <td>openbookqa</td>\n <td>acc</td>\n <td>36</td>\n </tr>\n <tr>\n <td>piqa</td>\n <td>acc</td>\n <td>80.03</td>\n </tr>\n <tr>\n <td>social_iqa</td>\n <td>acc</td>\n <td>50.31</td>\n </tr>\n <tr>\n <td>xquad_en **</td>\n <td>acc</td>\n <td>77.74</td>\n </tr>\n</tbody></table>\n\n\\* Current LM Evaluation Harness implementation is lacking correct pre-processing. These results are obtained with adequate pre-processing.\n\n\\*\\* This task is not yet available in the official Harness, we hope to add it soon.\n\n---\n\n## Ethical Considerations and Limitations\n\nWe examine the presence of undesired societal and cognitive biases present in this model using different benchmarks. For societal biases, \nwe test performance using the BBQ dataset (Parrish et al., 2022) in the original English and the Regard dataset (Sheng et al., 2019). \nWe report that while performance is high (accuracies between 0.69 and 0.87 depending on the social category) in disambiguated settings\nthe model performs very poorly in ambiguous settings, which is indicative of the presence of societal biases which need to be addressed in post-training phases.\n\nWe additionally analyse model generations using the Regard dataset and classifier in Catalan, Spanish, and English using backtranslation and manual revision of the \ntranslations. We find no statistically significant difference in regard between majority and minority groups for any regard types,\nwith the exception of negative regard in Catalan where model generations are actually slightly worse for social majorities. \nOur analyses on societal biases show that while these biases are capable of interfering with model performance as expressed in the results on the BBQ dataset,\ntheir tendency for representational harm is limited given the results of the Regard dataset. We highlight that our analyses of these biases are by no means exhaustive\nand are limited by the relative scarcity of adequate resources in all languages present in the training data. We aim to gradually extend and expand our analyses \nin future work.\n\nOur cognitive bias analysis focuses on positional effects in 0-shot settings, and majority class bias in few-shot settings.\nFor positional effects, we leverage the ARC Multiple Choice Question dataset (Clark et al., 2018).\nWe observe moderate to strong primacy effects, whereby the model shows a preference for answers towards the beginning of the list of provided answers.\nWe measure effects of majority class effects in few-shot settings using SST-2 (Socher et al., 2013). We detect moderate effects,\nimplying that outputs can be influenced by the prompts.\n\nWe highlight that these results can be expected from a pretrained model that has not yet been instruction-tuned or aligned.\nThese tests are performed in order to show the biases the model may contain.\nWe urge developers to take them into account and perform safety testing and tuning tailored to their specific applications of the model.\n\n---\n\n## Additional information\n\n### Author\nThe Language Technologies Unit from Barcelona Supercomputing Center.\n\n### Contact\nFor further information, please send an email to <langtech@bsc.es>.\n\n### Copyright\nCopyright(c) 2024 by Language Technologies Unit, Barcelona Supercomputing Center.\n\n### Funding\nThis work has been promoted and financed by the Government of Catalonia through the [Aina Project](https://projecteaina.cat/).\n\nThis work is funded by the _Ministerio para la Transformación Digital y de la Función Pública_ - Funded by EU – NextGenerationEU \nwithin the framework of [ILENIA Project](https://proyectoilenia.es/) with reference 2022/TL22/00215337.\n\n### Acknowledgements\n\nThis project has benefited from the contributions of numerous teams and institutions, mainly through data contributions, knowledge transfer or technical support. \n\nIn Catalonia, many institutions have been involved in the project. Our thanks to Òmnium Cultural, Parlament de Catalunya, Institut d'Estudis Aranesos, Racó Català, Vilaweb, ACN, Nació Digital, El món and Aquí Berguedà.\n\nAt the national level, we are especially grateful to our ILENIA project partners: CENID, HiTZ and CiTIUS for their participation. We also extend our genuine gratitude to the Spanish Senate and Congress, Fundación Dialnet, and the ‘Instituto Universitario de Sistemas Inteligentes y Aplicaciones Numéricas en Ingeniería (SIANI)’ of the University of Las Palmas de Gran Canaria. \n\nAt the international level, we thank the Welsh government, DFKI, Occiglot project, especially Malte Ostendorff, and The Common Crawl Foundation, especially Pedro Ortiz, for their collaboration. We would also like to give special thanks to the NVIDIA team, with whom we have met regularly, specially to: Ignacio Sarasua, Adam Henryk Grzywaczewski, Oleg Sudakov, Sergio Perez, Miguel Martinez, Felipes Soares and Meriem Bendris. Their constant support has been especially appreciated throughout the entire process.\n\nTheir valuable efforts have been instrumental in the development of this work.\n\n### Disclaimer\nBe aware that the model may contain biases or other unintended distortions. \nWhen third parties deploy systems or provide services based on this model, or use the model themselves, \nthey bear the responsibility for mitigating any associated risks and ensuring compliance with applicable regulations, \nincluding those governing the use of Artificial Intelligence.\n\nThe Barcelona Supercomputing Center, as the owner and creator of the model, shall not be held liable for any outcomes resulting from third-party use.\n\n### Citation\n```\n@misc{gonzalezagirre2025salamandratechnicalreport,\n title={Salamandra Technical Report}, \n author={Aitor Gonzalez-Agirre and Marc Pàmies and Joan Llop and Irene Baucells and Severino Da Dalt and Daniel Tamayo and José Javier Saiz and Ferran Espuña and Jaume Prats and Javier Aula-Blasco and Mario Mina and Adrián Rubio and Alexander Shvets and Anna Sallés and Iñaki Lacunza and Iñigo Pikabea and Jorge Palomar and Júlia Falcão and Lucía Tormo and Luis Vasquez-Reina and Montserrat Marimon and Valle Ruíz-Fernández and Marta Villegas},\n year={2025},\n eprint={2502.08489},\n archivePrefix={arXiv},\n primaryClass={cs.CL},\n url={https://arxiv.org/abs/2502.08489}, \n}\n```\n\n### License\n[Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0)\n\n## Model Index\n|Model|Base|Instruct|\n|:---:|:---:|:---:|\n|2B| [Link](https://huggingface.co/BSC-LT/salamandra-2b) | [Link](https://huggingface.co/BSC-LT/salamandra-2b-instruct) |\n|7B| [Link](https://huggingface.co/BSC-LT/salamandra-7b) | [Link](https://huggingface.co/BSC-LT/salamandra-7b-instruct) |\n|40B| [Link](https://huggingface.co/BSC-LT/ALIA-40b) | WiP |\n"},"matched_task":{"kind":"list like","value":["QUESTION_ANSWERING","TRANSLATION","SUMMARIZATION","PARAPHRASING"],"string":"[\n \"QUESTION_ANSWERING\",\n \"TRANSLATION\",\n \"SUMMARIZATION\",\n \"PARAPHRASING\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["BEAR","SCIELO"],"string":"[\n \"BEAR\",\n \"SCIELO\"\n]","isConversational":false}}},{"rowIdx":194,"cells":{"id":{"kind":"string","value":"HiTZ/GoLLIE-7B"},"author":{"kind":"string","value":"HiTZ"},"task_category":{"kind":"string","value":"text-generation"},"tags":{"kind":"list like","value":["transformers","pytorch","llama","text-generation","code","text-generation-inference","Information Extraction","IE","Named Entity Recogniton","Event Extraction","Relation Extraction","LLaMA","custom_code","en","dataset:ACE05","dataset:bc5cdr","dataset:conll2003","dataset:ncbi_disease","dataset:conll2012_ontonotesv5","dataset:rams","dataset:tacred","dataset:wnut_17","arxiv:2310.03668","license:llama2","autotrain_compatible","endpoints_compatible","region:us"],"string":"[\n \"transformers\",\n \"pytorch\",\n \"llama\",\n \"text-generation\",\n \"code\",\n \"text-generation-inference\",\n \"Information Extraction\",\n \"IE\",\n \"Named Entity Recogniton\",\n \"Event Extraction\",\n \"Relation Extraction\",\n \"LLaMA\",\n \"custom_code\",\n \"en\",\n \"dataset:ACE05\",\n \"dataset:bc5cdr\",\n \"dataset:conll2003\",\n \"dataset:ncbi_disease\",\n \"dataset:conll2012_ontonotesv5\",\n \"dataset:rams\",\n \"dataset:tacred\",\n \"dataset:wnut_17\",\n \"arxiv:2310.03668\",\n \"license:llama2\",\n \"autotrain_compatible\",\n \"endpoints_compatible\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2023-09-25T10:24:52","string":"2023-09-25T10:24:52"},"last_modified":{"kind":"timestamp","value":"2023-10-10T07:51:44","string":"2023-10-10T07:51:44"},"downloads":{"kind":"number","value":2215,"string":"2,215"},"likes":{"kind":"number","value":28,"string":"28"},"README":{"kind":"string","value":"---\ndatasets:\n- ACE05\n- bc5cdr\n- conll2003\n- ncbi_disease\n- conll2012_ontonotesv5\n- rams\n- tacred\n- wnut_17\nlanguage:\n- en\nlicense: llama2\nmetrics:\n- f1\npipeline_tag: text-generation\ntags:\n- code\n- text-generation-inference\n- Information Extraction\n- IE\n- Named Entity Recogniton\n- Event Extraction\n- Relation Extraction\n- LLaMA\n---\n\n<p align=\"center\">\n <br>\n <img src=\"https://github.com/hitz-zentroa/GoLLIE/raw/main/assets/GoLLIE.png\" style=\"height: 250px;\">\n <h2 align=\"center\"><b>G</b>uideline f<b>o</b>llowing <b>L</b>arge <b>L</b>anguage Model for <b>I</b>nformation <b>E</b>xtraction</h2>\n <br>\n\n\n# Model Card for GoLLIE 7B\n\n\n<p align=\"justify\">\nWe present GoLLIE, a Large Language Model trained to follow annotation guidelines. GoLLIE outperforms previous approaches on zero-shot Information Extraction and allows the user to perform inferences with annotation schemas defined on the fly. Different from previous approaches, GoLLIE is able to follow detailed definitions and does not only rely on the knowledge already encoded in the LLM. \n\n- 💻 Code: [https://github.com/osainz59/CoLLIE/](https://github.com/hitz-zentroa/GoLLIE)\n- 📒 Blog Post: [GoLLIE: Guideline-following Large Language Model for Information Extraction](https://hitz-zentroa.github.io/GoLLIE/)\n- 📖 Paper: [GoLLIE: Annotation Guidelines improve Zero-Shot Information-Extraction](https://arxiv.org/abs/2310.03668)\n- 🐕 GoLLIE Colection in the 🤗HuggingFace Hub: [HiTZ/gollie](https://huggingface.co/collections/HiTZ/gollie-651bf19ee315e8a224aacc4f)\n- 🚀 Example Jupyter Notebooks: [GoLLIE Notebooks](https://github.com/hitz-zentroa/GoLLIE/tree/main/notebooks)\n</p>\n\n<p align=\"center\">\n<img src=\"https://github.com/hitz-zentroa/GoLLIE/raw/main/assets/zero_shot_results.png\">\n</p>\n\n\n### Model Description\n\n- **Developed by:** [Oscar Sainz](https://osainz59.github.io/), [Iker García-Ferrero](https://ikergarcia1996.github.io/Iker-Garcia-Ferrero/), [Rodrigo Agerri](https://ragerri.github.io/), [Oier Lopez de Lacalle](https://oierldl.github.io/), [German Rigau](https://adimen.si.ehu.es/~rigau/) and [Eneko Agirre](https://eagirre.github.io/)\n- **Institution:** [HiTZ Basque Center for Language Technology](http://www.hitz.eus/) - [Ixa](https://www.ixa.eus/node/2?language=en), [University of the Basque Country UPV/EHU](https://www.ehu.eus/en/en-home)\n- **Model type:** Text Generation \n- **Language(s) (NLP):** English\n- **License:** LLaMA2 License for the base and merged model. Apache 2.0 for pre-trained LoRA Adapters\n- **Finetuned from model:** CODE-LLaMA2\n\n\n\n## Schema definition and inference example\n\nThe labels are represented as Python classes, and the guidelines or instructions are introduced as docstrings. The model start generating after the `result = [` line.\n```Python\n# Entity definitions\n@dataclass\nclass Launcher(Template):\n \"\"\"Refers to a vehicle designed primarily to transport payloads from the Earth's \n surface to space. Launchers can carry various payloads, including satellites, \n crewed spacecraft, and cargo, into various orbits or even beyond Earth's orbit. \n They are usually multi-stage vehicles that use rocket engines for propulsion.\"\"\"\n\n mention: str \n \"\"\"\n The name of the launcher vehicle. \n Such as: \"Sturn V\", \"Atlas V\", \"Soyuz\", \"Ariane 5\"\n \"\"\"\n space_company: str # The company that operates the launcher. Such as: \"Blue origin\", \"ESA\", \"Boeing\", \"ISRO\", \"Northrop Grumman\", \"Arianespace\"\n crew: List[str] # Names of the crew members boarding the Launcher. Such as: \"Neil Armstrong\", \"Michael Collins\", \"Buzz Aldrin\"\n \n\n@dataclass\nclass Mission(Template):\n \"\"\"Any planned or accomplished journey beyond Earth's atmosphere with specific objectives, \n either crewed or uncrewed. It includes missions to satellites, the International \n Space Station (ISS), other celestial bodies, and deep space.\"\"\"\n \n mention: str\n \"\"\"\n The name of the mission. \n Such as: \"Apollo 11\", \"Artemis\", \"Mercury\"\n \"\"\"\n date: str # The start date of the mission\n departure: str # The place from which the vehicle will be launched. Such as: \"Florida\", \"Houston\", \"French Guiana\"\n destination: str # The place or planet to which the launcher will be sent. Such as \"Moon\", \"low-orbit\", \"Saturn\"\n\n# This is the text to analyze\ntext = (\n \"The Ares 3 mission to Mars is scheduled for 2032. The Starship rocket build by SpaceX will take off from Boca Chica,\"\n \"carrying the astronauts Max Rutherford, Elena Soto, and Jake Martinez.\"\n)\n\n# The annotation instances that take place in the text above are listed here\nresult = [\n Mission(mention='Ares 3', date='2032', departure='Boca Chica', destination='Mars'),\n Launcher(mention='Starship', space_company='SpaceX', crew=['Max Rutherford', 'Elena Soto', 'Jake Martinez'])\n]\n```\n\n## How to Get Started with the Model\n\nPlease read our [🚀 Example Jupyter Notebooks](https://github.com/hitz-zentroa/GoLLIE/tree/main/notebooks) to get started with GoLLIE. \n\nThe best way to load the model is using our custom `load_model` fuction. However, you can also load them using the AutoModelForCausalLM class.\n\n**Important**: Our flash attention implementation has small numerical differences compared to the attention implementation in Huggingface.\nYou must use the flag `trust_remote_code=True` or you will get inferior results. Flash attention requires an available CUDA GPU. Running GOLLIE \npre-trained models on a CPU is not supported. We plan to address this in future releases. First, install flash attention 2:\n```bash\npip install flash-attn --no-build-isolation\npip install git+https://github.com/HazyResearch/flash-attention.git#subdirectory=csrc/rotary\n```\n\nThen you can load the model using\n\n```python\nimport torch\nfrom transformers import AutoTokenizer, AutoModelForCausalLM\n\ntokenizer = AutoTokenizer.from_pretrained(\"HiTZ/GoLLIE-7B\")\nmodel = AutoModelForCausalLM.from_pretrained(\"HiTZ/GoLLIE-7B\", trust_remote_code=True, torch_dtype=torch.bfloat16)\nmodel.to(\"cuda\")\n```\n\nRead our [🚀 Example Jupyter Notebooks](https://github.com/hitz-zentroa/GoLLIE/tree/main/notebooks) to learn how to easily define guidelines, generate model inputs and parse the output! \n\n\n\n### Training Data\n\nThis is the list of task used for training and evaluating GoLLIE. However, as demonstrated in the 🚀 [Create Custom Task notebook](https://github.com/hitz-zentroa/GoLLIE/blob/main/notebooks/Create%20Custom%20Task.ipynb) GoLLIE can perform a wide range of unseen tasks. \nFor more info, read our [📖Paper](https://arxiv.org/abs/2310.03668).\n\n<p align=\"center\">\n<img src=\"https://github.com/hitz-zentroa/GoLLIE/raw/main/assets/datasets.png\">\n</p>\n\n\n## Evaluation\n\n| Model | Supervised average F1 | Zero-shot average F1 | 🤗HuggingFace Hub |\n|---|:---------------------:|:--------------------:|:---------------------------------------------------------:|\n| GoLLIE-7B | 73.0 | 55.3 | [HiTZ/GoLLIE-7B](https://huggingface.co/HiTZ/GoLLIE-7B) |\n| GoLLIE-13B | 73.9 | 56.0 | [HiTZ/GoLLIE-13B](https://huggingface.co/HiTZ/GoLLIE-13B) |\n| GoLLIE-34B | **75.0** | **57.2** | [HiTZ/GoLLIE-34B](https://huggingface.co/HiTZ/GoLLIE-34B) |\n\n\n## Environmental Impact\n\n| Model | Hardware | FLOPs | Time (h) | CO<sup>2</sup>eq (kg) |\n|----------------|-------------------|---------------------------|-------------------|-------------------------------------|\n| GoLLIE 7B | 1xA100 | 11.9e<sup>18</sup> | 44.5 | 1.57 |\n| GoLLIE 13B | 1xA100 | 22.7e<sup>18</sup> | 79.5 | 2.80 |\n| GoLLIE 34B | 2xA100 | 55.8e<sup>18</sup> | 94.6 | 6.67 |\n\n\n\n## Citation\n```\n@misc{sainz2023gollie,\n title={GoLLIE: Annotation Guidelines improve Zero-Shot Information-Extraction}, \n author={Oscar Sainz and Iker García-Ferrero and Rodrigo Agerri and Oier Lopez de Lacalle and German Rigau and Eneko Agirre},\n year={2023},\n eprint={2310.03668},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n```"},"matched_task":{"kind":"list like","value":["RELATION_EXTRACTION","EVENT_EXTRACTION"],"string":"[\n \"RELATION_EXTRACTION\",\n \"EVENT_EXTRACTION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["BC5CDR","NCBI DISEASE"],"string":"[\n \"BC5CDR\",\n \"NCBI DISEASE\"\n]","isConversational":false}}},{"rowIdx":195,"cells":{"id":{"kind":"string","value":"GoToCompany/gemma2-9b-cpt-sahabatai-v1-instruct"},"author":{"kind":"string","value":"GoToCompany"},"task_category":{"kind":"null"},"tags":{"kind":"list like","value":["safetensors","gemma2","en","id","jv","su","arxiv:2309.06085","arxiv:2310.04928","arxiv:2311.07911","base_model:GoToCompany/gemma2-9b-cpt-sahabatai-v1-base","base_model:finetune:GoToCompany/gemma2-9b-cpt-sahabatai-v1-base","license:gemma","region:us"],"string":"[\n \"safetensors\",\n \"gemma2\",\n \"en\",\n \"id\",\n \"jv\",\n \"su\",\n \"arxiv:2309.06085\",\n \"arxiv:2310.04928\",\n \"arxiv:2311.07911\",\n \"base_model:GoToCompany/gemma2-9b-cpt-sahabatai-v1-base\",\n \"base_model:finetune:GoToCompany/gemma2-9b-cpt-sahabatai-v1-base\",\n \"license:gemma\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2024-11-06T04:51:58","string":"2024-11-06T04:51:58"},"last_modified":{"kind":"timestamp","value":"2024-11-06T04:51:58","string":"2024-11-06T04:51:58"},"downloads":{"kind":"number","value":2211,"string":"2,211"},"likes":{"kind":"number","value":35,"string":"35"},"README":{"kind":"string","value":"---\nbase_model:\n- GoToCompany/gemma2-9b-cpt-sahabatai-v1-base\nlanguage:\n- en\n- id\n- jv\n- su\nlicense: gemma\n---\n# Gemma2 9B CPT Sahabat-AI v1 Instruct\n\n**Sahabat-AI** (Indonesian language for “close friends”) is a collection of Large Language Models (LLMs) which has been pretrained and instruct-tuned for Indonesian language and its various dialects. Sahabat-AI ecosystem is co-initiated by Indonesian tech and telecommunication companies: GoTo Group and Indosat Ooredoo Hutchison.\n\nGemma2 9B CPT Sahabat-AI v1 Instruct is an Indonesian-focused model which has been fine-tuned with around **448,000 Indonesian instruction-completion pairs** alongside an Indonesian-dialect pool consisting of **96,000 instruction-completion pairs in Javanese** and **98,000 instruction-completion pairs in Sundanese**. Additionally, we added a pool of **129,000 instruction-completion pairs in English**.\n\n- **Co-initiated by:** PT GoTo Gojek Tokopedia Tbk, Indosat Ooredoo Hutchison\n- **Developed by:** PT GoTo Gojek Tokopedia Tbk, AI Singapore\n- **Model type:** Decoder\n- **Languages:** English, Indonesian, Javanese, Sundanese\n- **License:** [Gemma Community License](https://ai.google.dev/gemma/terms)\n\n## Model Details\n\n### Model Description\nWe performed instruction tuning in Indonesian, Javanese, Sundanese as well as English on our [continued pre-trained Gemma2 9B CPT Sahabat-AI v1](https://huggingface.co/GoToCompany/gemma2-9b-cpt-sahabatai-v1-base), a decoder model using the Gemma2 architecture, to create Gemma2 9B CPT Sahabat-AI v1 Instruct.\n\nFor tokenisation, the model employs the default tokenizer used in Gemma-2-9B. The model has a context length of 8192.\n\n### Benchmark Performance\nWe evaluated Gemma2 9B CPT Sahabat-AI V1 Instruct on both general language capabilities and instruction-following capabilities.\n\n#### General Language Capabilities\nFor the evaluation of general language capabilities, we employed the \n- [SEA HELM (also known as BHASA) evaluation benchmark](https://arxiv.org/abs/2309.06085v2) across a variety of tasks.\n - These tasks include Question Answering (QA), Sentiment Analysis (Sentiment), Toxicity Detection (Toxicity), Translation in both directions (Eng>Lang & Lang>Eng), Abstractive Summarization (Summ), Causal Reasoning (Causal) and Natural Language Inference (NLI).\n - We also added support for Javanese and Sundanese for the BHASA tasks whenever applicable\n- [IndoMMLU](https://arxiv.org/pdf/2310.04928)\n - These tasks include examination questions on Humanities, Indonesian language, Local languages and cultures, Social science and STEM across primary, middle, and high school levels.\n- and the common English tasks from the [HuggingFace LLM Leaderboard](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard).\n - These tasks consist of [IFEval, BBH, Math Lvl 5, GPQA, MuSR, and MMLU-PRO.](https://huggingface.co/docs/leaderboards/open_llm_leaderboard/about)\n - **Caveat**: Our results differ from the HuggingFace LLM Leaderboard because we have used [VLLM](https://docs.vllm.ai/en/latest/) as our inference platform. VLLM caps the context size at **4096 tokens** while HuggingFace was set to **8192 tokens**.\n\nNote: SEA HELM is implemented using prompts to elicit answers in a strict format. For all tasks, the model is expected to provide an answer tag from which the answer is automatically extracted. For tasks where options are provided, the answer should comprise one of the pre-defined options. The scores for each task is normalised to account for baseline performance due to random chance.\n\nThe evaluation was done **zero-shot** with native prompts on a sample of 100-1000 instances for each dataset.\n\n\n#### Instruction-following Capabilities\nSince Gemma2 9B CPT Sahabat-AI v1 Instruct is an instruction-following model, we also evaluated it on instruction-following capabilities with the [IFEval](https://arxiv.org/abs/2311.07911) dataset.\n\n As this dataset was in English, the linguists and native speakers in the team worked together to filter, localize and translate the dataset into the respective target languages to ensure that the examples remained reasonable, meaningful and natural.\n\n**IFEval**\n\nIFEval evaluates a model's ability to adhere to constraints provided in the prompt, for example beginning a response with a specific word/phrase or answering with a certain number of sections. Additionally, accuracy is normalized by the proportion of responses in the correct language (if the model performs the task correctly but responds in the wrong language, it is judged to have failed the task).