stair-lab/platinum_detect · Datasets at Hugging Face

name stringlengths 10 47	display_name stringlengths 5 41	short_display_name stringlengths 5 41 ⌀	description stringlengths 3 534	creator_organization stringlengths 4 22	access stringclasses 3 values	todo bool 2 classes	release_date stringdate 2019-10-23 00:00:00 2024-12-11 00:00:00 ⌀	num_parameters float64 1 1,210B ⌀
openthaigpt/openthaigpt-1.0.0-7b-chat	OpenThaiGPT v1.0.0 (7B)	OpenThaiGPT v1.0.0 (7B)	OpenThaiGPT v1.0.0 (7B) is a Thai language chat model based on Llama 2 that has been specifically fine-tuned for Thai instructions and enhanced by incorporating over 10,000 of the most commonly used Thai words into the dictionary. ([blog post](https://openthaigpt.aieat.or.th/openthaigpt-1.0.0-less-than-8-apr-2024-greater-than))	OpenThaiGPT	open	false	2024/4/8	7,000,000,000
qwen/qwen-vl-chat	Qwen-VL Chat	Qwen-VL Chat	Chat version of Qwen-VL ([paper](https://arxiv.org/abs/2308.12966)).	Alibaba Cloud	open	false	2023/8/24	null
qwen/qwen1.5-0.5b-chat	Qwen1.5 Chat (0.5B)	Qwen1.5 Chat (0.5B)	0.5B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))	Qwen	open	false	2024/2/5	null
qwen/qwen1.5-1.8b-chat	Qwen1.5 Chat (1.8B)	Qwen1.5 Chat (1.8B)	1.8B-parameter chat version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))	Qwen	open	false	2024/2/5	null
qwen/qwen1.5-110b-chat	Qwen1.5 Chat (110B)	Qwen1.5 Chat (110B)	110B-parameter chat version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. The 110B version also includes grouped query attention (GQA). ([blog](https://qwenlm.github.io/blog/qwen1.5-110b/))	Qwen	open	false	2024/4/25	null
qwen/qwen1.5-14b	Qwen1.5 (14B)	Qwen1.5 (14B)	14B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))	Qwen	open	false	2024/2/5	null
qwen/qwen1.5-32b	Qwen1.5 (32B)	Qwen1.5 (32B)	32B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. The 32B version also includes grouped query attention (GQA). ([blog](https://qwenlm.github.io/blog/qwen1.5-32b/))	Qwen	open	false	2024/4/2	null
qwen/qwen1.5-4b-chat	Qwen1.5 Chat (4B)	Qwen1.5 Chat (4B)	4B-parameter chat version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))	Qwen	open	false	2024/2/5	null
qwen/qwen1.5-72b	Qwen1.5 (72B)	Qwen1.5 (72B)	72B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))	Qwen	open	false	2024/2/5	null
qwen/qwen1.5-72b-chat	Qwen1.5 Chat (72B)	Qwen1.5 Chat (72B)	72B-parameter chat version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))	Qwen	open	false	2024/2/5	null
qwen/qwen1.5-7b	Qwen1.5 (7B)	Qwen1.5 (7B)	7B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))	Qwen	open	false	2024/2/5	null
qwen/qwen1.5-7b-chat	Qwen1.5 Chat (7B)	Qwen1.5 Chat (7B)	7B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))	Qwen	open	false	2024/2/5	null
qwen/qwen2-72b-instruct	Qwen2 Instruct (72B)	Qwen2 Instruct (72B)	72B-parameter chat version of the large language model series, Qwen2. Qwen2 uses Group Query Attention (GQA) and has extended context length support up to 128K tokens. ([blog](https://qwenlm.github.io/blog/qwen2/))	Qwen	open	false	2024/6/7	null
qwen/qwen2.5-72b-instruct-turbo	Qwen2.5 Instruct Turbo (72B)	Qwen2.5 Instruct Turbo (72B)	Qwen2.5 Instruct Turbo (72B) was trained on 18 trillion tokens and supports 29 languages, and shows improvements over Qwen2 in knowledge, coding, mathematics, instruction following, generating long texts, and processing structure data. ([blog](https://qwenlm.github.io/blog/qwen2.5/)) Turbo is Together's cost-efficient implementation, providing fast FP8 performance while maintaining quality, closely matching FP16 reference models. ([blog](https://www.together.ai/blog/together-inference-engine-2))	Qwen	open	false	2024/9/19	null
