Due to projects like Explore the LLMs specializing in model indexing, the custom list has been removed.
| Model | Link | Description | Date added |
|---|---|---|---|
| Moxin LLM | https://huggingface.co/moxin-org/moxin-llm-7b | Fully open data, open training 7B base and chat fine tuned model | 2024-12-20 |
| Bamba-9b | https://huggingface.co/blog/bamba | Hybrid Mamba2 model by IBM, Princeton, CMU, UIUC trained on open data with 2.5x throughput available for vLLM, TRL, llama and transformers | 2024-12-20 |
| granite 3.1 | https://huggingface.co/collections/ibm-granite/granite-31-language-models-6751dbbf2f3389bec5c6f02d | apache 2 licenced models from 1b, 2b, 3b and 8b with 128K context from IBM | 2024-12-20 |
| Command R7B | https://huggingface.co/CohereForAI/c4ai-command-r7b-12-2024 | open weights research 7B model with reasoning, summarization, question answering, coding, tool use and RAG capabilities | 2024-12-20 |
| DeepSeek-V2.5-1210-236B | https://huggingface.co/deepseek-ai/DeepSeek-V2.5-1210 | 1210 improvement over original V2.5 with Math, Coding and Reasoning improvements | 2024-12-20 |
| Granite-3.0-8b-lab-community | https://huggingface.co/instructlab/granite-3.0-8b-lab-community | community-driven, openly-reproducible base and instruct model with open data and method | 2024-12-20 |
| QwQ-32b | https://qwenlm.github.io/blog/qwq-32b-preview/ | Apache 2 licensed LLM from Alibaba Cloud's Qwen team, inspired by OpenAI's o1 reasoning model for test time compute via reasoning tokens to improve performance | 2024-12-02 |
| Sparse-Llama-3.1-8B-2of4 | https://neuralmagic.com/blog/24-sparse-llama-smaller-models-for-efficient-gpu-inference/ | 2:4 Sparse Llama: Smaller Models for Efficient GPU Inference | 2024-12-02 |
| CursorCore | https://huggingface.co/collections/TechxGenus/cursorcore-series-6706618c38598468866b60e2 | Coding LLMs for use within CursorCore and CursorWeb | |
| ichigo | https://huggingface.co/homebrewltd | an open research project extending text-based llama3 to have native "listening" ability, using an early fusion technique, with improved multiturn capabilities and refusal to process inaudible queries | |
| Zamba2 | https://www.zyphra.com/post/zamba2-7b | a 7B SOTA SML for running on-device with 25% faster first token time and 20% token per second rate compared to other architectures using Mamba2 blocks interleaved shared attention blocks and LoRA shared MLP block | |
| reader-lm | https://jina.ai/news/reader-lm-small-language-models-for-cleaning-and-converting-html-to-markdown | Jina AI's LLM to convert HTML to Markdown, making heuristics, cleanup and content identification an LLM task | |
| Pixtral | https://huggingface.co/mistralai/Pixtral-12B-2409 | 12B LLM with a 400M vision encoder for multi modal image and text inference and 128k sequence length by Mistral | |
| llama-3.2 | https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/ | small and medium sized vision LLMs in 11b and 90b and text only 1b and 3b models by Meta | |
| gemma2 2b | https://huggingface.co/bartowski/gemma-2-2b-it-GGUF | 2b small language model by google achieving SOTA performance for sub 3b models on LLM Leaderboard 2 | |
| DeepSeekCoderv2 | https://github.com/deepseek-ai/DeepSeek-Coder-V2?tab=readme-ov-file#2-model-downloads | 16b and 236b mixture of experts coding models with 128k context length | |
| codegemma | https://huggingface.co/google/codegemma-7b | google's coding models from 2b base, 7b base and 7b instruct | |