\n\n*Note*: IFEval was only used on Bahasa Indonesia. We are currently working on adding it for Javanese and Sundanese for our upcoming releases.\n\n#### Results\n\n#### Indonesian Results\n#### SEA HELM (also known as BHASA)\n<table style=\"border-collapse: collapse; width: 100%; font-size: 10px\">\n <tr>\n <th style=\"border: 2px solid black; padding: 8px; font-weight: bold;\">Language / Model Name [Instruct]</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Qwen2-7B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Qwen2.5-7B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Llama-3-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Llama-3.1-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">sea-lionv2.1-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">gemma-2-9B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">sahabatai-v1-8B</th>\n <th style=\"border: 2px solid black; padding: 8px;\">sahabatai-v1-9B</th>\n </tr>\n <tr>\n <td style=\"border: 2px solid black; padding: 8px; font-weight: bold;\">Overall (Bahasa Indonesia + Javanese + Sundanese)</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">36.963</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">42.988</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">37.805</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">45.866</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">46.880</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">56.359</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">53.725</td>\n <td style=\"border: 2px solid black; padding: 8px; background-color: lightgreen;\">61.169</td>\n </tr>\n <tr>\n <td style=\"border: 2px solid black; padding: 8px; font-weight: bold;\">Bahasa Indonesia</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">46.760</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">60.372</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">42.022</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">51.944</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">54.579</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">63.394</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">57.221</td>\n <td style=\"border: 2px solid black; padding: 8px; background-color: lightgreen;\">64.154</td>\n </tr>\n <tr>\n <td style=\"border: 2px solid black; padding: 8px; font-weight: bold;\">Javanese</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">33.956</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">40.625</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">41.739</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">47.587</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">48.012</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">56.468</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">56.460</td>\n <td style=\"border: 2px solid black; padding: 8px; background-color: lightgreen;\">64.439</td>\n </tr>\n <tr>\n <td style=\"border: 2px solid black; padding: 8px; font-weight: bold;\">Sundanese</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">30.173</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">27.969</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">29.654</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">38.068</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">38.050</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">49.216</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">47.495</td>\n <td style=\"border: 2px solid black; padding: 8px; background-color: lightgreen;\">54.913</td>\n </tr>\n</table>\n\n#### IndoMMLU\n<table style=\"border-collapse: collapse; width: 100%; font-size: 10px\">\n <tr>\n <th style=\"border: 2px solid black; padding: 8px; font-weight: bold;\">Model Name [Instruct]</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Qwen2-7B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Qwen2.5-7B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Meta-Llama-3-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Llama-3.1-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">sea-lionv2.1-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">gemma-2-9B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">sahabatai-v1-8B</th>\n <th style=\"border: 2px solid black; padding: 8px;\">sahabatai-v1-9B</th>\n </tr>\n <tr>\n <td style=\"border: 2px solid black; padding: 8px; font-weight: bold;\">Overall Results</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">53.0%</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">56.0%</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">51.9%</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">53.8%</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">54.4%</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">61.4%</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">55.6%</td>\n <td style=\"border: 2px solid black; padding: 8px; background-color: lightgreen;\">62.6%</td>\n </tr>\n</table>\n\n\n#### English Results\n<table style=\"border-collapse: collapse; width: 100%; font-size: 10px\">\n <tr>\n <th style=\"border: 2px solid black; padding: 8px;\">Model Name [Instruct]</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Qwen2-7B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Qwen2.5-7B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Llama-3-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">Llama-3.1-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">sea-lionv2.1-8B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">gemma-2-9B</th>\n <th style=\"border: 1px solid gray; padding: 8px;\">sahabatai-v1-8B</th>\n <th style=\"border: 2px solid black; padding: 8px;\">sahabatai-v1-9B</th>\n </tr>\n <tr>\n <td style=\"border: 2px solid black; padding: 8px; font-weight: bold;\">Average</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">24.48</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">27.75</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">23.91</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">27.98</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">24.52</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">26.44</td>\n <td style=\"border: 1px solid gray; padding: 8px;\">24.43</td>\n <td style=\"border: 1px solid black; padding: 8px; background-color: lightgreen;\">33.67</td>\n </tr>\n</table>\n\nGemma2 9B CPT Sahabat-AI v1 Instruct can be run using the 🤗 Transformers library \n```python\n# Please use transformers==4.45.0\n\nimport torch\nimport transformers\n\nmodel_id = \"GoToCompany/gemma2-9b-cpt-sahabatai-v1-instruct\"\n\npipeline = transformers.pipeline(\n \"text-generation\",\n model=model_id,\n model_kwargs={\"torch_dtype\": torch.bfloat16},\n device_map=\"auto\",\n)\n\nterminators = [\n pipeline.tokenizer.eos_token_id,\n pipeline.tokenizer.convert_tokens_to_ids(\"<|eot_id|>\")\n]\n\n# Javanese\nmessages = [\n {\"role\": \"user\", \"content\": \"Sopo wae sing ana ing Punakawan?\"}\n]\n\noutputs = pipeline(\n messages,\n max_new_tokens=256,\n eos_token_id=terminators,\n)\nprint(outputs[0][\"generated_text\"][-1])\n\n\n# Sundanese\nmessages = [\n {\"role\": \"user\", \"content\": \"Kumaha caritana si Kabayan?\"},\n]\n\noutputs = pipeline(\n messages,\n max_new_tokens=256,\n eos_token_id=terminators,\n)\nprint(outputs[0][\"generated_text\"][-1])\n```\n\n### Caveats\nIt is important for users to be aware that our model exhibits certain limitations that warrant consideration. Like many LLMs, the model can hallucinate and occasionally generates irrelevant content, introducing fictional elements that are not grounded in the provided context. Users should also exercise caution in interpreting and validating the model's responses due to the potential inconsistencies in its reasoning.\n\n## Limitations\n### Safety\n\nCurrent Sahabat-AI models, including this commercially permissive release, have not been aligned for safety. Developers and users should perform their own safety fine-tuning and related security measures. In no event shall the authors be held liable for any claim, damages, or other liability arising from the use of the released weights and codes.\n\n## Technical Specifications\n### Fine-Tuning Details\nGemma2 9B CPT Sahabat-AI v1 Instruct was built using a combination of a full parameter fine-tune, on-policy alignment, and model merges of the best performing checkpoints. The training process for fine-tuning was approximately 4 hours, with alignment taking 2 hours, both on 8x H100-80GB GPUs.\n\n## Data\nGemma2 9B CPT Sahabat-AI v1 Instruct was trained on a wide range of synthetic instructions, alongside publicly available instructions hand-curated by the team with the assistance of native speakers. In addition, special care was taken to ensure that the datasets used had commercially permissive licenses through verification with the original data source. \n\n## Call for Collaboration \n\nSahabat-AI (Indonesian language for “close friends”) a **local open source Large Language Model (LLM) ecosystem in Indonesian language**, co-initiated by Indonesian tech and telecommunication companies: GoTo Group and Indosat Ooredoo Hutchison. \nSahabat-AI ecosystem aims to empower Indonesians who want to develop AI-based services and applications using Bahasa Indonesia and its various local dialects. \n\nWe are supported by research centers and global tech experts such as AI Singapore and Tech Mahendra to train the model to gain general language understanding.\n\nWe also collaborate with key top Indonesia universities such as University of Indonesia, Gadjah Mada University, Bogor Institute of Agriculture, Bandung Institute of Technology, including top Indonesia media groups, such as Kompas Gramedia Group and Republika to train and enrich the model in Bahasa Indonesia, ensuring optimum provision of local context and cultural relevance.\n\nWe would like to invite **researchers, developers, and language enthusiasts** to actively contribute to the enhancement and expansion of Sahabat-AI. \nYour collaborations can involve:\n- Identifying and reporting technical issues\n- Sharing pre-training, instruction, and preference data\n- Improving documentation usability\n- Proposing and implementing new model evaluation tasks and metrics \n\nJoin us in shaping the future of Sahabat-AI by sharing your expertise and insights to make these models more accessible, accurate, and versatile.\n\nYou can contribute your ideas through [this form.](https://docs.google.com/forms/d/1_us969eQtEooYOn4XkvGkdP5VHOyCbO6L_sd9kTMnaA/edit)\n\n## The Development Team (in ascending alphabetical order)\n\n### AI Singapore\nChan Adwin<br>\nCheng Nicholas<br>\nChoa Esther<br>\nHuang Yuli<br>\nLau Wayne<br>\nLee Chwan Ren<br>\nLeong Wai Yi<br>\nLeong Wei Qi<br>\nLimkonchotiwat Peerat<br>\nLiu Bing Jie Darius<br>\nMontalan Jann Railey<br>\nNg Boon Cheong Raymond<br>\nNgui Jian Gang<br>\nNguyen Thanh Ngan<br>\nOng Brandon<br>\nOng Tat-Wee David<br>\nOng Zhi Hao<br>\nRengarajan Hamsawardhini<br>\nSiow Bryan<br>\nSusanto Yosephine<br>\nTai Ngee Chia<br>\nTan Choon Meng<br>\nTeng Walter<br>\nTeo Eng Sipp Leslie<br>\nTeo Wei Yi<br>\nTjhi William<br>\nYeo Yeow Tong<br>\nYong Xianbin<br>\n\n### PT GoTo Gojek Tokopedia Tbk\nAnissa Dininta<br>\nChau Shiau Ching<br>\nChoiri Hendra Hadhil<br>\nGoel Priyank<br>\nSaini Ajay Kumar<br>\nShalev Ofir<br>\nTan Daryl<br>\nTep Kilian Rithi<br>\nTiwari Anupam<br>\nWidjojo Daniel<br>\n\n## Acknowledgements\n\n [AI Singapore](https://aisingapore.org/) is a national programme supported by the National Research Foundation, Singapore and hosted by the National University of Singapore. \n\nAny opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not reflect the views of the National Research Foundation or the National University of Singapore.\n\n\n## Contact\n\nFor more info, please contact us using this [Sahabat-AI Inquiry Form.](https://docs.google.com/forms/d/1_us969eQtEooYOn4XkvGkdP5VHOyCbO6L_sd9kTMnaA/edit)\n\n## Disclaimer\n\nThis is the repository for the Instruct model.\nThe model has _not_ been aligned for safety.\nDevelopers and users should perform their own safety fine-tuning and related security measures.\nIn no event shall the authors be held liable for any claim, damages, or other liability arising from the use of the released weights and codes.\n\n\n## References\n### IndoMMLU Reference\n\n```bibtex\n@inproceedings{koto-etal-2023-indommlu,\n title = \"Large Language Models Only Pass Primary School Exams in {I}ndonesia: A Comprehensive Test on {I}ndo{MMLU}\",\n author = \"Fajri Koto and Nurul Aisyah and Haonan Li and Timothy Baldwin\",\n booktitle = \"Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP)\",\n month = December,\n year = \"2023\",\n address = \"Singapore\",\n publisher = \"Association for Computational Linguistics\",\n}\n}\n```"},"matched_task":{"kind":"list like","value":["QUESTION_ANSWERING","TRANSLATION","SUMMARIZATION"],"string":"[\n \"QUESTION_ANSWERING\",\n \"TRANSLATION\",\n \"SUMMARIZATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["CHIA"],"string":"[\n \"CHIA\"\n]","isConversational":false}}},{"rowIdx":196,"cells":{"id":{"kind":"string","value":"aliakseilabanau/bge-small-en"},"author":{"kind":"string","value":"aliakseilabanau"},"task_category":{"kind":"null"},"tags":{"kind":"list like","value":["safetensors","openvino","bert","mteb","sentence transformers","en","arxiv:2311.13534","arxiv:2310.07554","arxiv:2309.07597","base_model:BAAI/bge-small-en","base_model:quantized:BAAI/bge-small-en","license:mit","region:us"],"string":"[\n \"safetensors\",\n \"openvino\",\n \"bert\",\n \"mteb\",\n \"sentence transformers\",\n \"en\",\n \"arxiv:2311.13534\",\n \"arxiv:2310.07554\",\n \"arxiv:2309.07597\",\n \"base_model:BAAI/bge-small-en\",\n \"base_model:quantized:BAAI/bge-small-en\",\n \"license:mit\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2024-11-19T18:27:44","string":"2024-11-19T18:27:44"},"last_modified":{"kind":"timestamp","value":"2025-02-04T13:09:46","string":"2025-02-04T13:09:46"},"downloads":{"kind":"number","value":2173,"string":"2,173"},"likes":{"kind":"number","value":0,"string":"0"},"README":{"kind":"string","value":"---\nbase_model:\n- BAAI/bge-small-en\nlanguage:\n- en\nlicense: mit\ntags:\n- mteb\n- sentence transformers\n- openvino\nbase_model_relation: quantized\n---\n\nThe [BAAI/bge-small-en](https://huggingface.co/BAAI/bge-small-en) converted for [openvino backend](https://sbert.net/docs/sentence_transformer/usage/efficiency.html#openvino).\n\n```\nfrom sentence_transformers import SentenceTransformer\n\nmodel = SentenceTransformer(\"aliakseilabanau/bge-small-en\", backend=\"openvino\")\n\nsentences = [\"This is an example sentence\", \"Each sentence is converted\"]\nembeddings = model.encode(sentences)\n```\n\n\nThe original model card is below:\n\n--------\n\n**Recommend switching to newest [BAAI/bge-small-en-v1.5](https://huggingface.co/BAAI/bge-small-en-v1.5), which has more reasonable similarity distribution and same method of usage.**\n\n<h1 align=\"center\">FlagEmbedding</h1>\n\n\n<h4 align=\"center\">\n <p>\n <a href=#model-list>Model List</a> | \n <a href=#frequently-asked-questions>FAQ</a> |\n <a href=#usage>Usage</a> |\n <a href=\"#evaluation\">Evaluation</a> |\n <a href=\"#train\">Train</a> |\n <a href=\"#citation\">Citation</a> |\n <a href=\"#license\">License</a> \n <p>\n</h4>\n\nMore details please refer to our Github: [FlagEmbedding](https://github.com/FlagOpen/FlagEmbedding).\n\n\n[English](README.md) | [中文](https://github.com/FlagOpen/FlagEmbedding/blob/master/README_zh.md)\n\nFlagEmbedding focus on retrieval-augmented LLMs, consisting of following projects currently:\n\n- **Fine-tuning of LM** : [LM-Cocktail](https://github.com/FlagOpen/FlagEmbedding/tree/master/LM_Cocktail)\n- **Dense Retrieval**: [LLM Embedder](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/llm_embedder), [BGE Embedding](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/baai_general_embedding), [C-MTEB](https://github.com/FlagOpen/FlagEmbedding/tree/master/C_MTEB)\n- **Reranker Model**: [BGE Reranker](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/reranker)\n\n\n## News \n\n- 11/23/2023: Release [LM-Cocktail](https://github.com/FlagOpen/FlagEmbedding/tree/master/LM_Cocktail), a method to maintain general capabilities during fine-tuning by merging multiple language models. [Technical Report](https://arxiv.org/abs/2311.13534) :fire: \n- 10/12/2023: Release [LLM-Embedder](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/llm_embedder), a unified embedding model to support diverse retrieval augmentation needs for LLMs. [Technical Report](https://arxiv.org/pdf/2310.07554.pdf)\n- 09/15/2023: The [technical report](https://arxiv.org/pdf/2309.07597.pdf) of BGE has been released \n- 09/15/2023: The [massive training data](https://data.baai.ac.cn/details/BAAI-MTP) of BGE has been released \n- 09/12/2023: New models: \n - **New reranker model**: release cross-encoder models `BAAI/bge-reranker-base` and `BAAI/bge-reranker-large`, which are more powerful than embedding model. We recommend to use/fine-tune them to re-rank top-k documents returned by embedding models. \n - **update embedding model**: release `bge-*-v1.5` embedding model to alleviate the issue of the similarity distribution, and enhance its retrieval ability without instruction.\n \n\n<details>\n <summary>More</summary>\n<!-- ### More -->\n \n- 09/07/2023: Update [fine-tune code](https://github.com/FlagOpen/FlagEmbedding/blob/master/FlagEmbedding/baai_general_embedding/README.md): Add script to mine hard negatives and support adding instruction during fine-tuning. \n- 08/09/2023: BGE Models are integrated into **Langchain**, you can use it like [this](#using-langchain); C-MTEB **leaderboard** is [available](https://huggingface.co/spaces/mteb/leaderboard). \n- 08/05/2023: Release base-scale and small-scale models, **best performance among the models of the same size 🤗** \n- 08/02/2023: Release `bge-large-*`(short for BAAI General Embedding) Models, **rank 1st on MTEB and C-MTEB benchmark!** :tada: :tada: \n- 08/01/2023: We release the [Chinese Massive Text Embedding Benchmark](https://github.com/FlagOpen/FlagEmbedding/blob/master/C_MTEB) (**C-MTEB**), consisting of 31 test dataset. \n \n</details>\n\n\n## Model List\n\n`bge` is short for `BAAI general embedding`.\n\n| Model | Language | | Description | query instruction for retrieval [1] |\n|:-------------------------------|:--------:| :--------:| :--------:|:--------:|\n| [LM-Cocktail](https://huggingface.co/Shitao) | English | | fine-tuned models (Llama and BGE) which can be used to reproduce the results of LM-Cocktail | |\n| [BAAI/llm-embedder](https://huggingface.co/BAAI/llm-embedder) | English | [Inference](./FlagEmbedding/llm_embedder/README.md) [Fine-tune](./FlagEmbedding/llm_embedder/README.md) | a unified embedding model to support diverse retrieval augmentation needs for LLMs | See [README](./FlagEmbedding/llm_embedder/README.md) |\n| [BAAI/bge-reranker-large](https://huggingface.co/BAAI/bge-reranker-large) | Chinese and English | [Inference](#usage-for-reranker) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/reranker) | a cross-encoder model which is more accurate but less efficient [2] | |\n| [BAAI/bge-reranker-base](https://huggingface.co/BAAI/bge-reranker-base) | Chinese and English | [Inference](#usage-for-reranker) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/reranker) | a cross-encoder model which is more accurate but less efficient [2] | |\n| [BAAI/bge-large-en-v1.5](https://huggingface.co/BAAI/bge-large-en-v1.5) | English | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | version 1.5 with more reasonable similarity distribution | `Represent this sentence for searching relevant passages: ` |\n| [BAAI/bge-base-en-v1.5](https://huggingface.co/BAAI/bge-base-en-v1.5) | English | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | version 1.5 with more reasonable similarity distribution | `Represent this sentence for searching relevant passages: ` |\n| [BAAI/bge-small-en-v1.5](https://huggingface.co/BAAI/bge-small-en-v1.5) | English | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | version 1.5 with more reasonable similarity distribution | `Represent this sentence for searching relevant passages: ` |\n| [BAAI/bge-large-zh-v1.5](https://huggingface.co/BAAI/bge-large-zh-v1.5) | Chinese | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | version 1.5 with more reasonable similarity distribution | `为这个句子生成表示以用于检索相关文章：` |\n| [BAAI/bge-base-zh-v1.5](https://huggingface.co/BAAI/bge-base-zh-v1.5) | Chinese | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | version 1.5 with more reasonable similarity distribution | `为这个句子生成表示以用于检索相关文章：` |\n| [BAAI/bge-small-zh-v1.5](https://huggingface.co/BAAI/bge-small-zh-v1.5) | Chinese | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | version 1.5 with more reasonable similarity distribution | `为这个句子生成表示以用于检索相关文章：` |\n| [BAAI/bge-large-en](https://huggingface.co/BAAI/bge-large-en) | English | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | :trophy: rank **1st** in [MTEB](https://huggingface.co/spaces/mteb/leaderboard) leaderboard | `Represent this sentence for searching relevant passages: ` |\n| [BAAI/bge-base-en](https://huggingface.co/BAAI/bge-base-en) | English | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | a base-scale model but with similar ability to `bge-large-en` | `Represent this sentence for searching relevant passages: ` |\n| [BAAI/bge-small-en](https://huggingface.co/BAAI/bge-small-en) | English | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) |a small-scale model but with competitive performance | `Represent this sentence for searching relevant passages: ` |\n| [BAAI/bge-large-zh](https://huggingface.co/BAAI/bge-large-zh) | Chinese | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | :trophy: rank **1st** in [C-MTEB](https://github.com/FlagOpen/FlagEmbedding/tree/master/C_MTEB) benchmark | `为这个句子生成表示以用于检索相关文章：` |\n| [BAAI/bge-base-zh](https://huggingface.co/BAAI/bge-base-zh) | Chinese | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | a base-scale model but with similar ability to `bge-large-zh` | `为这个句子生成表示以用于检索相关文章：` |\n| [BAAI/bge-small-zh](https://huggingface.co/BAAI/bge-small-zh) | Chinese | [Inference](#usage-for-embedding-model) [Fine-tune](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) | a small-scale model but with competitive performance | `为这个句子生成表示以用于检索相关文章：` |\n\n\n[1\\]: If you need to search the relevant passages to a query, we suggest to add the instruction to the query; in other cases, no instruction is needed, just use the original query directly. In all cases, **no instruction** needs to be added to passages.