qwen/qwen2.5-7b-instruct-turbo	Qwen2.5 Instruct Turbo (7B)	Qwen2.5 Instruct Turbo (7B)	Qwen2.5 Instruct Turbo (7B) was trained on 18 trillion tokens and supports 29 languages, and shows improvements over Qwen2 in knowledge, coding, mathematics, instruction following, generating long texts, and processing structure data. ([blog](https://qwenlm.github.io/blog/qwen2.5/)) Turbo is Together's cost-efficient implementation, providing fast FP8 performance while maintaining quality, closely matching FP16 reference models. ([blog](https://www.together.ai/blog/together-inference-engine-2))	Qwen	open	false	2024/9/19	null
sail/sailor-14b-chat	Sailor Chat (14B)	Sailor Chat (14B)	Sailor is a suite of Open Language Models tailored for South-East Asia, focusing on languages such as Indonesian, Thai, Vietnamese, Malay, and Lao. These models were continually pre-trained from Qwen1.5. ([paper](https://arxiv.org/abs/2404.03608))	SAIL	open	false	2024/4/4	14,000,000,000
sail/sailor-7b-chat	Sailor Chat (7B)	Sailor Chat (7B)	Sailor is a suite of Open Language Models tailored for South-East Asia, focusing on languages such as Indonesian, Thai, Vietnamese, Malay, and Lao. These models were continually pre-trained from Qwen1.5. ([paper](https://arxiv.org/abs/2404.03608))	SAIL	open	false	2024/4/4	7,000,000,000
sambanova/sambalingo-thai-chat	SambaLingo-Thai-Chat	SambaLingo-Thai-Chat	SambaLingo-Thai-Chat is a chat model trained using direct preference optimization on SambaLingo-Thai-Base. SambaLingo-Thai-Base adapts Llama 2 (7B) to Thai by training on 38 billion tokens from the Thai split of the Cultura-X dataset. ([paper](https://arxiv.org/abs/2404.05829))	SambaLingo	open	false	2024/4/8	7,000,000,000
sambanova/sambalingo-thai-chat-70b	SambaLingo-Thai-Chat-70B	SambaLingo-Thai-Chat-70B	SambaLingo-Thai-Chat-70B is a chat model trained using direct preference optimization on SambaLingo-Thai-Base-70B. SambaLingo-Thai-Base-70B adapts Llama 2 (7B) to Thai by training on 26 billion tokens from the Thai split of the Cultura-X dataset. ([paper](https://arxiv.org/abs/2404.05829))	SambaLingo	open	false	2024/4/8	70,000,000,000
scb10x/llama-3-typhoon-v1.5x-70b-instruct	Typhoon 1.5X instruct (70B)	Typhoon 1.5X instruct (70B)	Llama-3-Typhoon-1.5X-70B-instruct is a 70 billion parameter instruct model designed for the Thai language based on Llama 3 Instruct. It utilizes the task-arithmetic model editing technique. ([blog](https://blog.opentyphoon.ai/typhoon-1-5x-our-experiment-designed-for-application-use-cases-7b85d9e9845c))	SCB10X	open	false	2024/5/29	70,000,000,000
scb10x/llama-3-typhoon-v1.5x-8b-instruct	Typhoon 1.5X instruct (8B)	Typhoon 1.5X instruct (8B)	Llama-3-Typhoon-1.5X-8B-instruct is a 8 billion parameter instruct model designed for the Thai language based on Llama 3 Instruct. It utilizes the task-arithmetic model editing technique. ([blog](https://blog.opentyphoon.ai/typhoon-1-5x-our-experiment-designed-for-application-use-cases-7b85d9e9845c))	SCB10X	open	false	2024/5/29	8,000,000,000
scb10x/typhoon-7b	Typhoon (7B)	Typhoon (7B)	Typhoon (7B) is pretrained Thai large language model with 7 billion parameters based on Mistral 7B. ([paper](https://arxiv.org/abs/2312.13951))	SCB10X	open	false	2023/12/21	7,000,000,000
scb10x/typhoon-v1.5-72b-instruct	Typhoon v1.5 Instruct (72B)	Typhoon v1.5 Instruct (72B)	Typhoon v1.5 Instruct (72B) is a pretrained Thai large language model with 72 billion parameters based on Qwen1.5-72B. ([blog](https://blog.opentyphoon.ai/typhoon-1-5-release-a9364cb8e8d7))	SCB10X	open	false	2024/5/8	72,000,000,000
scb10x/typhoon-v1.5-8b-instruct	Typhoon v1.5 Instruct (8B)	Typhoon v1.5 Instruct (8B)	Typhoon v1.5 Instruct (8B) is a pretrained Thai large language model with 8 billion parameters based on Llama 3 8B. ([blog](https://blog.opentyphoon.ai/typhoon-1-5-release-a9364cb8e8d7))	SCB10X	open	false	2024/5/8	8,000,000,000
snorkelai/Snorkel-Mistral-PairRM-DPO	Snorkel Mistral PairRM DPO	Snorkel Mistral PairRM DPO	Snorkel Mistral PairRM DPO is a multimodal model trained on 7B parameters with a 32K token sequence length. ([blog](https://snorkelai.com/snorkel-mistral-pairrm-dpo/))	Snorkel AI	open	false	2024/4/18	7,000,000,000