| granite | https://huggingface.co/collections/ibm-granite/granite-code-models-6624c5cec322e4c148c8b330 | IBMs code models available in 3b, 8b, 20b size as base and instruct variants with up to 128k context size | |
| codeqwen1.5 | https://huggingface.co/Qwen/CodeQwen1.5-7B | base and chat models with 7B parameters and good quality | |
| Qwen2 | https://huggingface.co/collections/Qwen/qwen2-6659360b33528ced941e557f | English and Chinese models from 0.5b, 1.5b, 7b, and 72b sizes with great performance and 128k context windows for the 7 and 72b models | |
| Phi | https://huggingface.co/collections/microsoft/phi-3-6626e15e9585a200d2d761e3 | Microsoft's small language and vision models with small and medium parameter sizes, short and long context lengths and great performance | |
| Yi-1.5 | [https://huggingface.co/01-ai/Yi-9B](https://huggingface.co/01-ai/Yi-1.5-34B-Chat | 9b model focusing on multilingual text understanding, available as 9B and 34B variants | |
| InternLM2.5 | https://huggingface.co/internlm/internlm2_5-7b-chat | 7B base and chat models focusing reasoning, math and tool use and 1M context window | |
| Mistral-Large | https://huggingface.co/mistralai/Mistral-Large-Instruct-2407 | a 123B sized model beating llama-3.1 and gpt-4o in several categories with a focus on multilinguality, coding, agentic tasks and reasoning. | |
| Llama-3.1 | https://ai.meta.com/blog/meta-llama-3-1/ | Metas most advanced model providing 8b, 70b and 405b base and instruction tuned models and 128k context window with on par quality of current SOTA closed source models | |
| Nuextract | https://huggingface.co/numind/NuExtract | is a structure extraction model based on phi-3-mini, allowing to instruct based on a json template that the model fills from unstructured text provided | |
| Mistral Nemo | https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407 | a 12B model by mistral and nvidia offering 128k context window offered as instruct and base models | |
| CodeGeeX4 | https://huggingface.co/THUDM/codegeex4-all-9b | 9B multilingual code generation model for chat and instruct with a 128k context length | |
| Mamba-Codestral | https://huggingface.co/mistralai/Mamba-Codestral-7B-v0.1 | by mistral based on the Mamba2 architecture performing on par with SOTA transformer based code models | |
| Aya-23 | https://huggingface.co/CohereForAI/aya-23-35B | 8B and 35B instruction tuned multi lingual model focusing on 23 languages | |
| Mistral-7b-instruct-v0.3 | https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3 | with function calling, new tokenizer and 32k max context | |
| CodeStral-22B | https://huggingface.co/mistralai/Codestral-22B-v0.1 | Coding model trained on 80+ languages with instruct and Fill in the Middle tasks, 32k max context | |
| Yuan2-M32 | https://huggingface.co/IEITYuan/Yuan2-M32-hf | Mixture of Experts with Attention Router, 32 Experts, 2 Active, TOtal 40B parameters, 3.7B active and max length of 16K | |
| DeepSeek-V2 | https://github.com/deepseek-ai/DeepSeek-V2#2-model-downloads | 21B Strong, Economical, and Efficient Mixture-of-Experts Language Model | |
| Granite | https://huggingface.co/ibm-granite | family of Code Models from IBM with 3b, 8b, 20b, 34b, base and instruct models for code completion and chat | |
| GemMoE | https://huggingface.co/Crystalcareai/GemMoE-Base-Random | An 8x8 Mixture Of Experts based on Gemma | |
| wavecoder-ultra-6.7b | https://huggingface.co/microsoft/wavecoder-ultra-6.7b | covering four general code-related tasks: code generation, code summary, code translation, and code repair | |
| Mixtral-8x22B-Instruct-v0.1 | https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1 | an instruct fine-tuned version of the Mixtral-8x22B-v0.1 | |