\n\n[2\\]: Different from embedding model, reranker uses question and document as input and directly output similarity instead of embedding. To balance the accuracy and time cost, cross-encoder is widely used to re-rank top-k documents retrieved by other simple models. \nFor examples, use bge embedding model to retrieve top 100 relevant documents, and then use bge reranker to re-rank the top 100 document to get the final top-3 results.\n\nAll models have been uploaded to Huggingface Hub, and you can see them at https://huggingface.co/BAAI. \nIf you cannot open the Huggingface Hub, you also can download the models at https://model.baai.ac.cn/models .\n\n\n## Frequently asked questions\n\n<details>\n <summary>1. How to fine-tune bge embedding model?</summary>\n\n <!-- ### How to fine-tune bge embedding model? -->\nFollowing this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) to prepare data and fine-tune your model. \nSome suggestions:\n- Mine hard negatives following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune#hard-negatives), which can improve the retrieval performance.\n- If you pre-train bge on your data, the pre-trained model cannot be directly used to calculate similarity, and it must be fine-tuned with contrastive learning before computing similarity.\n- If the accuracy of the fine-tuned model is still not high, it is recommended to use/fine-tune the cross-encoder model (bge-reranker) to re-rank top-k results. Hard negatives also are needed to fine-tune reranker.\n\n \n</details>\n\n<details>\n <summary>2. The similarity score between two dissimilar sentences is higher than 0.5</summary>\n\n <!-- ### The similarity score between two dissimilar sentences is higher than 0.5 -->\n**Suggest to use bge v1.5, which alleviates the issue of the similarity distribution.** \n\nSince we finetune the models by contrastive learning with a temperature of 0.01, \nthe similarity distribution of the current BGE model is about in the interval \\[0.6, 1\\].\nSo a similarity score greater than 0.5 does not indicate that the two sentences are similar.\n\nFor downstream tasks, such as passage retrieval or semantic similarity, \n**what matters is the relative order of the scores, not the absolute value.**\nIf you need to filter similar sentences based on a similarity threshold, \nplease select an appropriate similarity threshold based on the similarity distribution on your data (such as 0.8, 0.85, or even 0.9).\n\n</details>\n\n<details>\n <summary>3. When does the query instruction need to be used</summary>\n\n <!-- ### When does the query instruction need to be used -->\n\nFor the `bge-*-v1.5`, we improve its retrieval ability when not using instruction. \nNo instruction only has a slight degradation in retrieval performance compared with using instruction. \nSo you can generate embedding without instruction in all cases for convenience.\n \nFor a retrieval task that uses short queries to find long related documents, \nit is recommended to add instructions for these short queries.\n**The best method to decide whether to add instructions for queries is choosing the setting that achieves better performance on your task.**\nIn all cases, the documents/passages do not need to add the instruction. \n\n</details>\n\n\n## Usage \n\n### Usage for Embedding Model\n\nHere are some examples for using `bge` models with \n[FlagEmbedding](#using-flagembedding), [Sentence-Transformers](#using-sentence-transformers), [Langchain](#using-langchain), or [Huggingface Transformers](#using-huggingface-transformers).\n\n#### Using FlagEmbedding\n```\npip install -U FlagEmbedding\n```\nIf it doesn't work for you, you can see [FlagEmbedding](https://github.com/FlagOpen/FlagEmbedding/blob/master/FlagEmbedding/baai_general_embedding/README.md) for more methods to install FlagEmbedding.\n\n```python\nfrom FlagEmbedding import FlagModel\nsentences_1 = [\"样例数据-1\", \"样例数据-2\"]\nsentences_2 = [\"样例数据-3\", \"样例数据-4\"]\nmodel = FlagModel('BAAI/bge-large-zh-v1.5', \n query_instruction_for_retrieval=\"为这个句子生成表示以用于检索相关文章：\",\n use_fp16=True) # Setting use_fp16 to True speeds up computation with a slight performance degradation\nembeddings_1 = model.encode(sentences_1)\nembeddings_2 = model.encode(sentences_2)\nsimilarity = embeddings_1 @ embeddings_2.T\nprint(similarity)\n\n# for s2p(short query to long passage) retrieval task, suggest to use encode_queries() which will automatically add the instruction to each query\n# corpus in retrieval task can still use encode() or encode_corpus(), since they don't need instruction\nqueries = ['query_1', 'query_2']\npassages = [\"样例文档-1\", \"样例文档-2\"]\nq_embeddings = model.encode_queries(queries)\np_embeddings = model.encode(passages)\nscores = q_embeddings @ p_embeddings.T\n```\nFor the value of the argument `query_instruction_for_retrieval`, see [Model List](https://github.com/FlagOpen/FlagEmbedding/tree/master#model-list). \n\nBy default, FlagModel will use all available GPUs when encoding. Please set `os.environ[\"CUDA_VISIBLE_DEVICES\"]` to select specific GPUs.\nYou also can set `os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"\"` to make all GPUs unavailable.\n\n\n#### Using Sentence-Transformers\n\nYou can also use the `bge` models with [sentence-transformers](https://www.SBERT.net):\n\n```\npip install -U sentence-transformers\n```\n```python\nfrom sentence_transformers import SentenceTransformer\nsentences_1 = [\"样例数据-1\", \"样例数据-2\"]\nsentences_2 = [\"样例数据-3\", \"样例数据-4\"]\nmodel = SentenceTransformer('BAAI/bge-large-zh-v1.5')\nembeddings_1 = model.encode(sentences_1, normalize_embeddings=True)\nembeddings_2 = model.encode(sentences_2, normalize_embeddings=True)\nsimilarity = embeddings_1 @ embeddings_2.T\nprint(similarity)\n```\nFor s2p(short query to long passage) retrieval task, \neach short query should start with an instruction (instructions see [Model List](https://github.com/FlagOpen/FlagEmbedding/tree/master#model-list)). \nBut the instruction is not needed for passages.\n```python\nfrom sentence_transformers import SentenceTransformer\nqueries = ['query_1', 'query_2']\npassages = [\"样例文档-1\", \"样例文档-2\"]\ninstruction = \"为这个句子生成表示以用于检索相关文章：\"\n\nmodel = SentenceTransformer('BAAI/bge-large-zh-v1.5')\nq_embeddings = model.encode([instruction+q for q in queries], normalize_embeddings=True)\np_embeddings = model.encode(passages, normalize_embeddings=True)\nscores = q_embeddings @ p_embeddings.T\n```\n\n#### Using Langchain \n\nYou can use `bge` in langchain like this:\n```python\nfrom langchain.embeddings import HuggingFaceBgeEmbeddings\nmodel_name = \"BAAI/bge-large-en-v1.5\"\nmodel_kwargs = {'device': 'cuda'}\nencode_kwargs = {'normalize_embeddings': True} # set True to compute cosine similarity\nmodel = HuggingFaceBgeEmbeddings(\n model_name=model_name,\n model_kwargs=model_kwargs,\n encode_kwargs=encode_kwargs,\n query_instruction=\"为这个句子生成表示以用于检索相关文章：\"\n)\nmodel.query_instruction = \"为这个句子生成表示以用于检索相关文章：\"\n```\n\n\n#### Using HuggingFace Transformers\n\nWith the transformers package, you can use the model like this: First, you pass your input through the transformer model, then you select the last hidden state of the first token (i.e., [CLS]) as the sentence embedding.\n\n```python\nfrom transformers import AutoTokenizer, AutoModel\nimport torch\n# Sentences we want sentence embeddings for\nsentences = [\"样例数据-1\", \"样例数据-2\"]\n\n# Load model from HuggingFace Hub\ntokenizer = AutoTokenizer.from_pretrained('BAAI/bge-large-zh-v1.5')\nmodel = AutoModel.from_pretrained('BAAI/bge-large-zh-v1.5')\nmodel.eval()\n\n# Tokenize sentences\nencoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')\n# for s2p(short query to long passage) retrieval task, add an instruction to query (not add instruction for passages)\n# encoded_input = tokenizer([instruction + q for q in queries], padding=True, truncation=True, return_tensors='pt')\n\n# Compute token embeddings\nwith torch.no_grad():\n model_output = model(**encoded_input)\n # Perform pooling. In this case, cls pooling.\n sentence_embeddings = model_output[0][:, 0]\n# normalize embeddings\nsentence_embeddings = torch.nn.functional.normalize(sentence_embeddings, p=2, dim=1)\nprint(\"Sentence embeddings:\", sentence_embeddings)\n```\n\n### Usage for Reranker\n\nDifferent from embedding model, reranker uses question and document as input and directly output similarity instead of embedding. \nYou can get a relevance score by inputting query and passage to the reranker. \nThe reranker is optimized based cross-entropy loss, so the relevance score is not bounded to a specific range.\n\n\n#### Using FlagEmbedding\n```\npip install -U FlagEmbedding\n```\n\nGet relevance scores (higher scores indicate more relevance):\n```python\nfrom FlagEmbedding import FlagReranker\nreranker = FlagReranker('BAAI/bge-reranker-large', use_fp16=True) # Setting use_fp16 to True speeds up computation with a slight performance degradation\n\nscore = reranker.compute_score(['query', 'passage'])\nprint(score)\n\nscores = reranker.compute_score([['what is panda?', 'hi'], ['what is panda?', 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.']])\nprint(scores)\n```\n\n\n#### Using Huggingface transformers\n\n```python\nimport torch\nfrom transformers import AutoModelForSequenceClassification, AutoTokenizer\n\ntokenizer = AutoTokenizer.from_pretrained('BAAI/bge-reranker-large')\nmodel = AutoModelForSequenceClassification.from_pretrained('BAAI/bge-reranker-large')\nmodel.eval()\n\npairs = [['what is panda?', 'hi'], ['what is panda?', 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.']]\nwith torch.no_grad():\n inputs = tokenizer(pairs, padding=True, truncation=True, return_tensors='pt', max_length=512)\n scores = model(**inputs, return_dict=True).logits.view(-1, ).float()\n print(scores)\n```\n\n## Evaluation \n\n`baai-general-embedding` models achieve **state-of-the-art performance on both MTEB and C-MTEB leaderboard!**\nFor more details and evaluation tools see our [scripts](https://github.com/FlagOpen/FlagEmbedding/blob/master/C_MTEB/README.md). \n\n- **MTEB**: \n\n| Model Name | Dimension | Sequence Length | Average (56) | Retrieval (15) |Clustering (11) | Pair Classification (3) | Reranking (4) | STS (10) | Summarization (1) | Classification (12) |\n|:----:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|\n| [BAAI/bge-large-en-v1.5](https://huggingface.co/BAAI/bge-large-en-v1.5) | 1024 | 512 | **64.23** | **54.29** | 46.08 | 87.12 | 60.03 | 83.11 | 31.61 | 75.97 | \n| [BAAI/bge-base-en-v1.5](https://huggingface.co/BAAI/bge-base-en-v1.5) | 768 | 512 | 63.55 | 53.25 | 45.77 | 86.55 | 58.86 | 82.4 | 31.07 | 75.53 | \n| [BAAI/bge-small-en-v1.5](https://huggingface.co/BAAI/bge-small-en-v1.5) | 384 | 512 | 62.17 |51.68 | 43.82 | 84.92 | 58.36 | 81.59 | 30.12 | 74.14 | \n| [bge-large-en](https://huggingface.co/BAAI/bge-large-en) | 1024 | 512 | 63.98 | 53.9 | 46.98 | 85.8 | 59.48 | 81.56 | 32.06 | 76.21 | \n| [bge-base-en](https://huggingface.co/BAAI/bge-base-en) | 768 | 512 | 63.36 | 53.0 | 46.32 | 85.86 | 58.7 | 81.84 | 29.27 | 75.27 | \n| [gte-large](https://huggingface.co/thenlper/gte-large) | 1024 | 512 | 63.13 | 52.22 | 46.84 | 85.00 | 59.13 | 83.35 | 31.66 | 73.33 |\n| [gte-base](https://huggingface.co/thenlper/gte-base) \t| 768 | 512 | 62.39 | 51.14 | 46.2 | 84.57 | 58.61 | 82.3 | 31.17 | 73.01 |\n| [e5-large-v2](https://huggingface.co/intfloat/e5-large-v2) | 1024| 512 | 62.25 | 50.56 | 44.49 | 86.03 | 56.61 | 82.05 | 30.19 | 75.24 |\n| [bge-small-en](https://huggingface.co/BAAI/bge-small-en) | 384 | 512 | 62.11 | 51.82 | 44.31 | 83.78 | 57.97 | 80.72 | 30.53 | 74.37 | \n| [instructor-xl](https://huggingface.co/hkunlp/instructor-xl) | 768 | 512 | 61.79 | 49.26 | 44.74 | 86.62 | 57.29 | 83.06 | 32.32 | 61.79 |\n| [e5-base-v2](https://huggingface.co/intfloat/e5-base-v2) | 768 | 512 | 61.5 | 50.29 | 43.80 | 85.73 | 55.91 | 81.05 | 30.28 | 73.84 |\n| [gte-small](https://huggingface.co/thenlper/gte-small) | 384 | 512 | 61.36 | 49.46 | 44.89 | 83.54 | 57.7 | 82.07 | 30.42 | 72.31 |\n| [text-embedding-ada-002](https://platform.openai.com/docs/guides/embeddings) | 1536 | 8192 | 60.99 | 49.25 | 45.9 | 84.89 | 56.32 | 80.97 | 30.8 | 70.93 |\n| [e5-small-v2](https://huggingface.co/intfloat/e5-base-v2) | 384 | 512 | 59.93 | 49.04 | 39.92 | 84.67 | 54.32 | 80.39 | 31.16 | 72.94 |\n| [sentence-t5-xxl](https://huggingface.co/sentence-transformers/sentence-t5-xxl) | 768 | 512 | 59.51 | 42.24 | 43.72 | 85.06 | 56.42 | 82.63 | 30.08 | 73.42 |\n| [all-mpnet-base-v2](https://huggingface.co/sentence-transformers/all-mpnet-base-v2) \t| 768 | 514 \t| 57.78 | 43.81 | 43.69 | 83.04 | 59.36 | 80.28 | 27.49 | 65.07 |\n| [sgpt-bloom-7b1-msmarco](https://huggingface.co/bigscience/sgpt-bloom-7b1-msmarco) \t| 4096 | 2048 | 57.59 | 48.22 | 38.93 | 81.9 | 55.65 | 77.74 | 33.6 | 66.19 |\n\n\n\n- **C-MTEB**: \nWe create the benchmark C-MTEB for Chinese text embedding which consists of 31 datasets from 6 tasks. \nPlease refer to [C_MTEB](https://github.com/FlagOpen/FlagEmbedding/blob/master/C_MTEB/README.md) for a detailed introduction.\n \n| Model | Embedding dimension | Avg | Retrieval | STS | PairClassification | Classification | Reranking | Clustering |\n|:-------------------------------|:--------:|:--------:|:--------:|:--------:|:--------:|:--------:|:--------:|:--------:|\n| [**BAAI/bge-large-zh-v1.5**](https://huggingface.co/BAAI/bge-large-zh-v1.5) | 1024 | **64.53** | 70.46 | 56.25 | 81.6 | 69.13 | 65.84 | 48.99 | \n| [BAAI/bge-base-zh-v1.5](https://huggingface.co/BAAI/bge-base-zh-v1.5) | 768 | 63.13 | 69.49 | 53.72 | 79.75 | 68.07 | 65.39 | 47.53 | \n| [BAAI/bge-small-zh-v1.5](https://huggingface.co/BAAI/bge-small-zh-v1.5) | 512 | 57.82 | 61.77 | 49.11 | 70.41 | 63.96 | 60.92 | 44.18 | \n| [BAAI/bge-large-zh](https://huggingface.co/BAAI/bge-large-zh) | 1024 | 64.20 | 71.53 | 54.98 | 78.94 | 68.32 | 65.11 | 48.39 |\n| [bge-large-zh-noinstruct](https://huggingface.co/BAAI/bge-large-zh-noinstruct) | 1024 | 63.53 | 70.55 | 53 | 76.77 | 68.58 | 64.91 | 50.01 |\n| [BAAI/bge-base-zh](https://huggingface.co/BAAI/bge-base-zh) | 768 | 62.96 | 69.53 | 54.12 | 77.5 | 67.07 | 64.91 | 47.63 |\n| [multilingual-e5-large](https://huggingface.co/intfloat/multilingual-e5-large) | 1024 | 58.79 | 63.66 | 48.44 | 69.89 | 67.34 | 56.00 | 48.23 |\n| [BAAI/bge-small-zh](https://huggingface.co/BAAI/bge-small-zh) | 512 | 58.27 | 63.07 | 49.45 | 70.35 | 63.64 | 61.48 | 45.09 |\n| [m3e-base](https://huggingface.co/moka-ai/m3e-base) | 768 | 57.10 | 56.91 | 50.47 | 63.99 | 67.52 | 59.34 | 47.68 |\n| [m3e-large](https://huggingface.co/moka-ai/m3e-large) | 1024 | 57.05 | 54.75 | 50.42 | 64.3 | 68.2 | 59.66 | 48.88 |\n| [multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base) | 768 | 55.48 | 61.63 | 46.49 | 67.07 | 65.35 | 54.35 | 40.68 |\n| [multilingual-e5-small](https://huggingface.co/intfloat/multilingual-e5-small) | 384 | 55.38 | 59.95 | 45.27 | 66.45 | 65.85 | 53.86 | 45.26 |\n| [text-embedding-ada-002(OpenAI)](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings) | 1536 | 53.02 | 52.0 | 43.35 | 69.56 | 64.31 | 54.28 | 45.68 |\n| [luotuo](https://huggingface.co/silk-road/luotuo-bert-medium) | 1024 | 49.37 | 44.4 | 42.78 | 66.62 | 61 | 49.25 | 44.39 |\n| [text2vec-base](https://huggingface.co/shibing624/text2vec-base-chinese) | 768 | 47.63 | 38.79 | 43.41 | 67.41 | 62.19 | 49.45 | 37.66 |\n| [text2vec-large](https://huggingface.co/GanymedeNil/text2vec-large-chinese) | 1024 | 47.36 | 41.94 | 44.97 | 70.86 | 60.66 | 49.16 | 30.02 |\n\n\n- **Reranking**:\nSee [C_MTEB](https://github.com/FlagOpen/FlagEmbedding/blob/master/C_MTEB/) for evaluation script.\n\n| Model | T2Reranking | T2RerankingZh2En\\* | T2RerankingEn2Zh\\* | MMarcoReranking | CMedQAv1 | CMedQAv2 | Avg | \n|:-------------------------------|:--------:|:--------:|:--------:|:--------:|:--------:|:--------:|:--------:| \n| text2vec-base-multilingual | 64.66 | 62.94 | 62.51 | 14.37 | 48.46 | 48.6 | 50.26 | \n| multilingual-e5-small | 65.62 | 60.94 | 56.41 | 29.91 | 67.26 | 66.54 | 57.78 | \n| multilingual-e5-large | 64.55 | 61.61 | 54.28 | 28.6 | 67.42 | 67.92 | 57.4 | \n| multilingual-e5-base | 64.21 | 62.13 | 54.68 | 29.5 | 66.23 | 66.98 | 57.29 | \n| m3e-base | 66.03 | 62.74 | 56.07 | 17.51 | 77.05 | 76.76 | 59.36 | \n| m3e-large | 66.13 | 62.72 | 56.1 | 16.46 | 77.76 | 78.27 | 59.57 | \n| bge-base-zh-v1.5 | 66.49 | 63.25 | 57.02 | 29.74 | 80.47 | 84.88 | 63.64 | \n| bge-large-zh-v1.5 | 65.74 | 63.39 | 57.03 | 28.74 | 83.45 | 85.44 | 63.97 | \n| [BAAI/bge-reranker-base](https://huggingface.co/BAAI/bge-reranker-base) | 67.28 | 63.95 | 60.45 | 35.46 | 81.26 | 84.1 | 65.42 | \n| [BAAI/bge-reranker-large](https://huggingface.co/BAAI/bge-reranker-large) | 67.6 | 64.03 | 61.44 | 37.16 | 82.15 | 84.18 | 66.09 | \n\n\\* : T2RerankingZh2En and T2RerankingEn2Zh are cross-language retrieval tasks\n\n## Train\n\n### BAAI Embedding \n\nWe pre-train the models using [retromae](https://github.com/staoxiao/RetroMAE) and train them on large-scale pairs data using contrastive learning. \n**You can fine-tune the embedding model on your data following our [examples](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune).**\nWe also provide a [pre-train example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/pretrain).\nNote that the goal of pre-training is to reconstruct the text, and the pre-trained model cannot be used for similarity calculation directly, it needs to be fine-tuned.\nMore training details for bge see [baai_general_embedding](https://github.com/FlagOpen/FlagEmbedding/blob/master/FlagEmbedding/baai_general_embedding/README.md).\n\n\n\n### BGE Reranker\n\nCross-encoder will perform full-attention over the input pair, \nwhich is more accurate than embedding model (i.e., bi-encoder) but more time-consuming than embedding model.\nTherefore, it can be used to re-rank the top-k documents returned by embedding model.\nWe train the cross-encoder on a multilingual pair data, \nThe data format is the same as embedding model, so you can fine-tune it easily following our [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/reranker). \nMore details please refer to [./FlagEmbedding/reranker/README.md](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/reranker)\n\n\n\n\n## Citation\n\nIf you find this repository useful, please consider giving a star :star: and citation\n\n```\n@misc{bge_embedding,\n title={C-Pack: Packaged Resources To Advance General Chinese Embedding}, \n author={Shitao Xiao and Zheng Liu and Peitian Zhang and Niklas Muennighoff},\n year={2023},\n eprint={2309.07597},\n archivePrefix={arXiv},\n primaryClass={cs.CL}\n}\n```\n\n## License\nFlagEmbedding is licensed under the [MIT License](https://github.com/FlagOpen/FlagEmbedding/blob/master/LICENSE). The released models can be used for commercial purposes free of charge."},"matched_task":{"kind":"list like","value":["SEMANTIC_SIMILARITY","SUMMARIZATION"],"string":"[\n \"SEMANTIC_SIMILARITY\",\n \"SUMMARIZATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["BEAR"],"string":"[\n \"BEAR\"\n]","isConversational":false}}},{"rowIdx":197,"cells":{"id":{"kind":"string","value":"JosephusCheung/Guanaco"},"author":{"kind":"string","value":"JosephusCheung"},"task_category":{"kind":"string","value":"text-generation"},"tags":{"kind":"list like","value":["transformers","pytorch","llama","text-generation","guannaco","alpaca","conversational","en","zh","ja","de","dataset:JosephusCheung/GuanacoDataset","doi:10.57967/hf/0607","license:gpl-3.0","autotrain_compatible","text-generation-inference","region:us"],"string":"[\n \"transformers\",\n \"pytorch\",\n \"llama\",\n \"text-generation\",\n \"guannaco\",\n \"alpaca\",\n \"conversational\",\n \"en\",\n \"zh\",\n \"ja\",\n \"de\",\n \"dataset:JosephusCheung/GuanacoDataset\",\n \"doi:10.57967/hf/0607\",\n \"license:gpl-3.0\",\n \"autotrain_compatible\",\n \"text-generation-inference\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2023-04-08T03:03:14","string":"2023-04-08T03:03:14"},"last_modified":{"kind":"timestamp","value":"2023-05-29T12:48:21","string":"2023-05-29T12:48:21"},"downloads":{"kind":"number","value":2164,"string":"2,164"},"likes":{"kind":"number","value":230,"string":"230"},"README":{"kind":"string","value":"---\ndatasets:\n- JosephusCheung/GuanacoDataset\nlanguage:\n- en\n- zh\n- ja\n- de\nlicense: gpl-3.0\npipeline_tag: conversational\ntags:\n- llama\n- guannaco\n- alpaca\ninference: false\n---\n\n\n\n**You can run on Colab free T4 GPU now**\n\n[](https://colab.research.google.com/drive/1ocSmoy3ba1EkYu7JWT1oCw9vz8qC2cMk#scrollTo=zLORi5OcPcIJ)\n\n **It is highly recommended to use fp16 inference for this model, as 8-bit precision may significantly affect performance. If you require a more Consumer Hardware friendly version, please use the specialized quantized, only 5+GB V-Ram required** [JosephusCheung/GuanacoOnConsumerHardware](https://huggingface.co/JosephusCheung/GuanacoOnConsumerHardware).\n\n **You are encouraged to use the latest version of transformers from GitHub.**\n\nGuanaco is an advanced instruction-following language model built on Meta's LLaMA 7B model. Expanding upon the initial 52K dataset from the Alpaca model, an additional 534K+ entries have been incorporated, covering English, Simplified Chinese, Traditional Chinese (Taiwan), Traditional Chinese (Hong Kong), Japanese, Deutsch, and various linguistic and grammatical tasks. This wealth of data enables Guanaco to perform exceptionally well in multilingual environments.\n\nIn an effort to foster openness and replicability in research, we have made the Guanaco Dataset publicly accessible and we have released the model weights here. By providing these resources, we aim to inspire more researchers to pursue related research and collectively advance the development of instruction-following language models.\n\n[KBlueLeaf](https://huggingface.co/KBlueLeaf)’s invaluable contributions to the conceptual validation, [trained model](https://huggingface.co/KBlueLeaf/guanaco-7B-leh) and [inference development](https://github.com/KohakuBlueleaf/guanaco-lora) of the model would be gratefully acknowledged, without whose efforts the project shall never have come to fruition.\n\nWhen utilizing the Guanaco model, please bear in mind the following points:\n\nThe Guanaco model has not been filtered for harmful, biased, or explicit content. As a result, outputs that do not adhere to ethical norms may be generated during use. Please exercise caution when using the model in research or practical applications.\n1. ### Improved context and prompt role support:\n\n The new format is designed to be similar to ChatGPT, allowing for better integration with the Alpaca format and enhancing the overall user experience.\n\n Instruction is utilized as a few-shot context to support diverse inputs and responses, making it easier for the model to understand and provide accurate responses to user queries.