snowflake/snowflake-arctic-instruct	Arctic Instruct	Arctic Instruct	Arctic combines a 10B dense transformer model with a residual 128x3.66B MoE MLP resulting in 480B total and 17B active parameters chosen using a top-2 gating.	Snowflake	open	false	2024/4/24	482,000,000,000
stabilityai/stablelm-base-alpha-3b	StableLM-Base-Alpha (3B)	null	StableLM-Base-Alpha is a suite of 3B and 7B parameter decoder-only language models pre-trained on a diverse collection of English datasets with a sequence length of 4096 to push beyond the context window limitations of existing open-source language models.	Stability AI	open	true	2023/4/20	3,000,000,000
stabilityai/stablelm-base-alpha-7b	StableLM-Base-Alpha (7B)	null	StableLM-Base-Alpha is a suite of 3B and 7B parameter decoder-only language models pre-trained on a diverse collection of English datasets with a sequence length of 4096 to push beyond the context window limitations of existing open-source language models.	Stability AI	open	true	2023/4/20	7,000,000,000
stanford/alpaca-7b	Alpaca (7B)	null	Alpaca 7B is a model fine-tuned from the LLaMA 7B model on 52K instruction-following demonstrations	Stanford	open	false	2023/3/13	7,000,000,000
teknium/OpenHermes-2-Mistral-7B	OpenHermes 2 Mistral 7B	OpenHermes 2 Mistral 7B	OpenHermes 2 Mistral 7B is a multimodal model trained on 7B parameters with a 32K token sequence length. ([blog](https://teknium.com/openhermes-2-mistral-7b/))	Teknium	open	false	2024/4/18	7,000,000,000
teknium/OpenHermes-2.5-Mistral-7B	OpenHermes 2.5 Mistral 7B	OpenHermes 2.5 Mistral 7B	OpenHermes 2.5 Mistral 7B is a multimodal model trained on 7B parameters with a 32K token sequence length. ([blog](https://teknium.com/openhermes-2-5-mistral-7b/))	Teknium	open	false	2024/4/18	7,000,000,000
tiiuae/falcon-40b	Falcon (40B)	null	Falcon-40B is a 40B parameters causal decoder-only model built by TII and trained on 1,500B tokens of RefinedWeb enhanced with curated corpora.	TII UAE	open	false	2023/5/25	40,000,000,000
tiiuae/falcon-40b-instruct	Falcon-Instruct (40B)	null	Falcon-40B-Instruct is a 40B parameters causal decoder-only model built by TII based on Falcon-7B and finetuned on a mixture of chat/instruct datasets.	TII UAE	open	false	2023/5/25	40,000,000,000
tiiuae/falcon-7b	Falcon (7B)	null	Falcon-7B is a 7B parameters causal decoder-only model built by TII and trained on 1,500B tokens of RefinedWeb enhanced with curated corpora.	TII UAE	open	false	2023/3/15	7,000,000,000
tiiuae/falcon-7b-instruct	Falcon-Instruct (7B)	null	Falcon-7B-Instruct is a 7B parameters causal decoder-only model built by TII based on Falcon-7B and finetuned on a mixture of chat/instruct datasets.	TII UAE	open	false	2023/3/15	7,000,000,000
together/bloom	BLOOM (176B)	null	BLOOM (176B parameters) is an autoregressive model trained on 46 natural languages and 13 programming languages ([paper](https://arxiv.org/pdf/2211.05100.pdf)).	BigScience	open	false	2022/6/28	176,000,000,000
together/bloomz	BLOOMZ (176B)	null	BLOOMZ (176B parameters) is BLOOM that has been fine-tuned on natural language instructions ([details](https://huggingface.co/bigscience/bloomz)).	BigScience	open	true	2022/11/3	176,000,000,000
together/cerebras-gpt-13b	Cerebras GPT (13B)	null	Cerebras GPT is a family of open compute-optimal language models scaled from 111M to 13B parameters trained on the Eleuther Pile. ([paper](https://arxiv.org/pdf/2304.03208.pdf))	Cerebras	limited	true	2023/4/6	13,000,000,000
together/cerebras-gpt-6.7b	Cerebras GPT (6.7B)	null	Cerebras GPT is a family of open compute-optimal language models scaled from 111M to 13B parameters trained on the Eleuther Pile. ([paper](https://arxiv.org/pdf/2304.03208.pdf))	Cerebras	limited	true	2023/4/6	6,700,000,000
together/codegeex	CodeGeeX (13B)	null	CodeGeeX (13B parameters) is an open dense code model trained on more than 20 programming languages on a corpus of more than 850B tokens ([blog](http://keg.cs.tsinghua.edu.cn/codegeex/)).	Tsinghua	open	true	2022/9/19	13,000,000,000