| WizardLM-2-8x22B | https://huggingface.co/alpindale/WizardLM-2-8x22B | Microsoft's WizardLM 2 8x22B beating gpt-4-0314 on MT-Bench | |
| WizardLM-2-7B | https://huggingface.co/microsoft/WizardLM-2-7B | Microsoft's WizardLM 2 7B, release for 70B coming up backup0 | |
| aiXcoder | https://huggingface.co/aiXcoder/aixcoder-7b-base | 7B Code LLM for code completion, comprehension, generation | |
| Mixtral-8x22B-v0.1 | https://huggingface.co/v2ray/Mixtral-8x22B-v0.1 | Sparse MoE model with 176B total and 44B active parameters, 65k context size | |
| grok-1 | https://huggingface.co/xai-org/grok-1 | 314b MoE model by xAI | |
| DBRX | https://huggingface.co/databricks/dbrx-base | base and instruct MoE models from databricks with 132B total parameters and a larger number of smaller experts supporting RoPE and 32K context size | |
| command-r-plus | https://huggingface.co/CohereForAI/c4ai-command-r-plus | a 104B model with highly advanced capabilities including RAG and tool use for English, French, Spanish, Italian, German, Brazilian Portuguese, Japanese, Korean, Arabic, and Simplified Chinese | |
| StarCoder2 | https://huggingface.co/bigcode/starcoder2-15b | 15B, 7B and 3B code completion models trained on The Stack v2 | |
| command-r | https://www.maginative.com/article/cohere-launches-command-r-scalable-ai-model-for-enterprise-rag-and-tool-use/ | 35B optimized for retrieval augmented generation (RAG) and tool use supporting Embed and Rerank methodology. model weights | |
| AI21 Jamba | https://huggingface.co/ai21labs/Jamba-v0.1 | production-grade Mamba-based hybrid SSM-Transformer Model licensed under Apache 2.0 with 256K context and 52B MoE at 12B each | |
| Smaug-72B | https://huggingface.co/abacusai/Smaug-72B-v0.1 | Based on Qwen-72B and MoMo-72B-Lora then finetuned by Abacus.AI, is the best performing Open LLM on the HF leaderboard by Feb-2024 | |
| SLIM Model Family | https://huggingface.co/llmware | Small Specialized Function-Calling Models for Multi-Step Automation, focused on enterprise RAG workflows | |
| aya-101 | https://huggingface.co/CohereForAI/aya-101 | 13b model fine tuned open acess multilingual LLM from Cohere For AI | |
| seamlessM4T v2 | https://huggingface.co/docs/transformers/en/model_doc/seamless_m4t_v2 | Multimodal Audio and Text Translation between many languages | |
| SeaLLM | https://huggingface.co/SeaLLMs/SeaLLM-7B-v2 | multilingual LLM for Southeast Asian (SEA) languages 🇬🇧 🇨🇳 🇻🇳 🇮🇩 🇹🇭 🇲🇾 🇰🇭 🇱🇦 🇲🇲 🇵🇭 | |
| meditron | https://github.com/epfLLM/meditron | 7B and 70B Llama2 based LLM fine tuning adapted for the medical domain | |
| Mixtral of experts | https://mistral.ai/news/mixtral-of-experts/ | A high quality Sparse Mixture-of-Experts. | |
| Poro | https://huggingface.co/LumiOpen/Poro-34B | SiloGen model checkpoints of a family of multilingual open source LLMs covering all official European languages and code, news | |
| deepseek-coder | https://github.com/deepseek-ai/DeepSeek-Coder | code language models, trained on 2T tokens, 87% code 13% English / Chinese, up to 33B with 16K context size achieving SOTA performance on coding benchmarks | |
| openchat | https://github.com/imoneoi/openchat | Advancing Open-source Language Models with Mixed-Quality Data | |
| llmware RAG models | https://huggingface.co/llmware | small LLMs and sentence transformer embedding models specifically fine-tuned for RAG workflows | |
| HelixNet | https://huggingface.co/migtissera/HelixNet | Mixture of Experts with 3 Mistral-7B, LoRA, HelixNet-LMoE optimized version | |