\n\n The format is as follows:\n\n```\n### Instruction:\nUser: History User Input\nAssistant: History Assistant Answer\n### Input:\nSystem: Knowledge\nUser: New User Input\n### Response:\nNew Assistant Answer\n```\n\n This structured format allows for easier tracking of the conversation history and maintaining context throughout a multi-turn dialogue.\n\n3. ### Role-playing support:\n\n Guanaco now offers advanced role-playing support, similar to Character.AI, in English, Simplified Chinese, Traditional Chinese, Japanese, and Deutsch, making it more versatile for users from different linguistic backgrounds.\n\n Users can instruct the model to assume specific roles, historical figures, or fictional characters, as well as personalities based on their input. This allows for more engaging and immersive conversations.\n\n The model can use various sources of information to provide knowledge and context for the character's background and behavior, such as encyclopedic entries, first-person narrations, or a list of personality traits.\n\n The model will consistently output responses in the format \"Character Name: Reply\" to maintain the chosen role throughout the conversation, enhancing the user's experience.\n\n4. ### Rejection of answers and avoidance of erroneous responses:\n \n The model has been updated to handle situations where it lacks sufficient knowledge or is unable to provide a valid response more effectively.\n\n Reserved keywords have been introduced to indicate different scenarios and provide clearer communication with the user, use in System Prompt:\n\n NO IDEA: Indicates that the model lacks the necessary knowledge to provide an accurate answer, and will explain this to the user, encouraging them to seek alternative sources.\n \n FORBIDDEN: Indicates that the model refuses to answer due to specific reasons (e.g., legal, ethical, or safety concerns), which will be inferred based on the context of the query.\n\n SFW: Indicates that the model refuses to answer a question because it has been filtered for NSFW content, ensuring a safer and more appropriate user experience.\n\n6. ### Continuation of responses for ongoing topics:\n\n The Guanaco model can now continue answering questions or discussing topics upon the user's request, making it more adaptable and better suited for extended conversations.\n\n The contextual structure consisting of System, Assistant, and User roles allows the model to engage in multi-turn dialogues, maintain context-aware conversations, and provide more coherent responses.\n\n The model can now accommodate role specification and character settings, providing a more immersive and tailored conversational experience based on the user's preferences.\n\nIt is important to remember that Guanaco is a 7B-parameter model, and **any knowledge-based content should be considered potentially inaccurate**. We strongly recommend **providing verifiable sources in System Prompt, such as Wikipedia, for knowledge-based answers**. In the absence of sources, it is crucial to inform users of this limitation to prevent the dissemination of false information and to maintain transparency.\n\nDue to the differences in the format between this project and [Stanford Alpaca](https://github.com/tatsu-lab/stanford_alpaca), please refer to *Guanaco-lora: LoRA for training Multilingual Instruction-following LM based on LLaMA* (https://github.com/KohakuBlueleaf/guanaco-lora) for further training and inference our models.\n\n## Recent News\n\nWe've noticed a recent entrant in the field, the QLoRa method, which we find concerning due to its attempt to piggyback on the reputation of Guanaco. We strongly disapprove of such practices. QLoRa, as far as we can tell, lacks mathematical robustness and its performance significantly trails behind that of GPTQ and advancements such as PEFT fine-tuning, which have been successful in improving upon it.\n\nGuanaco has been diligent, consistently releasing multilingual datasets since March 2023, along with publishing weights that are not only an enhanced version of GPTQ but also support multimodal VQA and have been optimized for 4-bit. Despite the substantial financial investment of tens of thousands of dollars in distilling data from OpenAI's GPT models, we still consider these efforts to be incremental.\n\nWe, however, aim to move beyond the incremental:\n\n1. We strive to no longer rely on distillation data from OpenAI: We've found that relying on GPT-generated data impedes significant breakthroughs. Furthermore, this approach has proven to be disastrous when dealing with the imbalances in multilingual tasks.\n\n2. We're focusing on the enhancement of quantization structure and partial native 4-bit fine-tuning: We are deeply appreciative of the GPTQ-Llama project for paving the way in state-of-the-art LLM quantization. Its unique qualities, especially at the 7B size, are facilitating significant progress in multilingual and multimodal tasks.\n\n3. We plan to utilize visual data to adjust our language models: We believe this will fundamentally address the issues of language imbalance, translation inaccuracies, and the lack of graphical logic in LLM.\n\nWhile our work is still in the early stages, we're determined to break new ground in these areas. Our critique of QLoRa's practices does not stem from animosity but rather from the fundamental belief that innovation should be rooted in originality, integrity, and substantial progress.\n"},"matched_task":{"kind":"list like","value":["TRANSLATION"],"string":"[\n \"TRANSLATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["BEAR"],"string":"[\n \"BEAR\"\n]","isConversational":false}}},{"rowIdx":198,"cells":{"id":{"kind":"string","value":"mrm8488/modernbert-embed-base-ft-sts-spanish-matryoshka-768-64"},"author":{"kind":"string","value":"mrm8488"},"task_category":{"kind":"string","value":"sentence-similarity"},"tags":{"kind":"list like","value":["sentence-transformers","safetensors","modernbert","sentence-similarity","feature-extraction","generated_from_trainer","dataset_size:2697","loss:MatryoshkaLoss","loss:CoSENTLoss","arxiv:1908.10084","arxiv:2205.13147","base_model:nomic-ai/modernbert-embed-base","base_model:finetune:nomic-ai/modernbert-embed-base","model-index","autotrain_compatible","endpoints_compatible","region:us"],"string":"[\n \"sentence-transformers\",\n \"safetensors\",\n \"modernbert\",\n \"sentence-similarity\",\n \"feature-extraction\",\n \"generated_from_trainer\",\n \"dataset_size:2697\",\n \"loss:MatryoshkaLoss\",\n \"loss:CoSENTLoss\",\n \"arxiv:1908.10084\",\n \"arxiv:2205.13147\",\n \"base_model:nomic-ai/modernbert-embed-base\",\n \"base_model:finetune:nomic-ai/modernbert-embed-base\",\n \"model-index\",\n \"autotrain_compatible\",\n \"endpoints_compatible\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2025-01-10T21:44:41","string":"2025-01-10T21:44:41"},"last_modified":{"kind":"timestamp","value":"2025-01-10T21:50:00","string":"2025-01-10T21:50:00"},"downloads":{"kind":"number","value":2130,"string":"2,130"},"likes":{"kind":"number","value":2,"string":"2"},"README":{"kind":"string","value":"---\nbase_model: nomic-ai/modernbert-embed-base\nlibrary_name: sentence-transformers\nmetrics:\n- pearson_cosine\n- spearman_cosine\npipeline_tag: sentence-similarity\ntags:\n- sentence-transformers\n- sentence-similarity\n- feature-extraction\n- generated_from_trainer\n- dataset_size:2697\n- loss:MatryoshkaLoss\n- loss:CoSENTLoss\nwidget:\n- source_sentence: En un mercado de granjeros, se encuentra un hombre.\n sentences:\n - Un abogado de la CPI detenido en Libia está ahora mismo encarando un período de\n detención de 45 días\n - Un hombre está presente en un mercado donde se venden productos agrícolas directamente\n de los agricultores.\n - ¿Existe la posibilidad de que cambie de opinión si no se expresa de manera enérgica\n o muestra un comportamiento inapropiado?\n- source_sentence: Una mujer está posada en una postura con los brazos abiertos mientras\n otra persona le toma una fotografía.\n sentences:\n - Un hombre se encuentra parado en medio de una multitud sujetando un objeto de\n color blanco.\n - Las personas están cerca del agua.\n - Frente a una estatua de una vaca, hay una mujer, un niño pequeño y un bebé diminuto.\n- source_sentence: Un grupo de cuatro niños está observando los diferentes animales\n que están en el establo.\n sentences:\n - Evita apoyar todo tu peso en los brazos, ya que tus manos no están diseñadas para\n soportar esa presión constante.\n - Los niños están mirando atentamente a una oveja.\n - Un puma persigue a un oso grande en el bosque.\n- source_sentence: La gente se balancea saltando al agua mientras otros pescan en\n el fondo del mar.\n sentences:\n - Dos individuos observan el agua con atención.\n - Siempre golpeamos suavemente a nuestros hijos en la boca para mostrarles que su\n boca es lo que les causa dolor.\n - Aunque el sistema de prioridad al primero en llegar beneficia a dos participantes,\n no asegura definitivamente la exclusión de terceros.\n- source_sentence: El cordero está mirando hacia la cámara.\n sentences:\n - Manmohan en Teherán insta a NAM a tomar una posición clara sobre el conflicto\n en Siria\n - Un gato está mirando hacia la cámara también.\n - '\"Sí, no deseo estar presente durante este testimonio\", declaró tranquilamente\n Peterson, de 31 años, al juez cuando fue devuelto a su celda.'\nmodel-index:\n- name: SentenceTransformer based on nomic-ai/modernbert-embed-base\n results:\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts dev 768\n type: sts-dev-768\n metrics:\n - type: pearson_cosine\n value: 0.7498914121357008\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.7531670275662775\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts dev 512\n type: sts-dev-512\n metrics:\n - type: pearson_cosine\n value: 0.7468285624371191\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.7482342767593612\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts dev 256\n type: sts-dev-256\n metrics:\n - type: pearson_cosine\n value: 0.7419098803201045\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.7450577925521013\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts dev 128\n type: sts-dev-128\n metrics:\n - type: pearson_cosine\n value: 0.7262860099881795\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.7304432975238186\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts dev 64\n type: sts-dev-64\n metrics:\n - type: pearson_cosine\n value: 0.6973267849431932\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.7069603266334332\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts test 768\n type: sts-test-768\n metrics:\n - type: pearson_cosine\n value: 0.8673484326459211\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.8767387684433159\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts test 512\n type: sts-test-512\n metrics:\n - type: pearson_cosine\n value: 0.8665336885415594\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.8751868367625472\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts test 256\n type: sts-test-256\n metrics:\n - type: pearson_cosine\n value: 0.8568125590206718\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.8702353416571491\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts test 128\n type: sts-test-128\n metrics:\n - type: pearson_cosine\n value: 0.8485344363338887\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.8617402150766132\n name: Spearman Cosine\n - task:\n type: semantic-similarity\n name: Semantic Similarity\n dataset:\n name: sts test 64\n type: sts-test-64\n metrics:\n - type: pearson_cosine\n value: 0.8193790032247387\n name: Pearson Cosine\n - type: spearman_cosine\n value: 0.8419631939550043\n name: Spearman Cosine\n---\n\n# SentenceTransformer based on nomic-ai/modernbert-embed-base\n\nThis is a [sentence-transformers](https://www.SBERT.net) model finetuned from [nomic-ai/modernbert-embed-base](https://huggingface.co/nomic-ai/modernbert-embed-base) on the stsb_multi_es_augmented (private) dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.\n\n## Model Details\n\n### Model Description\n- **Model Type:** Sentence Transformer\n- **Base model:** [nomic-ai/modernbert-embed-base](https://huggingface.co/nomic-ai/modernbert-embed-base) <!-- at revision bb0033c9f3def40c3c5b26ff0b53c74f3320d703 -->\n- **Maximum Sequence Length:** 8192 tokens\n- **Output Dimensionality:** 768 dimensions\n- **Similarity Function:** Cosine Similarity\n- **Training Dataset:**\n - Private stsb dataset\n\n### Model Sources\n\n- **Documentation:** [Sentence Transformers Documentation](https://sbert.net)\n- **Repository:** [Sentence Transformers on GitHub](https://github.com/UKPLab/sentence-transformers)\n- **Hugging Face:** [Sentence Transformers on Hugging Face](https://huggingface.co/models?library=sentence-transformers)\n\n### Full Model Architecture\n\n```\nSentenceTransformer(\n (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False}) with Transformer model: ModernBertModel \n (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})\n (2): Normalize()\n)\n```\n\n## Usage\n\n### Direct Usage (Sentence Transformers)\n\nFirst install the Sentence Transformers library:\n\n```bash\npip install -U sentence-transformers\n```\n\nThen you can load this model and run inference.\n```python\nfrom sentence_transformers import SentenceTransformer\n\n# Download from the 🤗 Hub\nmodel = SentenceTransformer(\"mrm8488/modernbert-embed-base-ft-sts-spanish-matryoshka-768-64-5e\")\n# Run inference\nsentences = [\n 'El cordero está mirando hacia la cámara.',\n 'Un gato está mirando hacia la cámara también.',\n '\"Sí, no deseo estar presente durante este testimonio\", declaró tranquilamente Peterson, de 31 años, al juez cuando fue devuelto a su celda.',\n]\nembeddings = model.encode(sentences)\nprint(embeddings.shape)\n# [3, 768]\n\n# Get the similarity scores for the embeddings\nsimilarities = model.similarity(embeddings, embeddings)\nprint(similarities.shape)\n# [3, 3]\n```\n\n<!--\n### Direct Usage (Transformers)\n\n<details><summary>Click to see the direct usage in Transformers</summary>\n\n</details>\n-->\n\n<!--\n### Downstream Usage (Sentence Transformers)\n\nYou can finetune this model on your own dataset.\n\n<details><summary>Click to expand</summary>\n\n</details>\n-->\n\n<!--\n### Out-of-Scope Use\n\n*List how the model may foreseeably be misused and address what users ought not to do with the model.*\n-->\n\n## Evaluation\n\n### Metrics\n\n#### Semantic Similarity\n\n* Datasets: `sts-dev-768`, `sts-dev-512`, `sts-dev-256`, `sts-dev-128`, `sts-dev-64`, `sts-test-768`, `sts-test-512`, `sts-test-256`, `sts-test-128` and `sts-test-64`\n* Evaluated with [<code>EmbeddingSimilarityEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.evaluation.EmbeddingSimilarityEvaluator)\n\n| Metric | sts-dev-768 | sts-dev-512 | sts-dev-256 | sts-dev-128 | sts-dev-64 | sts-test-768 | sts-test-512 | sts-test-256 | sts-test-128 | sts-test-64 |\n|:--------------------|:------------|:------------|:------------|:------------|:-----------|:-------------|:-------------|:-------------|:-------------|:------------|\n| pearson_cosine | 0.7499 | 0.7468 | 0.7419 | 0.7263 | 0.6973 | 0.8673 | 0.8665 | 0.8568 | 0.8485 | 0.8194 |\n| **spearman_cosine** | **0.7532** | **0.7482** | **0.7451** | **0.7304** | **0.707** | **0.8767** | **0.8752** | **0.8702** | **0.8617** | **0.842** |\n\n<!--\n## Bias, Risks and Limitations\n\n*What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.*\n-->\n\n<!--\n### Recommendations\n\n*What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.*\n-->\n\n## Training Details\n\n### Training Dataset\n\n#### stsb_multi_es_augmented (private)\n\n* Size: 2,697 training samples\n* Columns: <code>sentence1</code>, <code>sentence2</code>, and <code>score</code>\n* Approximate statistics based on the first 1000 samples:\n | | sentence1 | sentence2 | score |\n |:--------|:----------------------------------------------------------------------------------|:-----------------------------------------------------------------------------------|:---------------------------------------------------------------|\n | type | string | string | float |\n | details | <ul><li>min: 9 tokens</li><li>mean: 28.42 tokens</li><li>max: 96 tokens</li></ul> | <ul><li>min: 10 tokens</li><li>mean: 28.01 tokens</li><li>max: 92 tokens</li></ul> | <ul><li>min: 0.0</li><li>mean: 2.72</li><li>max: 5.0</li></ul> |\n* Samples:\n | sentence1 | sentence2 | score |\n |:------------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------------------|:-------------------------------|\n | <code>El pájaro de tamaño reducido se posó con delicadeza en una rama cubierta de escarcha.</code> | <code>Un ave de color amarillo descansaba tranquilamente en una rama.</code> | <code>3.200000047683716</code> |\n | <code>Una chica está tocando la flauta en un parque.</code> | <code>Un grupo de músicos está tocando en un escenario al aire libre.</code> | <code>1.286</code> |\n | <code>La aclamada escritora británica, Doris Lessing, galardonada con el premio Nobel, fallece</code> | <code>La destacada autora británica, Doris Lessing, reconocida con el prestigioso Premio Nobel, muere</code> | <code>4.199999809265137</code> |\n* Loss: [<code>MatryoshkaLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#matryoshkaloss) with these parameters:\n ```json\n {\n \"loss\": \"CoSENTLoss\",\n \"matryoshka_dims\": [\n 768,\n 512,\n 256,\n 128,\n 64\n ],\n \"matryoshka_weights\": [\n 1,\n 1,\n 1,\n 1,\n 1\n ],\n \"n_dims_per_step\": -1\n }\n ```\n\n### Evaluation Dataset\n\n#### stsb_multi_es_augmented (private)\n\n* Size: 697 evaluation samples\n* Columns: <code>sentence1</code>, <code>sentence2</code>, and <code>score</code>\n* Approximate statistics based on the first 697 samples:\n | | sentence1 | sentence2 | score |\n |:--------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:--------------------------------------------------------------|\n | type | string | string | float |\n | details | <ul><li>min: 9 tokens</li><li>mean: 29.35 tokens</li><li>max: 87 tokens</li></ul> | <ul><li>min: 9 tokens</li><li>mean: 28.52 tokens</li><li>max: 81 tokens</li></ul> | <ul><li>min: 0.0</li><li>mean: 2.3</li><li>max: 5.0</li></ul> |\n* Samples:\n | sentence1 | sentence2 | score |\n |:--------------------------------------------------------------------------------------------------------------------------------------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------------------------|:-------------------------------|\n | <code>Un incendio ocurrido en un hospital psiquiátrico ruso resultó en la trágica muerte de 38 personas.</code> | <code>Se teme que el incendio en un hospital psiquiátrico ruso cause la pérdida de la vida de 38 individuos.</code> | <code>4.199999809265137</code> |\n | <code>\"Street dijo que el otro individuo a veces se siente avergonzado de su fiesta, lo cual provoca risas en la multitud\"</code> | <code>\"A veces, el otro tipo se encuentra avergonzado de su fiesta y no se le puede culpar.\"</code> | <code>3.5</code> |\n | <code>El veterano diplomático de Malasia tuvo un encuentro con Suu Kyi el miércoles en la casa del lago en Yangon donde permanece bajo arresto domiciliario.</code> | <code>Razali Ismail tuvo una reunión de 90 minutos con Suu Kyi, quien ganó el Premio Nobel de la Paz en 1991, en su casa del lago donde está recluida.</code> | <code>3.691999912261963</code> |\n* Loss: [<code>MatryoshkaLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#matryoshkaloss) with these parameters:\n ```json\n {\n \"loss\": \"CoSENTLoss\",\n \"matryoshka_dims\": [\n 768,\n 512,\n 256,\n 128,\n 64\n ],\n \"matryoshka_weights\": [\n 1,\n 1,\n 1,\n 1,\n 1\n ],\n \"n_dims_per_step\": -1\n }\n ```\n\n### Training Hyperparameters\n#### Non-Default Hyperparameters\n\n- `eval_strategy`: steps\n- `per_device_train_batch_size`: 16\n- `per_device_eval_batch_size`: 16\n- `num_train_epochs`: 5\n- `warmup_ratio`: 0.1\n- `bf16`: True\n\n#### All Hyperparameters\n<details><summary>Click to expand</summary>\n\n- `overwrite_output_dir`: False\n- `do_predict`: False\n- `eval_strategy`: steps\n- `prediction_loss_only`: True\n- `per_device_train_batch_size`: 16\n- `per_device_eval_batch_size`: 16\n- `per_gpu_train_batch_size`: None\n- `per_gpu_eval_batch_size`: None\n- `gradient_accumulation_steps`: 1\n- `eval_accumulation_steps`: None\n- `torch_empty_cache_steps`: None\n- `learning_rate`: 5e-05\n- `weight_decay`: 0.0\n- `adam_beta1`: 0.9\n- `adam_beta2`: 0.999\n- `adam_epsilon`: 1e-08\n- `max_grad_norm`: 1.0\n- `num_train_epochs`: 5\n- `max_steps`: -1\n- `lr_scheduler_type`: linear\n- `lr_scheduler_kwargs`: {}\n- `warmup_ratio`: 0.1\n- `warmup_steps`: 0\n- `log_level`: passive\n- `log_level_replica`: warning\n- `log_on_each_node`: True\n- `logging_nan_inf_filter`: True\n- `save_safetensors`: True\n- `save_on_each_node`: False\n- `save_only_model`: False\n- `restore_callback_states_from_checkpoint`: False\n- `no_cuda`: False\n- `use_cpu`: False\n- `use_mps_device`: False\n- `seed`: 42\n- `data_seed`: None\n- `jit_mode_eval`: False\n- `use_ipex`: False\n- `bf16`: True\n- `fp16`: False\n- `fp16_opt_level`: O1\n- `half_precision_backend`: auto\n- `bf16_full_eval`: False\n- `fp16_full_eval`: False\n- `tf32`: None\n- `local_rank`: 0\n- `ddp_backend`: None\n- `tpu_num_cores`: None\n- `tpu_metrics_debug`: False\n- `debug`: []\n- `dataloader_drop_last`: False\n- `dataloader_num_workers`: 0\n- `dataloader_prefetch_factor`: None\n- `past_index`: -1\n- `disable_tqdm`: False\n- `remove_unused_columns`: True\n- `label_names`: None\n- `load_best_model_at_end`: False\n- `ignore_data_skip`: False\n- `fsdp`: []\n- `fsdp_min_num_params`: 0\n- `fsdp_config`: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}\n- `fsdp_transformer_layer_cls_to_wrap`: None\n- `accelerator_config`: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}\n- `deepspeed`: None\n- `label_smoothing_factor`: 0.0\n- `optim`: adamw_torch\n- `optim_args`: None\n- `adafactor`: False\n- `group_by_length`: False\n- `length_column_name`: length\n- `ddp_find_unused_parameters`: None\n- `ddp_bucket_cap_mb`: None\n- `ddp_broadcast_buffers`: False\n- `dataloader_pin_memory`: True\n- `dataloader_persistent_workers`: False\n- `skip_memory_metrics`: True\n- `use_legacy_prediction_loop`: False\n- `push_to_hub`: False\n- `resume_from_checkpoint`: None\n- `hub_model_id`: None\n- `hub_strategy`: every_save\n- `hub_private_repo`: None\n- `hub_always_push`: False\n- `gradient_checkpointing`: False\n- `gradient_checkpointing_kwargs`: None\n- `include_inputs_for_metrics`: False\n- `include_for_metrics`: []\n- `eval_do_concat_batches`: True\n- `fp16_backend`: auto\n- `push_to_hub_model_id`: None\n- `push_to_hub_organization`: None\n- `mp_parameters`: \n- `auto_find_batch_size`: False\n- `full_determinism`: False\n- `torchdynamo`: None\n- `ray_scope`: last\n- `ddp_timeout`: 1800\n- `torch_compile`: False\n- `torch_compile_backend`: None\n- `torch_compile_mode`: None\n- `dispatch_batches`: None\n- `split_batches`: None\n- `include_tokens_per_second`: False\n- `include_num_input_tokens_seen`: False\n- `neftune_noise_alpha`: None\n- `optim_target_modules`: None\n- `batch_eval_metrics`: False\n- `eval_on_start`: False\n- `use_liger_kernel`: False\n- `eval_use_gather_object`: False\n- `average_tokens_across_devices`: False\n- `prompts`: None\n- `batch_sampler`: batch_sampler\n- `multi_dataset_batch_sampler`: proportional\n\n</details>\n\n### Training Logs\n| Epoch | Step | Training Loss | Validation Loss | sts-dev-768_spearman_cosine | sts-dev-512_spearman_cosine | sts-dev-256_spearman_cosine | sts-dev-128_spearman_cosine | sts-dev-64_spearman_cosine | sts-test-768_spearman_cosine | sts-test-512_spearman_cosine | sts-test-256_spearman_cosine | sts-test-128_spearman_cosine | sts-test-64_spearman_cosine |\n|:------:|:----:|:-------------:|:---------------:|:---------------------------:|:---------------------------:|:---------------------------:|:---------------------------:|:--------------------------:|:----------------------------:|:----------------------------:|:----------------------------:|:----------------------------:|:---------------------------:|\n| 0.5917 | 100 | 23.7709 | 22.5494 | 0.7185 | 0.7146 | 0.7055 | 0.6794 | 0.6570 | - | - | - | - | - |\n| 1.1834 | 200 | 22.137 | 22.7634 | 0.7449 | 0.7412 | 0.7439 | 0.7287 | 0.7027 | - | - | - | - | - |\n| 1.7751 | 300 | 21.5527 | 22.6985 | 0.7321 | 0.7281 | 0.7243 | 0.7063 | 0.6862 | - | - | - | - | - |\n| 2.3669 | 400 | 20.5745 | 24.0021 | 0.7302 | 0.7264 | 0.7221 | 0.7097 | 0.6897 | - | - | - | - | - |\n| 2.9586 | 500 | 20.0861 | 24.0091 | 0.7392 | 0.7361 | 0.7293 | 0.7124 | 0.6906 | - | - | - | - | - |\n| 3.5503 | 600 | 18.8191 | 26.9012 | 0.7502 | 0.7462 | 0.7399 | 0.7207 | 0.6960 | - | - | - | - | - |\n| 4.1420 | 700 | 18.3 | 29.0209 | 0.7496 | 0.7454 | 0.7432 | 0.7284 | 0.7065 | - | - | - | - | - |\n| 4.7337 | 800 | 17.6496 | 28.9536 | 0.7532 | 0.7482 | 0.7451 | 0.7304 | 0.7070 | - | - | - | - | - |\n| 5.0 | 845 | - | - | - | - | - | - | - | 0.8767 | 0.8752 | 0.8702 | 0.8617 | 0.8420 |\n\n\n### Framework Versions\n- Python: 3.10.12\n- Sentence Transformers: 3.3.1\n- Transformers: 4.48.0\n- PyTorch: 2.5.1+cu121\n- Accelerate: 1.2.1\n- Datasets: 3.2.0\n- Tokenizers: 0.21.0\n\n## Citation\n\n### BibTeX\n\n#### Sentence Transformers\n```bibtex\n@inproceedings{reimers-2019-sentence-bert,\n title = \"Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks\",\n author = \"Reimers, Nils and Gurevych, Iryna\",\n booktitle = \"Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing\",\n month = \"11\",\n year = \"2019\",\n publisher = \"Association for Computational Linguistics\",\n url = \"https://arxiv.org/abs/1908.10084\",\n}\n```\n\n#### MatryoshkaLoss\n```bibtex\n@misc{kusupati2024matryoshka,\n title={Matryoshka Representation Learning},\n author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},\n year={2024},\n eprint={2205.13147},\n archivePrefix={arXiv},\n primaryClass={cs.LG}\n}\n```\n\n#### CoSENTLoss\n```bibtex\n@online{kexuefm-8847,\n title={CoSENT: A more efficient sentence vector scheme than Sentence-BERT},\n author={Su Jianlin},\n year={2022},\n month={Jan},\n url={https://kexue.fm/archives/8847},\n}\n```\n\n<!--\n## Glossary\n\n*Clearly define terms in order to be accessible across audiences.*\n-->\n\n<!--\n## Model Card Authors\n\n*Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.*\n-->\n\n<!--\n## Model Card Contact\n\n*Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.*\n-->"},"matched_task":{"kind":"list like","value":["TEXT_CLASSIFICATION","SEMANTIC_SIMILARITY"],"string":"[\n \"TEXT_CLASSIFICATION\",\n \"SEMANTIC_SIMILARITY\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["CPI"],"string":"[\n \"CPI\"\n]","isConversational":false}}},{"rowIdx":199,"cells":{"id":{"kind":"string","value":"aisingapore/gemma2-9b-cpt-sea-lionv3-instruct"},"author":{"kind":"string","value":"aisingapore"},"task_category":{"kind":"string","value":"text-generation"},"tags":{"kind":"list like","value":["transformers","safetensors","gemma2","text-generation","conversational","en","zh","vi","id","th","fil","ta","ms","km","lo","my","jv","su","arxiv:2309.06085","arxiv:2311.07911","arxiv:2306.05685","base_model:aisingapore/gemma2-9b-cpt-sea-lionv3-base","base_model:finetune:aisingapore/gemma2-9b-cpt-sea-lionv3-base","license:gemma","autotrain_compatible","text-generation-inference","endpoints_compatible","region:us"],"string":"[\n \"transformers\",\n \"safetensors\",\n \"gemma2\",\n \"text-generation\",\n \"conversational\",\n \"en\",\n \"zh\",\n \"vi\",\n \"id\",\n \"th\",\n \"fil\",\n \"ta\",\n \"ms\",\n \"km\",\n \"lo\",\n \"my\",\n \"jv\",\n \"su\",\n \"arxiv:2309.06085\",\n \"arxiv:2311.07911\",\n \"arxiv:2306.05685\",\n \"base_model:aisingapore/gemma2-9b-cpt-sea-lionv3-base\",\n \"base_model:finetune:aisingapore/gemma2-9b-cpt-sea-lionv3-base\",\n \"license:gemma\",\n \"autotrain_compatible\",\n \"text-generation-inference\",\n \"endpoints_compatible\",\n \"region:us\"\n]","isConversational":false},"created_time":{"kind":"timestamp","value":"2024-10-30T03:19:20","string":"2024-10-30T03:19:20"},"last_modified":{"kind":"timestamp","value":"2024-12-19T12:57:01","string":"2024-12-19T12:57:01"},"downloads":{"kind":"number","value":2114,"string":"2,114"},"likes":{"kind":"number","value":10,"string":"10"},"README":{"kind":"string","value":"---\nbase_model:\n- aisingapore/gemma2-9b-cpt-sea-lionv3-base\nlanguage:\n- en\n- zh\n- vi\n- id\n- th\n- fil\n- ta\n- ms\n- km\n- lo\n- my\n- jv\n- su\nlibrary_name: transformers\nlicense: gemma\npipeline_tag: text-generation\n---\n\n<div>\n <img src=\"gemma_2_9b_sea-lion_v3_instruct_banner.png\"/>\n</div>\n\n# Gemma2 9B CPT SEA-LIONv3 Instruct\n\nSEA-LION is a collection of Large Language Models (LLMs) which have been pretrained and instruct-tuned for the Southeast Asia (SEA) region.\n\nGemma2 9B CPT SEA-LIONv3 Instruct is a multilingual model which has been fine-tuned with around **500,000 English instruction-completion pairs** alongside a larger pool of around **1,000,000 instruction-completion pairs** from other ASEAN languages, such as Indonesian, Thai and Vietnamese.\n\nSEA-LION stands for _Southeast Asian Languages In One Network_.\n\n- **Developed by:** Products Pillar, AI Singapore\n- **Funded by:** Singapore NRF\n- **Model type:** Decoder\n- **Languages supported:** Burmese, Chinese, English, Filipino, Indonesia, Javanese, Khmer, Lao, Malay, Sundanese, Tamil, Thai, Vietnamese\n- **License:** [Gemma Community License](https://ai.google.dev/gemma/terms)\n\n## Model Details\n### Model Description\nWe performed instruction tuning in English and also in ASEAN languages such as Indonesian, Thai and Vietnamese on our [continued pre-trained Gemma2 9B CPT SEA-LIONv3](https://huggingface.co/aisingapore/gemma2-9b-cpt-sea-lionv3-base), a decoder model using the Gemma2 architecture, to create Gemma2 9B CPT SEA-LIONv3 Instruct.\n\nFor tokenisation, the model employs the default tokenizer used in Gemma-2-9B. The model has a context length of 8192.\n\n### Benchmark Performance\nWe evaluated Gemma2 9B CPT SEA-LIONv3 Instruct on both general language capabilities and instruction-following capabilities.\n\n#### General Language Capabilities\nFor the evaluation of general language capabilities, we employed the [SEA HELM (also known as BHASA) evaluation benchmark](https://arxiv.org/abs/2309.06085v2) across a variety of tasks.\nThese tasks include Question Answering (QA), Sentiment Analysis (Sentiment), Toxicity Detection (Toxicity), Translation in both directions (Eng>Lang & Lang>Eng), Abstractive Summarization (Summ), Causal Reasoning (Causal) and Natural Language Inference (NLI).\n\nNote: SEA HELM is implemented using prompts to elicit answers in a strict format. For all tasks, the model is expected to provide an answer tag from which the answer is automatically extracted. For tasks where options are provided, the answer should comprise one of the pre-defined options. The scores for each task is normalised to account for baseline performance due to random chance.\n\nThe evaluation was done **zero-shot** with native prompts on a sample of 100-1000 instances for each dataset.\n\n#### Instruction-following Capabilities\nSince Gemma2 9B CPT SEA-LIONv3 Instruct is an instruction-following model, we also evaluated it on instruction-following capabilities with two datasets, [IFEval](https://arxiv.org/abs/2311.07911) and [MT-Bench](https://arxiv.org/abs/2306.05685).\n\nAs these two datasets were originally in English, the linguists and native speakers in the team worked together to filter, localize and translate the datasets into the respective target languages to ensure that the examples remained reasonable, meaningful and natural.\n\n**IFEval**\n\nIFEval evaluates a model's ability to adhere to constraints provided in the prompt, for example beginning a response with a specific word/phrase or answering with a certain number of sections. Additionally, accuracy is normalized by the proportion of responses in the correct language (if the model performs the task correctly but responds in the wrong language, it is judged to have failed the task).\n\n\n**MT-Bench**\n\nMT-Bench evaluates a model's ability to engage in multi-turn (2 turns) conversations and respond in ways that align with human needs. We use `gpt-4-1106-preview` as the judge model and compare against `gpt-3.5-turbo-0125` as the baseline model. The metric used is the weighted win rate against the baseline model (i.e. average win rate across each category: Math, Reasoning, STEM, Humanities, Roleplay, Writing, Extraction). A tie is given a score of 0.5.\n\n\nFor more details on Gemma2 9B CPT SEA-LIONv3 Instruct benchmark performance, please refer to the SEA HELM leaderboard, https://leaderboard.sea-lion.ai/\n\n\n### Usage\n**NOTE** This model has not been trained to use a system prompt or to use tool calling.\n\nGemma2 9B CPT SEA-LIONv3 Instruct can be run using the 🤗 Transformers library \n```python\n# Please use transformers==4.45.2\n\nimport transformers\nimport torch\n\nmodel_id = \"aisingapore/gemma2-9b-cpt-sea-lionv3-instruct\"\n\npipeline = transformers.pipeline(\n \"text-generation\",\n model=model_id,\n model_kwargs={\"torch_dtype\": torch.bfloat16},\n device_map=\"auto\",\n)\nmessages = [\n {\"role\": \"user\", \"content\": \"Apa sentimen dari kalimat berikut ini?\\nKalimat: Buku ini sangat membosankan.\\nJawaban: \"},\n]\n\noutputs = pipeline(\n messages,\n max_new_tokens=256,\n)\nprint(outputs[0][\"generated_text\"][-1])\n```\n\n### Caveats\nIt is important for users to be aware that our model exhibits certain limitations that warrant consideration. Like many LLMs, the model can hallucinate and occasionally generates irrelevant content, introducing fictional elements that are not grounded in the provided context. Users should also exercise caution in interpreting and validating the model's responses due to the potential inconsistencies in its reasoning.\n\n## Limitations\n### Safety\nCurrent SEA-LION models, including this commercially permissive release, have not been aligned for safety. Developers and users should perform their own safety fine-tuning and related security measures. In no event shall the authors be held liable for any claim, damages, or other liability arising from the use of the released weights and codes.\n\n## Technical Specifications\n### Fine-Tuning Details\nGemma2 9B CPT SEA-LIONv3 Instruct was built using a combination of a full parameter fine-tune, on-policy alignment, and model merges of the best performing checkpoints. The training process for fine-tuning was approximately 15 hours, with alignment taking 2 hours, both on 8x H100-80GB GPUs.\n\n## Data\nGemma2 9B CPT SEA-LIONv3 Instruct was trained on a wide range of synthetic instructions, alongside publicly available instructions hand-curated by the team with the assistance of native speakers. In addition, special care was taken to ensure that the datasets used had commercially permissive licenses through verification with the original data source. \n\n## Indonesian, Javanese & Sudanese Specific SEA-LION\nOur partners at GoTo have continued pretrained and instruction tuned a variant of Gemma2 9B CPT SEA-LIONv3, specifically enhancing its capabilities for Indonesian, Javanese, and Sundanese languages. Find the continued pretrained model at [Gemma2 9B CPT SahabatAIv1 Base](https://huggingface.co/GoToCompany/gemma2-9b-cpt-sahabatai-v1-base), and its corresponding instructioned tuned version at [Gemma2 9B CPT SahabatAIv1 Instruct](https://huggingface.co/GoToCompany/gemma2-9b-cpt-sahabatai-v1-instruct). \n\n## Call for Contributions\nWe encourage researchers, developers, and language enthusiasts to actively contribute to the enhancement and expansion of SEA-LION. Contributions can involve identifying and reporting bugs, sharing pre-training, instruction, and preference data, improving documentation usability, proposing and implementing new model evaluation tasks and metrics, or training versions of the model in additional Southeast Asian languages. Join us in shaping the future of SEA-LION by sharing your expertise and insights to make these models more accessible, accurate, and versatile. Please check out our GitHub for further information on the call for contributions.\n\n## The Team\nChan Adwin, Cheng Nicholas, Choa Esther, Huang Yuli, Hulagadri Adithya Venkatadri, Lau Wayne, Lee Chwan Ren, Leong Wai Yi, Leong Wei Qi, Limkonchotiwat Peerat, Liu Bing Jie Darius, Montalan Jann Railey, Ng Boon Cheong Raymond, Ngui Jian Gang, Nguyen Thanh Ngan, Ong Brandon, Ong Tat-Wee David, Ong Zhi Hao, Rengarajan Hamsawardhini, Siow Bryan, Susanto Yosephine, Tai Ngee Chia, Tan Choon Meng, Teng Walter, Teo Eng Sipp Leslie, Teo Wei Yi, Tjhi William, Yeo Yeow Tong, Yong Xianbin\n\n## Acknowledgements\n[AI Singapore](https://aisingapore.org/) is a national programme supported by the National Research Foundation, Singapore and hosted by the National University of Singapore. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not reflect the views of the National Research Foundation or the National University of Singapore. \n\n## Contact\nFor more info, please contact us using this [SEA-LION Inquiry Form](https://forms.gle/sLCUVb95wmGf43hi6)\n\n[Link to SEA-LION's GitHub repository](https://github.com/aisingapore/sealion)\n\n## Disclaimer\nThis is the repository for the commercial instruction-tuned model.\nThe model has _not_ been aligned for safety.\nDevelopers and users should perform their own safety fine-tuning and related security measures.\nIn no event shall the authors be held liable for any claims, damages, or other liabilities arising from the use of the released weights and codes."},"matched_task":{"kind":"list like","value":["QUESTION_ANSWERING","TRANSLATION","SUMMARIZATION"],"string":"[\n \"QUESTION_ANSWERING\",\n \"TRANSLATION\",\n \"SUMMARIZATION\"\n]","isConversational":false},"matched_bigbio_names":{"kind":"list like","value":["CHIA"],"string":"[\n \"CHIA\"\n]","isConversational":false}}}],"truncated":false,"partial":false},"paginationData":{"pageIndex":1,"numItemsPerPage":100,"numTotalItems":1838,"offset":100,"length":100}},"jwt":"eyJhbGciOiJFZERTQSJ9.eyJyZWFkIjp0cnVlLCJwZXJtaXNzaW9ucyI6eyJyZXBvLmNvbnRlbnQucmVhZCI6dHJ1ZX0sImlhdCI6MTc3NzIxNTU4NSwic3ViIjoiL2RhdGFzZXRzL0V1YW55dS9tZXJnZWRfdGFza19kYXRhc2V0X21vZGVscyIsImV4cCI6MTc3NzIxOTE4NSwiaXNzIjoiaHR0cHM6Ly9odWdnaW5nZmFjZS5jbyJ9.AY1Orn8RtuZS4C67p7SZ8BYHiZHZ8Bm-70pa_E6WJPgjzTGr3Rndy_7NJT6y_FJpO_5wrBn1fdMlaiP3RX5vDQ","displayUrls":true,"splitSizeSummaries":[{"config":"default","split":"train","numRows":1838,"numBytesParquet":47475604}]},"discussionsStats":{"closed":0,"open":0,"total":0},"fullWidth":true,"hasGatedAccess":true,"hasFullAccess":true,"isEmbedded":false,"savedQueries":{"community":[],"user":[]}}">
id stringlengths 11 95 | author stringlengths 3 36 | task_category stringclasses 16
values | tags listlengths 1 4.05k | created_time timestamp[s]date 2022-03-02 23:29:04 2025-03-18 02:34:30 | last_modified timestamp[s]date 2021-05-13 19:09:22 2025-03-18 03:19:02 | downloads int64 0 15.6M | likes int64 0 4.86k | README stringlengths 246 1.01M | matched_task listlengths 1 8 | matched_bigbio_names listlengths 1 8 |
|---|---|---|---|---|---|---|---|---|---|---|
ibm-granite/granite-embedding-278m-multilingual | ibm-granite | sentence-similarity | [
"transformers",
"pytorch",
"safetensors",
"xlm-roberta",
"feature-extraction",
"language",
"granite",
"embeddings",
"multilingual",
"mteb",
"sentence-transformers",
"sentence-similarity",
"en",
"ar",
"cs",
"de",
"es",
"fr",
"it",
"ja",
"ko",
"nl",
"pt",
"zh",
"arxiv:0... | 2024-12-04T20:29:12 | 2025-03-04T15:05:37 | 21,581 | 34 | ---
language:
- en
- ar
- cs
- de
- es
- fr
- it
- ja
- ko
- nl
- pt
- zh
library_name: transformers
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- language
- granite
- embeddings
- multilingual
- mteb
- sentence-transformers
model-index:
- name: ibm-granite/granite-embedding-278m-multilingual
results:... | [
"TRANSLATION",
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
BAAI/bge-large-zh | BAAI | feature-extraction | [
"transformers",
"pytorch",
"safetensors",
"bert",
"feature-extraction",
"zh",
"arxiv:2310.07554",
"arxiv:2309.07597",
"license:mit",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2023-08-02T07:13:44 | 2023-10-12T03:38:28 | 21,308 | 320 | ---
language:
- zh
license: mit
---
**Recommend switching to newest [BAAI/bge-large-zh-v1.5](https://huggingface.co/BAAI/bge-large-zh-v1.5), which has more reasonable similarity distribution and same method of usage.**
<h1 align="center">FlagEmbedding</h1>
<h4 align="center">
<p>
<a href=#model-list>Mo... | [
"SEMANTIC_SIMILARITY",
"SUMMARIZATION"
] | [
"BEAR"
] |
EleutherAI/pythia-1b-deduped | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2304.01373",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endp... | 2023-02-14T00:07:42 | 2023-07-10T15:04:31 | 20,123 | 19 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research [(see paper)](https://arxiv.org/pdf/2304.01373.pdf).
It contains two sets of eight models... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"SCIQ"
] |
EleutherAI/gpt-neox-20b | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"en",
"dataset:EleutherAI/pile",
"arxiv:2204.06745",
"arxiv:2101.00027",
"arxiv:2201.07311",
"arxiv:2104.09864",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_c... | 2022-04-07T20:28:29 | 2024-01-31T20:30:35 | 19,595 | 555 | ---
datasets:
- EleutherAI/pile
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
---
GPT-NeoX-20B is a 20 billion parameter autoregressive language model trained
on [the Pile](https://pile.eleuther.ai/) using the [GPT-NeoX
library](https://github.com/EleutherAI/gpt-neox). Its architecture intentionally... | [
"TRANSLATION"
] | [
"SCIQ"
] |
Linq-AI-Research/Linq-Embed-Mistral | Linq-AI-Research | feature-extraction | [
"sentence-transformers",
"safetensors",
"mistral",
"feature-extraction",
"mteb",
"transformers",
"en",
"arxiv:2210.07316",
"arxiv:2310.06825",
"arxiv:2401.00368",
"arxiv:2104.08663",
"license:cc-by-nc-4.0",
"model-index",
"autotrain_compatible",
"text-generation-inference",
"text-embed... | 2024-05-29T03:41:40 | 2024-06-05T12:50:34 | 19,366 | 69 | ---
language:
- en
license: cc-by-nc-4.0
tags:
- mteb
- transformers
- sentence-transformers
model-index:
- name: Linq-Embed-Mistral
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: ... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
bigscience/bloom-3b | bigscience | text-generation | [
"transformers",
"pytorch",
"safetensors",
"bloom",
"text-generation",
"ak",
"ar",
"as",
"bm",
"bn",
"ca",
"code",
"en",
"es",
"eu",
"fon",
"fr",
"gu",
"hi",
"id",
"ig",
"ki",
"kn",
"lg",
"ln",
"ml",
"mr",
"ne",
"nso",
"ny",
"or",
"pa",
"pt",
"rn",
... | 2022-05-19T11:52:27 | 2023-04-14T08:43:29 | 17,019 | 90 | ---
language:
- ak
- ar
- as
- bm
- bn
- ca
- code
- en
- es
- eu
- fon
- fr
- gu
- hi
- id
- ig
- ki
- kn
- lg
- ln
- ml
- mr
- ne
- nso
- ny
- or
- pa
- pt
- rn
- rw
- sn
- st
- sw
- ta
- te
- tn
- ts
- tum
- tw
- ur
- vi
- wo
- xh
- yo
- zh
- zhs
- zht
- zu
license: bigscience-bloom-rail-1.0
pipeline_tag: text-gener... | [
"QUESTION_ANSWERING",
"SUMMARIZATION"
] | [
"PUBMEDQA",
"SCIQ"
] |
Lihuchen/pearl_base | Lihuchen | feature-extraction | [
"sentence-transformers",
"pytorch",
"safetensors",
"bert",
"feature-extraction",
"Phrase Representation",
"String Matching",
"Fuzzy Join",
"Entity Retrieval",
"transformers",
"en",
"arxiv:2401.10407",
"license:apache-2.0",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints... | 2024-02-16T22:00:37 | 2024-03-06T16:14:00 | 16,855 | 3 | ---
language:
- en
license: apache-2.0
tags:
- Phrase Representation
- String Matching
- Fuzzy Join
- Entity Retrieval
- transformers
- sentence-transformers
---
## PEARL-base
[Learning High-Quality and General-Purpose Phrase Representations](https://arxiv.org/pdf/2401.10407.pdf). <br>
[Lihu Chen](https://chenlihu.com... | [
"SEMANTIC_SIMILARITY"
] | [
"BC5CDR"
] |
sileod/deberta-v3-base-tasksource-nli | sileod | zero-shot-classification | [
"transformers",
"pytorch",
"safetensors",
"deberta-v2",
"text-classification",
"deberta-v3-base",
"deberta-v3",
"deberta",
"nli",
"natural-language-inference",
"multitask",
"multi-task",
"pipeline",
"extreme-multi-task",
"extreme-mtl",
"tasksource",
"zero-shot",
"rlhf",
"zero-sho... | 2023-01-13T13:47:22 | 2024-08-13T21:12:51 | 15,941 | 123 | ---
datasets:
- glue
- nyu-mll/multi_nli
- multi_nli
- super_glue
- anli
- tasksource/babi_nli
- sick
- snli
- scitail
- OpenAssistant/oasst1
- universal_dependencies
- hans
- qbao775/PARARULE-Plus
- alisawuffles/WANLI
- metaeval/recast
- sileod/probability_words_nli
- joey234/nan-nli
- pietrolesci/nli_fever
- pietrole... | [
"TEXT_CLASSIFICATION"
] | [
"HEAD-QA",
"JNLPBA",
"MEDQA",
"NCBI DISEASE",
"SCICITE",
"SCIQ",
"SCITAIL"
] |
sdadas/mmlw-e5-small | sdadas | sentence-similarity | [
"sentence-transformers",
"pytorch",
"safetensors",
"bert",
"feature-extraction",
"sentence-similarity",
"transformers",
"mteb",
"pl",
"arxiv:2402.13350",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2023-11-17T18:40:08 | 2024-10-30T14:16:51 | 15,170 | 0 | ---
language: pl
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
- mteb
widget:
- source_sentence: 'query: Jak dożyć 100 lat?'
sentences:
- 'passage: Trzeba zdrowo się odżywiać i uprawiać sport.'