| Mistral-7B-german-assistant-v3 | https://huggingface.co/flozi00/Mistral-7B-german-assistant-v3 | finetuned version for german instructions and conversations in style of Alpaca. "### Assistant:" "### User:", trained with a context length of 8k tokens. The dataset used is deduplicated and cleaned, with no codes inside. The focus is on instruction following and conversational tasks | |
| WizardMath-70B-V1.0 | https://huggingface.co/WizardLM/WizardMath-70B-V1.0 | SOTA Mathematical Reasoning | |
| leo-hessianai-13b-chat-bilingual | https://huggingface.co/LeoLM/leo-hessianai-13b-chat-bilingual | based on llama-2 13b is a fine tune of the base leo-hessianai-13b for chat | |
| em_german_leo_mistral | https://huggingface.co/jphme/em_german_leo_mistral | LeoLM Mistral fine tune of LeoLM with german instructions | |
| SauerkrautLM-13B-v1 | https://huggingface.co/VAGOsolutions/SauerkrautLM-13b-v1 | fine tuned llama-2 13b on a mix of German data augmentation and translations, SauerkrautLM-7b-v1-mistral German SauerkrautLM-7b fine-tuned using QLoRA on 1 A100 80GB with Axolotl | |
| CodeShell | https://github.com/WisdomShell/codeshell/blob/main/README_EN.md | code LLM with 7b parameters trained on 500b tokens, context length of 8k outperforming CodeLlama and Starcoder on humaneval, weights | |
| sqlcoder | https://github.com/defog-ai/sqlcoder | 15B parameter model that outperforms gpt-3.5-turbo for natural language to SQL generation tasks | |
| ChatGLM2-6B | https://github.com/THUDM/ChatGLM2-6B | v2 of the GLM 6B open bilingual EN/CN model | |
| baichuan-7b | https://github.com/baichuan-inc/baichuan-7B | Baichuan Intelligent Technology developed baichuan-7B, an open-source language model with 7 billion parameters trained on 1.2 trillion tokens. Supporting Chinese and English, it achieves top performance on authoritative benchmarks (C-EVAL, MMLU) | |
| salesforce/CodeT5 | https://github.com/salesforce/codet5 | code assistant, has released their codet5+ 16b and other model sizes | |
| VPGTrans | https://vpgtrans.github.io/ | Transfer Visual Prompt Generator across LLMs and the VL-Vicuna model is a novel VL-LLM. Paper, code | |
| replit-code | https://huggingface.co/replit/ | focused on Code Completion. The model has been trained on a subset of the Stack Dedup v1.2 dataset. | |
| Visual-med-alpaca | https://github.com/cambridgeltl/visual-med-alpaca | fine-tuning llama-7b on self instruct for the biomedical domain. Models locked behind a request form. | |
| Multimodal-GPT | https://github.com/open-mmlab/Multimodal-GPT | multi-modal visual/language chatbot, using llama with custom LoRA weights and openflamingo-9B. | |
| mPLUG-Owl | https://github.com/X-PLUG/mPLUG-Owl | Multimodal finetuned model for visual/language tasks | |
| MOSS by Fudan University | https://github.com/OpenLMLab/MOSS | a 16b Chinese/English custom foundational model with additional models fine tuned on sft and plugin usage | |
| BigCode | Open Scientific collaboration to train a coding LLM | https://huggingface.co/bigcode | |
| CodeGeeX 13B | Multi Language Code Generation Model | https://huggingface.co/spaces/THUDM/CodeGeeX | |
| RWKV: Parallelizable RNN with Transformer-level LLM Performance | https://github.com/BlinkDL/RWKV-LM | ||
| GeoV/GeoV-9b | https://huggingface.co/GeoV/GeoV-9b | 9B parameter, in-progress training to 300B tokens (33:1) | |
| LAION OpenFlamingo | Multi Modal Model and training architecture | https://github.com/mlfoundations/open_flamingo | |
| Cerebras GPT-13b | https://huggingface.co/cerebras | (release notes) |