- 'passage: Trzeba pić a... | [
"SEMANTIC_SIMILARITY"
] | [
"SCIFACT"
] |
EleutherAI/pythia-1.4b-deduped-v0 | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"pythia_v0",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_c... | 2022-10-18T03:03:34 | 2023-07-09T16:02:25 | 15,140 | 5 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
- pythia_v0
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research. It contains two sets of eight models of sizes
70M, 160M, 410M, 1B, 1.4B, 2.8... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"SCIQ"
] |
Snowflake/snowflake-arctic-embed-m-long | Snowflake | sentence-similarity | [
"sentence-transformers",
"onnx",
"safetensors",
"nomic_bert",
"feature-extraction",
"sentence-similarity",
"mteb",
"arctic",
"snowflake-arctic-embed",
"transformers.js",
"custom_code",
"arxiv:2407.18887",
"arxiv:2405.05374",
"arxiv:2104.09864",
"license:apache-2.0",
"model-index",
"a... | 2024-04-12T13:52:31 | 2024-12-13T20:53:23 | 15,016 | 36 | ---
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- mteb
- arctic
- snowflake-arctic-embed
- transformers.js
model-index:
- name: snowflake-arctic-m-long
results:
- task:
type: Classification
dataset:
name: MTEB AmazonC... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
EleutherAI/pythia-12b | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"en",
"dataset:EleutherAI/pile",
"arxiv:2304.01373",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible"... | 2023-02-28T18:48:12 | 2024-07-09T15:50:54 | 14,944 | 135 | ---
datasets:
- EleutherAI/pile
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research [(see paper)](https://arxiv.org/pdf/2304.01373.pdf).
It contains two sets of eight models of sizes
70M, 1... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"SCIQ"
] |
intfloat/e5-large | intfloat | sentence-similarity | [
"sentence-transformers",
"pytorch",
"safetensors",
"bert",
"mteb",
"Sentence Transformers",
"sentence-similarity",
"en",
"arxiv:2212.03533",
"arxiv:2104.08663",
"arxiv:2210.07316",
"license:mit",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible"... | 2022-12-26T06:03:12 | 2023-08-07T04:59:49 | 14,071 | 74 | ---
language:
- en
license: mit
tags:
- mteb
- Sentence Transformers
- sentence-similarity
- sentence-transformers
model-index:
- name: e5-large
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
... | [
"SEMANTIC_SIMILARITY",
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
stanford-crfm/BioMedLM | stanford-crfm | text-generation | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"dataset:pubmed",
"arxiv:2403.18421",
"license:bigscience-bloom-rail-1.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | 2022-12-14T08:14:59 | 2024-03-28T13:57:14 | 13,701 | 417 | ---
datasets:
- pubmed
license: bigscience-bloom-rail-1.0
widget:
- text: Photosynthesis is
---
# Model Card for BioMedLM 2.7B
Note: This model was previously known as PubMedGPT 2.7B, but we have changed it due to a request from the NIH which holds the trademark for "PubMed".
Paper: [BioMedLM: A 2.7B Parameter Langu... | [
"QUESTION_ANSWERING"
] | [
"MEDQA"
] |
ibm-granite/granite-embedding-125m-english | ibm-granite | sentence-similarity | [
"sentence-transformers",
"pytorch",
"safetensors",
"roberta",
"feature-extraction",
"language",
"granite",
"embeddings",
"mteb",
"transformers",
"sentence-similarity",
"en",
"arxiv:0000.00000",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
... | 2024-12-04T20:28:08 | 2025-03-04T15:09:35 | 13,457 | 9 | ---
language:
- en
library_name: sentence-transformers
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- language
- granite
- embeddings
- mteb
- transformers
model-index:
- name: ibm-granite/granite-embedding-125m-english
results:
- task:
type: Classification
dataset:
name: MTEB AmazonC... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
BSC-LT/salamandra-7b-instruct | BSC-LT | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"conversational",
"bg",
"ca",
"code",
"cs",
"cy",
"da",
"de",
"el",
"en",
"es",
"et",
"eu",
"fi",
"fr",
"ga",
"gl",
"hr",
"hu",
"it",
"lt",
"lv",
"mt",
"nl",
"nn",
"oc",
"pl",
"pt",
"ro",
"ru... | 2024-09-30T11:13:21 | 2025-02-20T16:48:03 | 12,988 | 54 | ---
base_model:
- BSC-LT/salamandra-7b
datasets:
- oscar-corpus/colossal-oscar-1.0
- HuggingFaceFW/fineweb-edu
- joelniklaus/eurlex_resources
- joelito/legal-mc4
- projecte-aina/CATalog
- UFRGS/brwac
- community-datasets/hrwac
- danish-foundation-models/danish-gigaword
- HiTZ/euscrawl
- PleIAs/French-PD-Newspapers
- Pl... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION",
"PARAPHRASING"
] | [
"BEAR",
"SCIELO"
] |
EleutherAI/pythia-1.4b-deduped | EleutherAI | text-generation | [
"transformers",
"pytorch",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2304.01373",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible"... | 2023-02-09T21:42:04 | 2023-06-08T13:03:28 | 12,669 | 19 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research [(see paper)](https://arxiv.org/pdf/2304.01373.pdf).
It contains two sets of eight models... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"SCIQ"
] |
intfloat/e5-small | intfloat | sentence-similarity | [
"sentence-transformers",
"pytorch",
"onnx",
"safetensors",
"bert",
"mteb",
"Sentence Transformers",
"sentence-similarity",
"en",
"arxiv:2212.03533",
"arxiv:2104.08663",
"arxiv:2210.07316",
"license:mit",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_c... | 2022-12-07T06:48:03 | 2023-08-07T04:58:08 | 11,580 | 41 | ---
language:
- en
license: mit
tags:
- mteb
- Sentence Transformers
- sentence-similarity
- sentence-transformers
model-index:
- name: e5-small
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
... | [
"SEMANTIC_SIMILARITY",
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
hkunlp/instructor-base | hkunlp | sentence-similarity | [
"sentence-transformers",
"pytorch",
"t5",
"text-embedding",
"embeddings",
"information-retrieval",
"beir",
"text-classification",
"language-model",
"text-clustering",
"text-semantic-similarity",
"text-evaluation",
"prompt-retrieval",
"text-reranking",
"feature-extraction",
"sentence-si... | 2022-12-20T05:59:40 | 2023-01-21T06:31:16 | 11,563 | 116 | ---
language: en
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- text-embedding
- embeddings
- information-retrieval
- beir
- text-classification
- language-model
- text-clustering
- text-semantic-similarity
- text-evaluation
- prompt-retrieval
- text-reranking
- sentence-transformers
- feature-extraction... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
BAAI/bge-base-zh | BAAI | feature-extraction | [
"transformers",
"pytorch",
"safetensors",
"bert",
"feature-extraction",
"zh",
"arxiv:2310.07554",
"arxiv:2309.07597",
"license:mit",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2023-08-05T08:02:30 | 2023-10-12T03:37:45 | 11,361 | 52 | ---
language:
- zh
license: mit
---
**Recommend switching to newest [BAAI/bge-base-zh-v1.5](https://huggingface.co/BAAI/bge-base-zh-v1.5), which has more reasonable similarity distribution and same method of usage.**
<h1 align="center">FlagEmbedding</h1>
<h4 align="center">
<p>
<a href=#model-list>Mode... | [
"SEMANTIC_SIMILARITY",
"SUMMARIZATION"
] | [
"BEAR"
] |
GritLM/GritLM-7B | GritLM | text-generation | [
"transformers",
"pytorch",
"safetensors",
"mistral",
"text-generation",
"mteb",
"conversational",
"custom_code",
"dataset:GritLM/tulu2",
"arxiv:2402.09906",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | 2024-02-11T15:55:35 | 2024-02-16T10:14:51 | 11,128 | 97 | ---
datasets:
- GritLM/tulu2
license: apache-2.0
pipeline_tag: text-generation
tags:
- mteb
inference: true
model-index:
- name: GritLM-7B
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
s... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
m42-health/Llama3-Med42-70B | m42-health | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"m42",
"health",
"healthcare",
"clinical-llm",
"conversational",
"en",
"arxiv:2408.06142",
"license:llama3",
"autotrain_compatible",
"text-generation-inference",
"region:us"
] | 2024-06-27T13:19:46 | 2024-08-20T05:11:30 | 10,777 | 44 | ---
language:
- en
license: llama3
license_name: llama3
pipeline_tag: text-generation
tags:
- m42
- health
- healthcare
- clinical-llm
inference: false
---
# **Med42-v2 - A Suite of Clinically-aligned Large Language Models**
Med42-v2 is a suite of open-access clinical large language models (LLM) instruct and preference... | [
"QUESTION_ANSWERING",
"SUMMARIZATION"
] | [
"MEDQA"
] |
McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp-supervised | McGill-NLP | sentence-similarity | [
"peft",
"safetensors",
"text-embedding",
"embeddings",
"information-retrieval",
"beir",
"text-classification",
"language-model",
"text-clustering",
"text-semantic-similarity",
"text-evaluation",
"text-reranking",
"feature-extraction",
"sentence-similarity",
"Sentence Similarity",
"natu... | 2024-04-30T02:35:26 | 2024-04-30T03:48:00 | 10,725 | 48 | ---
language:
- en
library_name: peft
license: mit
pipeline_tag: sentence-similarity
tags:
- text-embedding
- embeddings
- information-retrieval
- beir
- text-classification
- language-model
- text-clustering
- text-semantic-similarity
- text-evaluation
- text-reranking
- feature-extraction
- sentence-similarity
- Sent... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
NovaSearch/jasper_en_vision_language_v1 | NovaSearch | null | [
"sentence-transformers",
"safetensors",
"jasper_vl",
"mteb",
"custom_code",
"en",
"dataset:BAAI/Infinity-MM",
"dataset:HuggingFaceFW/fineweb-edu",
"arxiv:2412.19048",
"base_model:NovaSearch/stella_en_1.5B_v5",
"base_model:finetune:NovaSearch/stella_en_1.5B_v5",
"model-index",
"region:us"
] | 2024-12-11T03:23:03 | 2025-01-24T02:03:17 | 10,537 | 46 | ---
base_model:
- dunzhang/stella_en_1.5B_v5
- google/siglip-so400m-patch14-384
datasets:
- BAAI/Infinity-MM
- HuggingFaceFW/fineweb-edu
language:
- en
tags:
- mteb
- sentence-transformers
model-index:
- name: jasper_en_vision_language_v1
results:
- task:
type: Classification
dataset:
name: MTEB Ama... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
RichardErkhov/Alibaba-NLP_-_gte-Qwen2-7B-instruct-gguf | RichardErkhov | null | [
"gguf",
"arxiv:2308.03281",
"endpoints_compatible",
"region:us",
"conversational"
] | 2024-06-22T13:20:44 | 2024-06-22T15:22:37 | 10,421 | 7 | ---
{}
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
gte-Qwen2-7B-instruct - GGUF
- Model creator: https://huggingface.co/Alibaba-NLP/
- Original model: https://huggi... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
EleutherAI/pythia-6.9b-deduped | EleutherAI | text-generation | [
"transformers",
"pytorch",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2304.01373",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible"... | 2023-02-25T17:56:57 | 2023-06-08T13:05:19 | 10,266 | 8 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research [(see paper)](https://arxiv.org/pdf/2304.01373.pdf).
It contains two sets of eight models... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"SCIQ"
] |
EleutherAI/pythia-2.8b-deduped | EleutherAI | text-generation | [
"transformers",
"pytorch",
"safetensors",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2304.01373",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endp... | 2023-02-10T22:26:20 | 2023-07-09T16:06:37 | 9,120 | 14 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research [(see paper)](https://arxiv.org/pdf/2304.01373.pdf).
It contains two sets of eight models... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"SCIQ"
] |
BAAI/bge-small-zh | BAAI | feature-extraction | [
"transformers",
"pytorch",
"bert",
"feature-extraction",
"zh",
"arxiv:2310.07554",
"arxiv:2309.07597",
"license:mit",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2023-08-05T08:03:22 | 2023-10-12T03:37:29 | 9,020 | 18 | ---
language:
- zh
license: mit
---
**Recommend switching to newest [BAAI/bge-small-zh-v1.5](https://huggingface.co/BAAI/bge-small-zh-v1.5), which has more reasonable similarity distribution and same method of usage.**
<h1 align="center">FlagEmbedding</h1>
<h4 align="center">
<p>
<a href=#model-list>Mo... | [
"SEMANTIC_SIMILARITY",
"SUMMARIZATION"
] | [
"BEAR"
] |
unicamp-dl/translation-en-pt-t5 | unicamp-dl | translation | [
"transformers",
"pytorch",
"t5",
"text2text-generation",
"translation",
"en",
"pt",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | 2022-03-02T23:29:05 | 2021-10-11T03:47:21 | 9,018 | 20 | ---
datasets:
- EMEA
- ParaCrawl 99k
- CAPES
- Scielo
- JRC-Acquis
- Biomedical Domain Corpora
language:
- en
- pt
metrics:
- bleu
tags:
- translation
---
# Introduction
This repository brings an implementation of T5 for translation in EN-PT tasks using a modest hardware setup. We propose some changes in tokenizator... | [
"TRANSLATION"
] | [
"SCIELO"
] |
HIT-TMG/KaLM-embedding-multilingual-mini-v1 | HIT-TMG | sentence-similarity | [
"sentence-transformers",
"safetensors",
"qwen2",
"feature-extraction",
"sentence-similarity",
"mteb",
"arxiv:2501.01028",
"license:mit",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2024-08-27T08:56:33 | 2025-01-03T07:26:50 | 8,131 | 20 | ---
license: mit
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- mteb
model-index:
- name: KaLM-Embedding
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en-ext)
type: mteb/amazon_counter... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
sdadas/mmlw-roberta-large | sdadas | sentence-similarity | [
"sentence-transformers",
"pytorch",
"safetensors",
"roberta",
"feature-extraction",
"sentence-similarity",
"transformers",
"mteb",
"pl",
"arxiv:2402.13350",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2023-11-17T19:08:47 | 2024-10-25T04:30:01 | 7,993 | 13 | ---
language: pl
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
- mteb
widget:
- source_sentence: 'zapytanie: Jak dożyć 100 lat?'
sentences:
- Trzeba zdrowo się odżywiać i uprawiać sport.
- Trzeba pić alkohol, imprezowa... | [
"SEMANTIC_SIMILARITY"
] | [
"SCIFACT"
] |
michiyasunaga/BioLinkBERT-large | michiyasunaga | text-classification | [
"transformers",
"pytorch",
"bert",
"feature-extraction",
"exbert",
"linkbert",
"biolinkbert",
"fill-mask",
"question-answering",
"text-classification",
"token-classification",
"en",
"dataset:pubmed",
"arxiv:2203.15827",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | 2022-03-08T06:20:38 | 2022-03-31T00:54:57 | 7,776 | 33 | ---
datasets:
- pubmed
language: en
license: apache-2.0
tags:
- bert
- exbert
- linkbert
- biolinkbert
- feature-extraction
- fill-mask
- question-answering
- text-classification
- token-classification
widget:
- text: Sunitinib is a tyrosine kinase inhibitor
---
## BioLinkBERT-large
BioLinkBERT-l... | [
"TEXT_CLASSIFICATION",
"QUESTION_ANSWERING"
] | [
"BLURB",
"MEDQA",
"PUBMEDQA"
] |
ai-sage/Giga-Embeddings-instruct | ai-sage | feature-extraction | [
"safetensors",
"gigarembed",
"mteb",
"feature-extraction",
"custom_code",
"ru",
"en",
"license:mit",
"model-index",
"region:us"
] | 2024-12-11T12:25:30 | 2025-02-04T09:21:03 | 7,701 | 43 | ---
language:
- ru
- en
license: mit
pipeline_tag: feature-extraction
tags:
- mteb
model-index:
- name: giga-embeddings-instruct
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en-ext)
type: mteb/amazon_counterfactual
config: en-ext
spl... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
Lajavaness/bilingual-embedding-small | Lajavaness | sentence-similarity | [
"sentence-transformers",
"safetensors",
"bilingual",
"feature-extraction",
"sentence-similarity",
"transformers",
"sentence-embedding",
"mteb",
"custom_code",
"fr",
"en",
"arxiv:2010.08240",
"arxiv:1911.02116",
"arxiv:1908.10084",
"license:apache-2.0",
"model-index",
"autotrain_compa... | 2024-07-17T03:13:47 | 2024-11-20T14:45:37 | 6,982 | 4 | ---
language:
- fr
- en
library_name: sentence-transformers
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
- sentence-embedding
- mteb
model-index:
- name: bilingual-embedding-small
results:
- task:
type: Clustering... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
parasail-ai/GritLM-7B-vllm | parasail-ai | text-generation | [
"pytorch",
"safetensors",
"mistral",
"mteb",
"vllm",
"text-generation",
"conversational",
"custom_code",
"dataset:GritLM/tulu2",
"arxiv:2402.09906",
"license:apache-2.0",
"model-index",
"region:us"
] | 2024-11-27T18:52:58 | 2024-12-06T03:01:43 | 6,918 | 1 | ---
datasets:
- GritLM/tulu2
license: apache-2.0
pipeline_tag: text-generation
tags:
- mteb
- vllm
inference: true
model-index:
- name: GritLM-7B
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
minishlab/potion-base-4M | minishlab | null | [
"model2vec",
"onnx",
"safetensors",
"embeddings",
"static-embeddings",
"mteb",
"sentence-transformers",
"license:mit",
"model-index",
"region:us"
] | 2024-10-29T09:40:20 | 2025-01-21T19:12:35 | 6,850 | 7 | ---
library_name: model2vec
license: mit
tags:
- embeddings
- static-embeddings
- mteb
- sentence-transformers
model-index:
- name: potion-base-4M
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en-ext)
type: mteb/amazon_counterfactual
config... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
abhinand/MedEmbed-small-v0.1 | abhinand | null | [
"sentence-transformers",
"safetensors",
"bert",
"medembed",
"medical-embedding",
"clinical-embedding",
"information-retrieval",
"mteb",
"en",
"dataset:MedicalQARetrieval",
"dataset:NFCorpus",
"dataset:PublicHealthQA",
"dataset:TRECCOVID",
"dataset:ArguAna",
"base_model:BAAI/bge-small-en-... | 2024-10-20T11:47:50 | 2024-10-23T14:49:50 | 6,846 | 8 | ---
base_model:
- BAAI/bge-small-en-v1.5
datasets:
- MedicalQARetrieval
- NFCorpus
- PublicHealthQA
- TRECCOVID
- ArguAna
language: en
license: apache-2.0
metrics:
- nDCG
- MAP
- Recall
- Precision
- MRR
tags:
- medembed
- medical-embedding
- clinical-embedding
- information-retrieval
- sentence-transformers
- mteb
mod... | [
"QUESTION_ANSWERING",
"SUMMARIZATION"
] | [
"BIOSSES",
"MEDICAL DATA",
"SCIFACT"
] |
EleutherAI/pythia-12b-deduped | EleutherAI | text-generation | [
"transformers",
"pytorch",
"gpt_neox",
"text-generation",
"causal-lm",
"pythia",
"en",
"dataset:EleutherAI/the_pile_deduplicated",
"arxiv:2304.01373",
"arxiv:2101.00027",
"arxiv:2201.07311",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible"... | 2023-02-27T14:01:34 | 2023-06-08T13:07:07 | 6,792 | 51 | ---
datasets:
- EleutherAI/the_pile_deduplicated
language:
- en
license: apache-2.0
tags:
- pytorch
- causal-lm
- pythia
---
The *Pythia Scaling Suite* is a collection of models developed to facilitate
interpretability research [(see paper)](https://arxiv.org/pdf/2304.01373.pdf).
It contains two sets of eight models... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"SCIQ"
] |
ibm-granite/granite-embedding-107m-multilingual | ibm-granite | sentence-similarity | [
"sentence-transformers",
"pytorch",
"safetensors",
"xlm-roberta",
"feature-extraction",
"language",
"granite",
"embeddings",
"multilingual",
"mteb",
"transformers",
"sentence-similarity",
"en",
"ar",
"cs",
"de",
"es",
"fr",
"it",
"ja",
"ko",
"nl",
"pt",
"zh",
"arxiv:0... | 2024-12-04T20:29:00 | 2025-03-04T15:11:23 | 6,783 | 19 | ---
language:
- en
- ar
- cs
- de
- es
- fr
- it
- ja
- ko
- nl
- pt
- zh
library_name: sentence-transformers
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- language
- granite
- embeddings
- multilingual
- mteb
- transformers
model-index:
- name: ibm-granite/granite-embedding-107m-multilingual
results:... | [
"TRANSLATION",
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
minishlab/potion-base-2M | minishlab | null | [
"model2vec",
"onnx",
"safetensors",
"embeddings",
"static-embeddings",
"mteb",
"sentence-transformers",
"license:mit",
"model-index",
"region:us"
] | 2024-10-29T09:48:42 | 2025-01-21T19:13:44 | 6,772 | 10 | ---
library_name: model2vec
license: mit
tags:
- embeddings
- static-embeddings
- mteb
- sentence-transformers
model-index:
- name: potion-base-2M
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en-ext)
type: mteb/amazon_counterfactual
config... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
abhinand/MedEmbed-base-v0.1 | abhinand | null | [
"sentence-transformers",
"safetensors",
"bert",
"medembed",
"medical-embedding",
"clinical-embedding",
"information-retrieval",
"en",
"dataset:MedicalQARetrieval",
"dataset:NFCorpus",
"dataset:PublicHealthQA",
"dataset:TRECCOVID",
"dataset:ArguAna",
"base_model:BAAI/bge-base-en-v1.5",
"b... | 2024-10-20T11:46:20 | 2024-10-21T06:49:35 | 6,316 | 14 | ---
base_model:
- BAAI/bge-base-en-v1.5
datasets:
- MedicalQARetrieval
- NFCorpus
- PublicHealthQA
- TRECCOVID
- ArguAna
language: en
license: apache-2.0
metrics:
- nDCG
- MAP
- Recall
- Precision
- MRR
tags:
- medembed
- medical-embedding
- clinical-embedding
- information-retrieval
- sentence-transformers
---
# MedE... | [
"QUESTION_ANSWERING"
] | [
"MEDICAL DATA"
] |
aisingapore/llama3-8b-cpt-sea-lionv2.1-instruct | aisingapore | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"conversational",
"en",
"id",
"ta",
"th",
"vi",
"arxiv:2309.06085",
"arxiv:2311.07911",
"arxiv:2306.05685",
"base_model:aisingapore/llama3-8b-cpt-sea-lionv2-base",
"base_model:finetune:aisingapore/llama3-8b-cpt-sea-lionv2-base",
... | 2024-07-30T00:31:49 | 2024-12-19T13:11:44 | 6,256 | 17 | ---
base_model:
- aisingapore/llama3-8b-cpt-sea-lionv2-base
language:
- en
- id
- ta
- th
- vi
license: llama3
new_version: aisingapore/llama3.1-8b-cpt-sea-lionv3-instruct
---
# Llama3 8B CPT SEA-Lionv2.1 Instruct
SEA-LION is a collection of Large Language Models (LLMs) which has been pretrained and instruct-tuned for ... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION"
] | [
"CHIA"
] |
microsoft/BioGPT-Large | microsoft | text-generation | [
"transformers",
"pytorch",
"biogpt",
"text-generation",
"medical",
"en",
"dataset:pubmed",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | 2023-02-03T16:17:26 | 2023-02-05T06:18:14 | 6,013 | 188 | ---
datasets:
- pubmed
language:
- en
library_name: transformers
license: mit
pipeline_tag: text-generation
tags:
- medical
widget:
- text: COVID-19 is
inference:
parameters:
max_new_tokens: 50
---
## BioGPT
Pre-trained language models have attracted increasing attention in the biomedical domain, inspired by th... | [
"RELATION_EXTRACTION"
] | [
"BC5CDR",
"PUBMEDQA"
] |
nvidia/NV-Embed-v1 | nvidia | null | [
"sentence-transformers",
"safetensors",
"nvembed",
"mteb",
"custom_code",
"en",
"arxiv:2210.07316",
"arxiv:2405.17428",
"license:cc-by-nc-4.0",
"model-index",
"region:us"
] | 2024-05-23T01:20:16 | 2024-11-30T23:02:27 | 5,894 | 426 | ---
language:
- en
license: cc-by-nc-4.0
tags:
- mteb
- sentence-transformers
model-index:
- name: NV-Embed-v1
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: test
revision: e... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
Lajavaness/bilingual-embedding-base | Lajavaness | sentence-similarity | [
"sentence-transformers",
"safetensors",
"bilingual",
"feature-extraction",
"sentence-similarity",
"transformers",
"sentence-embedding",
"mteb",
"custom_code",
"arxiv:2010.08240",
"arxiv:1911.02116",
"arxiv:1908.10084",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpo... | 2024-06-26T15:22:37 | 2024-11-18T11:21:39 | 5,585 | 6 | ---
library_name: sentence-transformers
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
- sentence-embedding
- mteb
model-index:
- name: bilingual-embedding-base
results:
- task:
type: Clustering
dataset:
n... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
Mihaiii/Bulbasaur | Mihaiii | sentence-similarity | [
"sentence-transformers",
"onnx",
"safetensors",
"bert",
"feature-extraction",
"sentence-similarity",
"gte",
"mteb",
"dataset:Mihaiii/qa-assistant",
"license:mit",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2024-04-27T09:53:29 | 2024-04-30T07:30:42 | 5,457 | 2 | ---
datasets:
- Mihaiii/qa-assistant
library_name: sentence-transformers
license: mit
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- gte
- mteb
model-index:
- name: Bulbasaur
results:
- task:
type: Classification
dataset:
name: MTEB Amazo... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
Backedman/TriviaAnsweringMachineREAL | Backedman | question-answering | [
"transformers",
"pytorch",
"TFIDF-QA",
"question-answering",
"custom_code",
"en",
"license:mit",
"region:us"
] | 2024-05-07T01:04:47 | 2024-05-12T22:28:55 | 5,270 | 0 | ---
language:
- en
license: mit
pipeline_tag: question-answering
---
The evaluation of this project is to answer trivia questions. You do
not need to do well at this task, but you should submit a system that
completes the task or create adversarial questions in that setting. This will help the whole class share data ... | [
"TRANSLATION"
] | [
"MEDAL"
] |
Salesforce/SFR-Embedding-2_R | Salesforce | feature-extraction | [
"sentence-transformers",
"safetensors",
"mistral",
"feature-extraction",
"mteb",
"transformers",
"en",
"license:cc-by-nc-4.0",
"model-index",
"autotrain_compatible",
"text-generation-inference",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2024-06-14T19:34:13 | 2025-02-04T21:05:42 | 5,249 | 84 | ---
language:
- en
license: cc-by-nc-4.0
tags:
- mteb
- sentence-transformers
- transformers
model-index:
- name: Salesforce/SFR-Embedding-2_R
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
Hiveurban/multilingual-e5-large-pooled | Hiveurban | feature-extraction | [
"sentence-transformers",
"pytorch",
"onnx",
"safetensors",
"xlm-roberta",
"mteb",
"Sentence Transformers",
"sentence-similarity",
"feature-extraction",
"multilingual",
"af",
"am",
"ar",
"as",
"az",
"be",
"bg",
"bn",
"br",
"bs",
"ca",
"cs",
"cy",
"da",
"de",
"el",
... | 2024-09-24T12:15:49 | 2024-09-24T13:06:40 | 5,176 | 1 | ---
language:
- multilingual
- af
- am
- ar
- as
- az
- be
- bg
- bn
- br
- bs
- ca
- cs
- cy
- da
- de
- el
- en
- eo
- es
- et
- eu
- fa
- fi
- fr
- fy
- ga
- gd
- gl
- gu
- ha
- he
- hi
- hr
- hu
- hy
- id
- is
- it
- ja
- jv
- ka
- kk
- km
- kn
- ko
- ku
- ky
- la
- lo
- lt
- lv
- mg
- mk
- ml
- mn
- mr
- ms
- my
-... | [
"SEMANTIC_SIMILARITY",
"TRANSLATION",
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
billatsectorflow/stella_en_400M_v5 | billatsectorflow | sentence-similarity | [
"sentence-transformers",
"pytorch",
"safetensors",
"new",
"feature-extraction",
"mteb",
"transformers",
"sentence-similarity",
"custom_code",
"arxiv:2205.13147",
"license:mit",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2025-01-22T10:55:37 | 2025-01-22T11:45:51 | 5,141 | 3 | ---
license: mit
tags:
- mteb
- sentence-transformers
- transformers
- sentence-similarity
model-index:
- name: stella_en_400M_v5
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: tes... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
nes470/pipeline-as-repo | nes470 | question-answering | [
"transformers",
"pytorch",
"QA-umd-quizbowl",
"question-answering",
"custom_code",
"license:mit",
"region:us"
] | 2024-05-06T15:27:54 | 2024-05-16T22:18:19 | 5,081 | 0 | ---
library_name: transformers
license: mit
---
Names: Nuran, Joshua, Robert
The evaluation of this project is to answer trivia questions. You do
not need to do well at this task, but you should submit a system that
completes the task or create adversarial questions in that setting. This will help the whol... | [
"TRANSLATION"
] | [
"MEDAL"
] |
gliner-community/gliner_small-v2.5 | gliner-community | token-classification | [
"gliner",
"pytorch",
"token-classification",
"multilingual",
"dataset:urchade/pile-mistral-v0.1",
"arxiv:2311.08526",
"license:apache-2.0",
"region:us"
] | 2024-06-17T18:11:25 | 2024-06-18T12:54:21 | 4,860 | 5 | ---
datasets:
- urchade/pile-mistral-v0.1
language:
- multilingual
library_name: gliner
license: apache-2.0
pipeline_tag: token-classification
---
# About
GLiNER is a Named Entity Recognition (NER) model capable of identifying any entity type using a bidirectional transformer encoder (BERT-like). It provides a practi... | [
"NAMED_ENTITY_RECOGNITION"
] | [
"ANATEM",
"BC5CDR"
] |
minishlab/potion-base-32M | minishlab | null | [
"model2vec",
"onnx",
"safetensors",
"embeddings",
"static-embeddings",
"mteb",
"sentence-transformers",
"license:mit",
"model-index",
"region:us"
] | 2025-01-22T17:10:47 | 2025-01-29T10:59:27 | 4,720 | 10 | ---
library_name: model2vec
license: mit
tags:
- embeddings
- static-embeddings
- mteb
- sentence-transformers
model-index:
- name: potion-base-32M
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en-ext)
type: mteb/amazon_counterfactual
confi... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
michiyasunaga/BioLinkBERT-base | michiyasunaga | text-classification | [
"transformers",
"pytorch",
"bert",
"feature-extraction",
"exbert",
"linkbert",
"biolinkbert",
"fill-mask",
"question-answering",
"text-classification",
"token-classification",
"en",
"dataset:pubmed",
"arxiv:2203.15827",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | 2022-03-08T07:22:12 | 2022-03-31T00:51:21 | 4,645 | 36 | ---
datasets:
- pubmed
language: en
license: apache-2.0
tags:
- bert
- exbert
- linkbert
- biolinkbert
- feature-extraction
- fill-mask
- question-answering
- text-classification
- token-classification
widget:
- text: Sunitinib is a tyrosine kinase inhibitor
---
## BioLinkBERT-base
BioLinkBERT-ba... | [
"TEXT_CLASSIFICATION",
"QUESTION_ANSWERING"
] | [
"BLURB",
"MEDQA",
"PUBMEDQA"
] |
joe32140/ModernBERT-base-msmarco | joe32140 | sentence-similarity | [
"sentence-transformers",
"onnx",
"safetensors",
"modernbert",
"sentence-similarity",
"feature-extraction",
"generated_from_trainer",
"dataset_size:11662655",
"loss:CachedMultipleNegativesRankingLoss",
"en",
"dataset:sentence-transformers/msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1",
"arxi... | 2024-12-20T05:00:55 | 2025-01-26T00:05:31 | 4,630 | 8 | ---
base_model: answerdotai/ModernBERT-base
datasets:
- sentence-transformers/msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1
language:
- en
library_name: sentence-transformers
metrics:
- cosine_accuracy
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- sentence-similarity
- feature-extraction
- genera... | [
"TEXT_CLASSIFICATION",
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
TaylorAI/bge-micro | TaylorAI | sentence-similarity | [
"sentence-transformers",
"pytorch",
"onnx",
"safetensors",
"bert",
"feature-extraction",
"sentence-similarity",
"transformers",
"mteb",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2023-10-07T06:46:18 | 2024-03-05T18:27:18 | 4,418 | 23 | ---
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
- mteb
model-index:
- name: bge_micro
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
SeaLLMs/SeaLLMs-v3-7B-Chat | SeaLLMs | text-generation | [
"transformers",
"safetensors",
"qwen2",
"text-generation",
"sea",
"multilingual",
"conversational",
"en",
"zh",
"id",
"vi",
"th",
"ms",
"tl",
"ta",
"jv",
"lo",
"km",
"my",
"arxiv:2407.19672",
"arxiv:2306.05179",
"license:other",
"autotrain_compatible",
"text-generation-... | 2024-07-03T03:02:02 | 2024-09-02T03:02:23 | 4,214 | 52 | ---
language:
- en
- zh
- id
- vi
- th
- ms
- tl
- ta
- jv
- lo
- km
- my
license: other
license_name: seallms
license_link: https://huggingface.co/SeaLLMs/SeaLLM-13B-Chat/blob/main/LICENSE
tags:
- sea
- multilingual
---
# *SeaLLMs-v3* - Large Language Models for Southeast Asia
<p align="center">
<a href="https://dam... | [
"TRANSLATION"
] | [
"CHIA"
] |
mixedbread-ai/mxbai-embed-2d-large-v1 | mixedbread-ai | feature-extraction | [
"sentence-transformers",
"onnx",
"safetensors",
"bert",
"feature-extraction",
"mteb",
"transformers.js",
"transformers",
"en",
"arxiv:2402.14776",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2024-03-04T11:17:53 | 2025-03-13T04:16:28 | 4,153 | 37 | ---
language:
- en
library_name: sentence-transformers
license: apache-2.0
tags:
- mteb
- transformers.js
- transformers
model-index:
- name: mxbai-embed-2d-large-v1
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactu... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
Almawave/Velvet-2B | Almawave | text-generation | [
"transformers",
"safetensors",
"mistral",
"text-generation",
"vllm",
"conversational",
"en",
"it",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"region:us"
] | 2025-02-10T15:49:00 | 2025-02-20T14:41:42 | 3,990 | 37 | ---
language:
- en
- it
library_name: transformers
license: apache-2.0
tags:
- vllm
inference: false
extra_gated_description: If you want to learn more about how we process your personal
data, please read our <a href="https://www.almawave.com/privacy-policy/">Privacy
Policy</a>.
---
# Model Card for Velvet-2B
Vel... | [
"TEXT_CLASSIFICATION",
"QUESTION_ANSWERING",
"TEXTUAL_ENTAILMENT",
"TRANSLATION",
"SUMMARIZATION",
"PARAPHRASING"
] | [
"SCIQ"
] |
jxm/cde-small-v2 | jxm | feature-extraction | [
"sentence-transformers",
"safetensors",
"feature-extraction",
"mteb",
"transformers",
"modernbert",
"custom_code",
"arxiv:2410.02525",
"base_model:answerdotai/ModernBERT-base",
"base_model:finetune:answerdotai/ModernBERT-base",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
... | 2025-01-13T18:04:14 | 2025-02-03T23:41:38 | 3,933 | 78 | ---
base_model: answerdotai/ModernBERT-base
tags:
- mteb
- transformers
- sentence-transformers
- modernbert
model-index:
- name: cde-small-v2
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"MEDAL",
"SCIFACT"
] |
aisingapore/llama3.1-8b-cpt-sea-lionv3-instruct | aisingapore | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"conversational",
"en",
"zh",
"vi",
"id",
"th",
"fil",
"ta",
"ms",
"km",
"lo",
"my",
"jv",
"su",
"arxiv:2309.06085",
"arxiv:2311.07911",
"arxiv:2306.05685",
"base_model:aisingapore/llama3.1-8b-cpt-sea-lionv3-base",
... | 2024-12-11T10:20:41 | 2024-12-19T12:49:17 | 3,931 | 4 | ---
base_model:
- aisingapore/llama3.1-8b-cpt-sea-lionv3-base
language:
- en
- zh
- vi
- id
- th
- fil
- ta
- ms
- km
- lo
- my
- jv
- su
library_name: transformers
license: llama3.1
pipeline_tag: text-generation
---
<div>
<img src="llama_3.1_8b_sea-lion_v3_instruct_banner.png"/>
</div>
# Llama3.1 8B CPT SEA-LIONv3... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"CHIA"
] |
bigscience/T0_3B | bigscience | text2text-generation | [
"transformers",
"pytorch",
"safetensors",
"t5",
"text2text-generation",
"en",
"dataset:bigscience/P3",
"arxiv:2110.08207",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | 2022-03-02T23:29:05 | 2024-03-09T13:39:07 | 3,910 | 100 | ---
datasets:
- bigscience/P3
language: en
license: apache-2.0
widget:
- text: A is the son's of B's uncle. What is the family relationship between A and
B?
- text: 'Reorder the words in this sentence: justin and name bieber years is my am
I 27 old.'
- text: "Task: copy but say the opposite.\n PSG won its match... | [
"COREFERENCE_RESOLUTION",
"TEXTUAL_ENTAILMENT",
"SUMMARIZATION"
] | [
"SCIQ"
] |
jinaai/jina-embedding-s-en-v1 | jinaai | sentence-similarity | [
"sentence-transformers",
"pytorch",
"t5",
"finetuner",
"mteb",
"feature-extraction",
"sentence-similarity",
"custom_code",
"en",
"dataset:jinaai/negation-dataset",
"arxiv:2307.11224",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | 2023-07-06T11:37:14 | 2025-01-06T16:31:58 | 3,739 | 26 | ---
datasets:
- jinaai/negation-dataset
language: en
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- finetuner
- mteb
- sentence-transformers
- feature-extraction
- sentence-similarity
model-index:
- name: jina-embedding-s-en-v1
results:
- task:
type: Classification
dataset:
name: MTEB... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"LINNAEUS",
"SCIFACT"
] |
Almawave/Velvet-14B | Almawave | text-generation | [
"transformers",
"safetensors",
"mistral",
"text-generation",
"vllm",
"conversational",
"en",
"fr",
"de",
"es",
"it",
"pt",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"region:us"
] | 2025-01-31T00:53:58 | 2025-02-20T14:38:44 | 3,649 | 129 | ---
language:
- en
- fr
- de
- es
- it
- pt
library_name: transformers
license: apache-2.0
tags:
- vllm
inference: false
extra_gated_description: If you want to learn more about how we process your personal
data, please read our <a href="https://www.almawave.com/privacy-policy/">Privacy
Policy</a>.
---
# Model Car... | [
"TEXT_CLASSIFICATION",
"QUESTION_ANSWERING",
"TEXTUAL_ENTAILMENT",
"TRANSLATION",
"SUMMARIZATION",
"PARAPHRASING"
] | [
"SCIQ"
] |
BSC-LT/salamandra-2b | BSC-LT | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"bg",
"ca",
"code",
"cs",
"cy",
"da",
"de",
"el",
"en",
"es",
"et",
"eu",
"fi",
"fr",
"ga",
"gl",
"hr",
"hu",
"it",
"lt",
"lv",
"mt",
"nl",
"nn",
"oc",
"pl",
"pt",
"ro",
"ru",
"sh",
"sk",
... | 2024-09-30T13:44:13 | 2025-02-20T16:36:39 | 3,633 | 23 | ---
datasets:
- oscar-corpus/colossal-oscar-1.0
- HuggingFaceFW/fineweb-edu
- joelniklaus/eurlex_resources
- joelito/legal-mc4
- projecte-aina/CATalog
- UFRGS/brwac
- community-datasets/hrwac
- danish-foundation-models/danish-gigaword
- HiTZ/euscrawl
- PleIAs/French-PD-Newspapers
- PleIAs/French-PD-Books
- AI-team-UoA/... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION",
"PARAPHRASING"
] | [
"BEAR",
"SCIELO"
] |
croissantllm/CroissantLLMChat-v0.1 | croissantllm | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"legal",
"code",
"text-generation-inference",
"art",
"conversational",
"fr",
"en",
"dataset:croissantllm/croissant_dataset",
"dataset:croissantllm/CroissantLLM-2201-sft",
"dataset:cerebras/SlimPajama-627B",
"dataset:uonlp/Cultura... | 2024-01-24T09:18:45 | 2024-04-26T10:02:01 | 3,614 | 50 | ---
datasets:
- croissantllm/croissant_dataset
- croissantllm/CroissantLLM-2201-sft
- cerebras/SlimPajama-627B
- uonlp/CulturaX
- pg19
- bigcode/starcoderdata
language:
- fr
- en
license: mit
pipeline_tag: text-generation
tags:
- legal
- code
- text-generation-inference
- art
---
# CroissantLLMChat (190k steps + Chat)... | [
"TRANSLATION"
] | [
"CRAFT"
] |
epfl-llm/meditron-7b | epfl-llm | text-generation | [
"transformers",
"pytorch",
"safetensors",
"llama",
"text-generation",
"en",
"dataset:epfl-llm/guidelines",
"arxiv:2311.16079",
"base_model:meta-llama/Llama-2-7b",
"base_model:finetune:meta-llama/Llama-2-7b",
"license:llama2",
"autotrain_compatible",
"text-generation-inference",
"endpoints_... | 2023-11-08T16:03:23 | 2023-12-07T19:38:26 | 3,594 | 272 | ---
base_model: meta-llama/Llama-2-7b
datasets:
- epfl-llm/guidelines
language:
- en
license: llama2
metrics:
- accuracy
- perplexity
---
<img width=50% src="meditron_LOGO.png" alt="Alt text" title="Meditron-logo">
# Model Card for Meditron-7B-v1.0
Meditron is a suite of open-source medical Large Language Models (LLMs... | [
"QUESTION_ANSWERING"
] | [
"MEDQA",
"PUBMEDQA"
] |
dwzhu/e5-base-4k | dwzhu | sentence-similarity | [
"transformers",
"pytorch",
"bert",
"feature-extraction",
"mteb",
"sentence-similarity",
"en",
"arxiv:2404.12096",
"arxiv:2104.08663",
"arxiv:2210.07316",
"license:mit",
"model-index",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2024-03-28T09:32:27 | 2024-05-14T08:22:21 | 3,389 | 10 | ---
language:
- en
license: mit
tags:
- mteb
- sentence-similarity
model-index:
- name: e5-base-4k
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: test
revision: e8379541af4e3... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
microsoft/BioGPT-Large-PubMedQA | microsoft | text-generation | [
"transformers",
"pytorch",
"biogpt",
"text-generation",
"medical",
"en",
"dataset:pubmed_qa",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | 2023-02-03T20:33:43 | 2023-02-04T07:50:25 | 3,367 | 108 | ---
datasets:
- pubmed_qa
language:
- en
library_name: transformers
license: mit
metrics:
- accuracy
pipeline_tag: text-generation
tags:
- medical
widget:
- text: 'question: Can ''high-risk'' human papillomaviruses (HPVs) be detected in
human breast milk? context: Using polymerase chain reaction techniques, we eval... | [
"RELATION_EXTRACTION"
] | [
"BC5CDR",
"PUBMEDQA"
] |
lightonai/modernbert-embed-large | lightonai | sentence-similarity | [
"sentence-transformers",
"onnx",
"safetensors",
"modernbert",
"feature-extraction",
"sentence-similarity",
"mteb",
"transformers.js",
"en",
"arxiv:2402.01613",
"arxiv:2412.13663",
"base_model:answerdotai/ModernBERT-large",
"base_model:quantized:answerdotai/ModernBERT-large",
"license:apach... | 2025-01-13T10:21:29 | 2025-01-14T10:02:51 | 3,361 | 20 | ---
base_model:
- answerdotai/ModernBERT-large
- lightonai/modernbert-embed-large-unsupervised
language:
- en
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- mteb
- transformers.js
model-index:
- name: modernbert-embed-large
results:
-... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
bijaygurung/stella_en_400M_v5 | bijaygurung | sentence-similarity | [
"sentence-transformers",
"pytorch",
"safetensors",
"new",
"feature-extraction",
"mteb",
"transformers",
"sentence-similarity",
"custom_code",
"arxiv:2205.13147",
"license:mit",
"model-index",
"autotrain_compatible",
"text-embeddings-inference",
"endpoints_compatible",
"region:us"
] | 2024-09-30T06:12:02 | 2024-09-30T17:03:05 | 3,336 | 4 | ---
license: mit
tags:
- mteb
- sentence-transformers
- transformers
- sentence-similarity
model-index:
- name: stella_en_400M_v5
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
split: tes... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
minishlab/potion-retrieval-32M | minishlab | null | [
"model2vec",
"onnx",
"safetensors",
"embeddings",
"static-embeddings",
"sentence-transformers",
"license:mit",
"region:us"
] | 2025-01-23T15:05:16 | 2025-01-29T11:00:09 | 3,271 | 17 | ---
library_name: model2vec
license: mit
model_name: potion-retrieval-32M
tags:
- embeddings
- static-embeddings
- sentence-transformers
---
# potion-retrieval-32M Model Card
<div align="center">
<img width="35%" alt="Model2Vec logo" src="https://raw.githubusercontent.com/MinishLab/model2vec/main/assets/images/lo... | [
"SUMMARIZATION"
] | [
"PUBMEDQA"
] |
BSC-LT/salamandra-2b-instruct | BSC-LT | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"conversational",
"bg",
"ca",
"code",
"cs",
"cy",
"da",
"de",
"el",
"en",
"es",
"et",
"eu",
"fi",
"fr",
"ga",
"gl",
"hr",
"hu",
"it",
"lt",
"lv",
"mt",
"nl",
"nn",
"oc",
"pl",
"pt",
"ro",
"ru... | 2024-09-30T13:44:40 | 2025-02-20T16:45:39 | 3,244 | 20 | ---
base_model:
- BSC-LT/salamandra-2b
datasets:
- oscar-corpus/colossal-oscar-1.0
- HuggingFaceFW/fineweb-edu
- joelniklaus/eurlex_resources
- joelito/legal-mc4
- projecte-aina/CATalog
- UFRGS/brwac
- community-datasets/hrwac
- danish-foundation-models/danish-gigaword
- HiTZ/euscrawl
- PleIAs/French-PD-Newspapers
- Pl... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION",
"PARAPHRASING"
] | [
"BEAR",
"SCIELO"
] |
tensorblock/gte-Qwen2-7B-instruct-GGUF | tensorblock | sentence-similarity | [
"sentence-transformers",
"gguf",
"mteb",
"transformers",
"Qwen2",
"sentence-similarity",
"TensorBlock",
"GGUF",
"base_model:Alibaba-NLP/gte-Qwen2-7B-instruct",
"base_model:quantized:Alibaba-NLP/gte-Qwen2-7B-instruct",
"license:apache-2.0",
"model-index",
"autotrain_compatible",
"endpoints_... | 2024-11-11T10:44:19 | 2024-11-16T01:05:41 | 3,236 | 8 | ---
base_model: Alibaba-NLP/gte-Qwen2-7B-instruct
license: apache-2.0
tags:
- mteb
- sentence-transformers
- transformers
- Qwen2
- sentence-similarity
- TensorBlock
- GGUF
model-index:
- name: gte-qwen2-7B-instruct
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassi... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
tasksource/ModernBERT-large-nli | tasksource | zero-shot-classification | [
"transformers",
"safetensors",
"modernbert",
"text-classification",
"instruct",
"natural-language-inference",
"nli",
"zero-shot-classification",
"en",
"dataset:nyu-mll/glue",
"dataset:facebook/anli",
"base_model:answerdotai/ModernBERT-large",
"base_model:finetune:answerdotai/ModernBERT-large... | 2025-01-04T00:56:11 | 2025-01-04T12:03:47 | 3,223 | 5 | ---
base_model:
- answerdotai/ModernBERT-large
datasets:
- nyu-mll/glue
- facebook/anli
language:
- en
library_name: transformers
license: apache-2.0
pipeline_tag: zero-shot-classification
tags:
- instruct
- natural-language-inference
- nli
---
# Model Card for Model ID
This model is ModernBERT multi-task fine-tuned ... | [
"SUMMARIZATION"
] | [
"SCIFACT",
"SCITAIL"
] |
pszemraj/long-t5-tglobal-base-16384-book-summary | pszemraj | summarization | [
"transformers",
"pytorch",
"rust",
"onnx",
"safetensors",
"longt5",
"text2text-generation",
"summarization",
"summary",
"booksum",
"long-document",
"long-form",
"dataset:kmfoda/booksum",
"arxiv:2112.07916",
"arxiv:2105.08209",
"doi:10.57967/hf/2078",
"license:apache-2.0",
"license:... | 2022-06-27T16:37:26 | 2025-01-21T13:58:25 | 3,139 | 134 | ---
datasets:
- kmfoda/booksum
license:
- apache-2.0
- bsd-3-clause
metrics:
- rouge
tags:
- summarization
- summary
- booksum
- long-document
- long-form
widget:
- text: large earthquakes along a given fault segment do not occur at random intervals
because it takes time to accumulate the strain energy for the rupt... | [
"QUESTION_ANSWERING",
"SUMMARIZATION"
] | [
"BEAR"
] |
xmanii/maux-gte-persian-v2 | xmanii | sentence-similarity | [
"sentence-transformers",
"safetensors",
"new",
"sentence-similarity",
"feature-extraction",
"generated_from_trainer",
"dataset_size:10000",
"loss:CosineSimilarityLoss",
"custom_code",
"dataset:xmanii/maux-gte-10k-public",
"arxiv:1908.10084",
"base_model:Alibaba-NLP/gte-multilingual-base",
"b... | 2024-12-25T10:05:32 | 2024-12-25T10:06:01 | 3,130 | 3 | ---
base_model: Alibaba-NLP/gte-multilingual-base
datasets:
- xmanii/maux-gte-10k-public
library_name: sentence-transformers
metrics:
- pearson_cosine
- spearman_cosine
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- sentence-similarity
- feature-extraction
- generated_from_trainer
- dataset_size:1000... | [
"TEXT_CLASSIFICATION",
"SEMANTIC_SIMILARITY"
] | [
"CPI"
] |
MaartenGr/BERTopic_Wikipedia | MaartenGr | text-classification | [
"bertopic",
"text-classification",
"region:us"
] | 2023-05-30T08:03:05 | 2023-05-31T17:58:03 | 3,095 | 43 | ---
library_name: bertopic
pipeline_tag: text-classification
tags:
- bertopic
---
# Wikipedia
This is a [BERTopic](https://github.com/MaartenGr/BERTopic) model.
BERTopic is a flexible and modular topic modeling framework that allows for the generation of easily interpretable topics from large datasets.
* Trained o... | [
"TRANSLATION"
] | [
"BEAR",
"MEDAL"
] |
VietAI/envit5-translation | VietAI | translation | [
"transformers",
"pytorch",
"tf",
"jax",
"t5",
"text2text-generation",
"translation",
"vi",
"en",
"dataset:cc100",
"license:openrail",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | 2022-10-06T14:53:36 | 2022-11-21T09:59:08 | 2,996 | 36 | ---
datasets:
- cc100
language:
- vi
- en
license: openrail
tags:
- translation
widget:
- text: 'vi: VietAI là tổ chức phi lợi nhuận với sứ mệnh ươm mầm tài năng về trí tuệ
nhân tạo và xây dựng một cộng đồng các chuyên gia trong lĩnh vực trí tuệ nhân
tạo đẳng cấp quốc tế tại Việt Nam.'
---
# EnViT5 Translation... | [
"TRANSLATION"
] | [
"CHIA"
] |
aisingapore/llama3.1-70b-cpt-sea-lionv3-instruct | aisingapore | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"conversational",
"en",
"zh",
"vi",
"id",
"th",
"fil",
"ta",
"ms",
"km",
"lo",
"my",
"jv",
"su",
"arxiv:2309.06085",
"arxiv:2311.07911",
"arxiv:2306.05685",
"base_model:aisingapore/llama3.1-70b-cpt-sea-lionv3-base",... | 2024-12-11T10:22:38 | 2024-12-19T13:58:00 | 2,958 | 1 | ---
base_model:
- aisingapore/llama3.1-70b-cpt-sea-lionv3-base
language:
- en
- zh
- vi
- id
- th
- fil
- ta
- ms
- km
- lo
- my
- jv
- su
library_name: transformers
license: llama3.1
pipeline_tag: text-generation
base_model_relation: finetune
---
<div>
<img src="llama_3.1_70b_sea-lion_v3_instruct_banner.png"/>
</di... | [
"QUESTION_ANSWERING",
"TRANSLATION"
] | [
"CHIA"
] |
cnmoro/snowflake-arctic-embed-m-v2.0-cpu | cnmoro | sentence-similarity | [
"sentence-transformers",
"safetensors",
"gte",
"feature-extraction",
"sentence-similarity",
"mteb",
"arctic",
"snowflake-arctic-embed",
"transformers.js",
"custom_code",
"af",
"ar",
"az",
"be",
"bg",
"bn",
"ca",
"ceb",
"cs",
"cy",
"da",
"de",
"el",
"en",
"es",
"et",... | 2025-01-22T20:22:34 | 2025-01-22T20:37:49 | 2,856 | 2 | ---
language:
- af
- ar
- az
- be
- bg
- bn
- ca
- ceb
- cs
- cy
- da
- de
- el
- en
- es
- et
- eu
- fa
- fi
- fr
- gl
- gu
- he
- hi
- hr
- ht
- hu
- hy
- id
- is
- it
- ja
- jv
- ka
- kk
- km
- kn
- ko
- ky
- lo
- lt
- lv
- mk
- ml
- mn
- mr
- ms
- my
- ne
- nl
- pa
- pl
- pt
- qu
- ro
- ru
- si
- sk
- sl
- so
- sq
... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
maddes8cht/tiiuae-falcon-40b | maddes8cht | null | [
"gguf",
"en",
"de",
"es",
"fr",
"dataset:tiiuae/falcon-refinedweb",
"arxiv:2205.14135",
"arxiv:1911.02150",
"arxiv:2101.00027",
"arxiv:2005.14165",
"arxiv:2104.09864",
"arxiv:2306.01116",
"license:apache-2.0",
"region:us"
] | 2023-09-15T09:48:14 | 2023-10-20T05:22:35 | 2,737 | 1 | ---
datasets:
- tiiuae/falcon-refinedweb
language:
- en
- de
- es
- fr
license: apache-2.0
inference: false
---
[]()
I am continuously enhancing the structure of these model descriptions, and they now provide even more comprehensive informati... | [
"SUMMARIZATION"
] | [
"BEAR"
] |
McGill-NLP/LLM2Vec-Sheared-LLaMA-mntp-unsup-simcse | McGill-NLP | sentence-similarity | [
"peft",
"safetensors",
"text-embedding",
"embeddings",
"information-retrieval",
"beir",
"text-classification",
"language-model",
"text-clustering",
"text-semantic-similarity",
"text-evaluation",
"text-reranking",
"feature-extraction",
"sentence-similarity",
"Sentence Similarity",
"natu... | 2024-04-04T14:10:21 | 2024-04-11T19:55:37 | 2,528 | 1 | ---
language:
- en
library_name: peft
license: mit
pipeline_tag: sentence-similarity
tags:
- text-embedding
- embeddings
- information-retrieval
- beir
- text-classification
- language-model
- text-clustering
- text-semantic-similarity
- text-evaluation
- text-reranking
- feature-extraction
- sentence-similarity
- Sent... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
gmonsoon/gemma2-9b-cpt-sahabatai-v1-instruct-GGUF | gmonsoon | null | [
"gguf",
"en",
"id",
"jv",
"su",
"arxiv:2309.06085",
"arxiv:2310.04928",
"arxiv:2311.07911",
"base_model:GoToCompany/gemma2-9b-cpt-sahabatai-v1-instruct",
"base_model:quantized:GoToCompany/gemma2-9b-cpt-sahabatai-v1-instruct",
"license:gemma",
"endpoints_compatible",
"region:us",
"conversat... | 2024-11-14T11:35:56 | 2024-11-15T19:05:41 | 2,514 | 5 | ---
base_model:
- GoToCompany/gemma2-9b-cpt-sahabatai-v1-instruct
language:
- en
- id
- jv
- su
license: gemma
---
# Gemma2 9B CPT Sahabat-AI v1 Instruct
**Sahabat-AI** (Indonesian language for “close friends”) is a collection of Large Language Models (LLMs) which has been pretrained and instruct-tuned for Indonesian ... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION"
] | [
"CHIA"
] |
blevlabs/stella_en_v5 | blevlabs | sentence-similarity | [
"sentence-transformers",
"pytorch",
"safetensors",
"qwen2",
"text-generation",
"mteb",
"transformers",
"sentence-similarity",
"custom_code",
"arxiv:2205.13147",
"license:mit",
"model-index",
"autotrain_compatible",
"text-generation-inference",
"text-embeddings-inference",
"endpoints_co... | 2024-12-05T20:06:08 | 2024-12-05T20:18:30 | 2,494 | 3 | ---
license: mit
tags:
- mteb
- sentence-transformers
- transformers
- sentence-similarity
model-index:
- name: stella_en_1.5B_v5
results:
- task:
type: Classification
dataset:
name: MTEB AmazonCounterfactualClassification (en)
type: mteb/amazon_counterfactual
config: en
... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
gmonsoon/llama3-8b-cpt-sahabatai-v1-instruct-GGUF | gmonsoon | null | [
"gguf",
"en",
"id",
"jv",
"su",
"arxiv:2309.06085",
"arxiv:2310.04928",
"arxiv:2311.07911",
"base_model:GoToCompany/llama3-8b-cpt-sahabatai-v1-instruct",
"base_model:quantized:GoToCompany/llama3-8b-cpt-sahabatai-v1-instruct",
"license:llama3",
"endpoints_compatible",
"region:us",
"conversa... | 2024-11-14T17:04:01 | 2024-11-15T17:08:53 | 2,445 | 1 | ---
base_model:
- GoToCompany/llama3-8b-cpt-sahabatai-v1-instruct
language:
- en
- id
- jv
- su
license: llama3
---
# Llama3 8B CPT Sahabat-AI v1 Instruct
**Sahabat-AI** (Indonesian language for “close friends”) is a collection of Large Language Models (LLMs) which has been pretrained and instruct-tuned for Indonesian... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION"
] | [
"CHIA"
] |
RichardErkhov/Alibaba-NLP_-_gte-Qwen2-1.5B-instruct-gguf | RichardErkhov | null | [
"gguf",
"arxiv:2308.03281",
"endpoints_compatible",
"region:us",
"conversational"
] | 2024-07-14T08:23:13 | 2024-07-14T10:05:13 | 2,409 | 0 | ---
{}
---
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
gte-Qwen2-1.5B-instruct - GGUF
- Model creator: https://huggingface.co/Alibaba-NLP/
- Original model: https://hug... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
TheBloke/meditron-70B-GPTQ | TheBloke | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"medical",
"health",
"llama2",
"en",
"dataset:bigbio/med_qa",
"dataset:medmcqa",
"dataset:bigbio/pubmed_qa",
"dataset:epfl-llm/guidelines",
"arxiv:2311.16079",
"base_model:epfl-llm/meditron-70b",
"base_model:quantized:epfl-llm/me... | 2023-11-30T17:10:33 | 2023-11-30T21:24:40 | 2,391 | 4 | ---
base_model: epfl-llm/meditron-70b
datasets:
- bigbio/med_qa
- medmcqa
- bigbio/pubmed_qa
- epfl-llm/guidelines
language:
- en
license: llama2
metrics:
- accuracy
- perplexity
model_name: Meditron 70B
pipeline_tag: text-generation
tags:
- medical
- health
- llama2
inference: false
model_creator: EPFL LLM Team
model_... | [
"QUESTION_ANSWERING"
] | [
"MEDQA",
"PUBMEDQA"
] |
Alibaba-NLP/gme-Qwen2-VL-7B-Instruct | Alibaba-NLP | sentence-similarity | [
"sentence-transformers",
"safetensors",
"qwen2_vl",
"image-text-to-text",
"mteb",
"transformers",
"Qwen2-VL",
"sentence-similarity",
"vidore",
"en",
"zh",
"arxiv:2412.16855",
"base_model:Qwen/Qwen2-VL-7B-Instruct",
"base_model:finetune:Qwen/Qwen2-VL-7B-Instruct",
"license:apache-2.0",
... | 2024-12-21T04:00:17 | 2025-01-21T11:53:23 | 2,333 | 26 | ---
base_model:
- Qwen/Qwen2-VL-7B-Instruct
language:
- en
- zh
license: apache-2.0
tags:
- mteb
- sentence-transformers
- transformers
- Qwen2-VL
- sentence-similarity
- vidore
model-index:
- name: gme-Qwen2-VL-7B-Instruct
results:
- task:
type: STS
dataset:
name: MTEB AFQMC
type: C-MTEB/AFQM... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp-unsup-simcse | McGill-NLP | sentence-similarity | [
"peft",
"safetensors",
"text-embedding",
"embeddings",
"information-retrieval",
"beir",
"text-classification",
"language-model",
"text-clustering",
"text-semantic-similarity",
"text-evaluation",
"text-reranking",
"feature-extraction",
"sentence-similarity",
"Sentence Similarity",
"natu... | 2024-04-30T02:45:32 | 2024-04-30T03:42:49 | 2,287 | 4 | ---
language:
- en
library_name: peft
license: mit
pipeline_tag: sentence-similarity
tags:
- text-embedding
- embeddings
- information-retrieval
- beir
- text-classification
- language-model
- text-clustering
- text-semantic-similarity
- text-evaluation
- text-reranking
- feature-extraction
- sentence-similarity
- Sent... | [
"SUMMARIZATION"
] | [
"BIOSSES",
"SCIFACT"
] |
Dampish/StellarX-4B-V0.2 | Dampish | text-generation | [
"transformers",
"pytorch",
"gpt_neox",
"text-generation",
"arxiv:2204.06745",
"license:cc-by-nc-sa-4.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | 2023-06-03T21:05:01 | 2023-09-18T12:13:30 | 2,264 | 2 | ---
license: cc-by-nc-sa-4.0
---
# StellarX: A Base Model by Dampish and Arkane
StellarX is a powerful autoregressive language model designed for various natural language processing tasks. It has been trained on a massive dataset containing 810 billion tokens(trained on 300B tokens), trained on "redpajama," and is bui... | [
"TRANSLATION"
] | [
"SCIQ"
] |
epfl-llm/meditron-70b | epfl-llm | text-generation | [
"transformers",
"pytorch",
"safetensors",
"llama",
"text-generation",
"medical",
"health",
"llama2",
"en",
"dataset:bigbio/med_qa",
"dataset:medmcqa",
"dataset:bigbio/pubmed_qa",
"dataset:epfl-llm/guidelines",
"arxiv:2311.16079",
"base_model:meta-llama/Llama-2-70b",
"base_model:finetun... | 2023-11-08T13:57:04 | 2023-12-07T19:39:04 | 2,262 | 234 | ---
base_model: meta-llama/Llama-2-70b
datasets:
- bigbio/med_qa
- medmcqa
- bigbio/pubmed_qa
- epfl-llm/guidelines
language:
- en
license: llama2
metrics:
- accuracy
- perplexity
pipeline_tag: text-generation
tags:
- medical
- health
- llama2
---
<img width=50% src="meditron_LOGO.png" alt="Alt text" title="Meditron-l... | [
"QUESTION_ANSWERING"
] | [
"MEDQA",
"PUBMEDQA"
] |
Dampish/StellarX-4B-V0 | Dampish | text-generation | [
"transformers",
"pytorch",
"gpt_neox",
"text-generation",
"arxiv:2204.06745",
"license:cc-by-nc-sa-4.0",
"autotrain_compatible",
"text-generation-inference",
"endpoints_compatible",
"region:us"
] | 2023-05-27T19:04:49 | 2023-12-03T19:52:22 | 2,229 | 1 | ---
license: cc-by-nc-sa-4.0
---
# StellarX: A Base Model by Dampish and Arkane
StellarX is a powerful autoregressive language model designed for various natural language processing tasks. It has been trained on a massive dataset containing 810 billion tokens, trained on "redpajama," and is built upon the popular GPT-... | [
"TRANSLATION"
] | [
"SCIQ"
] |
BSC-LT/salamandra-7b | BSC-LT | text-generation | [
"transformers",
"safetensors",
"llama",
"text-generation",
"bg",
"ca",
"code",
"cs",
"cy",
"da",
"de",
"el",
"en",
"es",
"et",
"eu",
"fi",
"fr",
"ga",
"gl",
"hr",
"hu",
"it",
"lt",
"lv",
"mt",
"nl",
"nn",
"oc",
"pl",
"pt",
"ro",
"ru",
"sh",
"sk",
... | 2024-09-30T06:47:04 | 2025-02-20T16:42:39 | 2,229 | 27 | ---
datasets:
- oscar-corpus/colossal-oscar-1.0
- HuggingFaceFW/fineweb-edu
- joelniklaus/eurlex_resources
- joelito/legal-mc4
- projecte-aina/CATalog
- UFRGS/brwac
- community-datasets/hrwac
- danish-foundation-models/danish-gigaword
- HiTZ/euscrawl
- PleIAs/French-PD-Newspapers
- PleIAs/French-PD-Books
- AI-team-UoA/... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION",
"PARAPHRASING"
] | [
"BEAR",
"SCIELO"
] |
HiTZ/GoLLIE-7B | HiTZ | text-generation | [
"transformers",
"pytorch",
"llama",
"text-generation",
"code",
"text-generation-inference",
"Information Extraction",
"IE",
"Named Entity Recogniton",
"Event Extraction",
"Relation Extraction",
"LLaMA",
"custom_code",
"en",
"dataset:ACE05",
"dataset:bc5cdr",
"dataset:conll2003",
"d... | 2023-09-25T10:24:52 | 2023-10-10T07:51:44 | 2,215 | 28 | ---
datasets:
- ACE05
- bc5cdr
- conll2003
- ncbi_disease
- conll2012_ontonotesv5
- rams
- tacred
- wnut_17
language:
- en
license: llama2
metrics:
- f1
pipeline_tag: text-generation
tags:
- code
- text-generation-inference
- Information Extraction
- IE
- Named Entity Recogniton
- Event Extraction
- Relation Extraction... | [
"RELATION_EXTRACTION",
"EVENT_EXTRACTION"
] | [
"BC5CDR",
"NCBI DISEASE"
] |
GoToCompany/gemma2-9b-cpt-sahabatai-v1-instruct | GoToCompany | null | [
"safetensors",
"gemma2",
"en",
"id",
"jv",
"su",
"arxiv:2309.06085",
"arxiv:2310.04928",
"arxiv:2311.07911",
"base_model:GoToCompany/gemma2-9b-cpt-sahabatai-v1-base",
"base_model:finetune:GoToCompany/gemma2-9b-cpt-sahabatai-v1-base",
"license:gemma",
"region:us"
] | 2024-11-06T04:51:58 | 2024-11-06T04:51:58 | 2,211 | 35 | ---
base_model:
- GoToCompany/gemma2-9b-cpt-sahabatai-v1-base
language:
- en
- id
- jv
- su
license: gemma
---
# Gemma2 9B CPT Sahabat-AI v1 Instruct
**Sahabat-AI** (Indonesian language for “close friends”) is a collection of Large Language Models (LLMs) which has been pretrained and instruct-tuned for Indonesian lang... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION"
] | [
"CHIA"
] |
aliakseilabanau/bge-small-en | aliakseilabanau | null | [
"safetensors",
"openvino",
"bert",
"mteb",
"sentence transformers",
"en",
"arxiv:2311.13534",
"arxiv:2310.07554",
"arxiv:2309.07597",
"base_model:BAAI/bge-small-en",
"base_model:quantized:BAAI/bge-small-en",
"license:mit",
"region:us"
] | 2024-11-19T18:27:44 | 2025-02-04T13:09:46 | 2,173 | 0 | ---
base_model:
- BAAI/bge-small-en
language:
- en
license: mit
tags:
- mteb
- sentence transformers
- openvino
base_model_relation: quantized
---
The [BAAI/bge-small-en](https://huggingface.co/BAAI/bge-small-en) converted for [openvino backend](https://sbert.net/docs/sentence_transformer/usage/efficiency.html#openvin... | [
"SEMANTIC_SIMILARITY",
"SUMMARIZATION"
] | [
"BEAR"
] |
JosephusCheung/Guanaco | JosephusCheung | text-generation | [
"transformers",
"pytorch",
"llama",
"text-generation",
"guannaco",
"alpaca",
"conversational",
"en",
"zh",
"ja",
"de",
"dataset:JosephusCheung/GuanacoDataset",
"doi:10.57967/hf/0607",
"license:gpl-3.0",
"autotrain_compatible",
"text-generation-inference",
"region:us"
] | 2023-04-08T03:03:14 | 2023-05-29T12:48:21 | 2,164 | 230 | ---
datasets:
- JosephusCheung/GuanacoDataset
language:
- en
- zh
- ja
- de
license: gpl-3.0
pipeline_tag: conversational
tags:
- llama
- guannaco
- alpaca
inference: false
---
**You can run on Colab free T4 GPU now**
[![Open In Colab](... | [
"TRANSLATION"
] | [
"BEAR"
] |
mrm8488/modernbert-embed-base-ft-sts-spanish-matryoshka-768-64 | mrm8488 | sentence-similarity | [
"sentence-transformers",
"safetensors",
"modernbert",
"sentence-similarity",
"feature-extraction",
"generated_from_trainer",
"dataset_size:2697",
"loss:MatryoshkaLoss",
"loss:CoSENTLoss",
"arxiv:1908.10084",
"arxiv:2205.13147",
"base_model:nomic-ai/modernbert-embed-base",
"base_model:finetun... | 2025-01-10T21:44:41 | 2025-01-10T21:50:00 | 2,130 | 2 | ---
base_model: nomic-ai/modernbert-embed-base
library_name: sentence-transformers
metrics:
- pearson_cosine
- spearman_cosine
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- sentence-similarity
- feature-extraction
- generated_from_trainer
- dataset_size:2697
- loss:MatryoshkaLoss
- loss:CoSENTLoss
w... | [
"TEXT_CLASSIFICATION",
"SEMANTIC_SIMILARITY"
] | [
"CPI"
] |
aisingapore/gemma2-9b-cpt-sea-lionv3-instruct | aisingapore | text-generation | [
"transformers",
"safetensors",
"gemma2",
"text-generation",
"conversational",
"en",
"zh",
"vi",
"id",
"th",
"fil",
"ta",
"ms",
"km",
"lo",
"my",
"jv",
"su",
"arxiv:2309.06085",
"arxiv:2311.07911",
"arxiv:2306.05685",
"base_model:aisingapore/gemma2-9b-cpt-sea-lionv3-base",
... | 2024-10-30T03:19:20 | 2024-12-19T12:57:01 | 2,114 | 10 | ---
base_model:
- aisingapore/gemma2-9b-cpt-sea-lionv3-base
language:
- en
- zh
- vi
- id
- th
- fil
- ta
- ms
- km
- lo
- my
- jv
- su
library_name: transformers
license: gemma
pipeline_tag: text-generation
---
<div>
<img src="gemma_2_9b_sea-lion_v3_instruct_banner.png"/>
</div>
# Gemma2 9B CPT SEA-LIONv3 Instruct... | [
"QUESTION_ANSWERING",
"TRANSLATION",
"SUMMARIZATION"
] | [
"CHIA"
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.