together/codegen	CodeGen (16B)	null	CodeGen (16B parameters) is an open dense code model trained for multi-turn program synthesis ([blog](https://arxiv.org/pdf/2203.13474.pdf)).	Tsinghua	open	true	2022/3/25	16,000,000,000
together/flan-t5-xxl	Flan-T5 (11B)	null	Flan-T5 (11B parameters) is T5 fine-tuned on 1.8K tasks ([paper](https://arxiv.org/pdf/2210.11416.pdf)).	Google	open	false	null	null
together/galactica-120b	Galactica (120B)	null	Galactica (120B parameters) is trained on 48 million papers, textbooks, lectures notes, compounds and proteins, scientific websites, etc. ([paper](https://galactica.org/static/paper.pdf)).	Meta	open	true	2022/11/15	120,000,000,000
together/galactica-30b	Galactica (30B)	null	Galactica (30B parameters) is trained on 48 million papers, textbooks, lectures notes, compounds and proteins, scientific websites, etc. ([paper](https://galactica.org/static/paper.pdf)).	Meta	open	true	2022/11/15	30,000,000,000
together/glm	GLM (130B)	null	GLM (130B parameters) is an open bilingual (English & Chinese) bidirectional dense model that was trained using General Language Model (GLM) procedure ([paper](https://arxiv.org/pdf/2210.02414.pdf)).	Tsinghua	open	false	2022/8/4	130,000,000,000
together/gpt-j-6b	GPT-J (6B)	null	GPT-J (6B parameters) autoregressive language model trained on The Pile ([details](https://arankomatsuzaki.wordpress.com/2021/06/04/gpt-j/)).	EleutherAI	open	false	2021/6/4	6,000,000,000
together/gpt-neox-20b	GPT-NeoX (20B)	null	GPT-NeoX (20B parameters) autoregressive language model trained on The Pile ([paper](https://arxiv.org/pdf/2204.06745.pdf)).	EleutherAI	open	false	2022/2/2	20,000,000,000
together/gpt-neoxt-chat-base-20b	GPT-NeoXT-Chat-Base (20B)	null	GPT-NeoXT-Chat-Base (20B) is fine-tuned from GPT-NeoX, serving as a base model for developing open-source chatbots.	Together	open	true	2023/3/8	20,000,000,000
together/h3-2.7b	H3 (2.7B)	null	H3 (2.7B parameters) is a decoder-only language model based on state space models ([paper](https://arxiv.org/abs/2212.14052)).	HazyResearch	open	true	2023/1/23	2,700,000,000
together/koala-13b	Koala (13B)	null	Koala (13B) is a chatbot fine-tuned from Llama (13B) on dialogue data gathered from the web. ([blog post](https://bair.berkeley.edu/blog/2023/04/03/koala/))	UC Berkeley	open	true	2022/4/3	13,000,000,000
together/opt-1.3b	OPT (1.3B)	null	Open Pre-trained Transformers (1.3B parameters) is a suite of decoder-only pre-trained transformers that are fully and responsibly shared with interested researchers ([paper](https://arxiv.org/pdf/2205.01068.pdf)).	Meta	open	false	2022/5/2	1,300,000,000
together/opt-175b	OPT (175B)	null	Open Pre-trained Transformers (175B parameters) is a suite of decoder-only pre-trained transformers that are fully and responsibly shared with interested researchers ([paper](https://arxiv.org/pdf/2205.01068.pdf)).	Meta	open	false	2022/5/2	175,000,000,000
together/opt-6.7b	OPT (6.7B)	null	Open Pre-trained Transformers (6.7B parameters) is a suite of decoder-only pre-trained transformers that are fully and responsibly shared with interested researchers ([paper](https://arxiv.org/pdf/2205.01068.pdf)).	Meta	open	false	2022/5/2	6,700,000,000
together/opt-66b	OPT (66B)	null	Open Pre-trained Transformers (66B parameters) is a suite of decoder-only pre-trained transformers that are fully and responsibly shared with interested researchers ([paper](https://arxiv.org/pdf/2205.01068.pdf)).	Meta	open	false	2022/5/2	66,000,000,000
together/opt-iml-175b	OPT-IML (175B)	null	OPT-IML (175B parameters) is a suite of decoder-only transformer LMs that are multi-task fine-tuned on 2000 datasets ([paper](https://arxiv.org/pdf/2212.12017.pdf)).	Meta	open	true	2022/12/22	175,000,000,000
together/opt-iml-30b	OPT-IML (30B)	null	OPT-IML (30B parameters) is a suite of decoder-only transformer LMs that are multi-task fine-tuned on 2000 datasets ([paper](https://arxiv.org/pdf/2212.12017.pdf)).	Meta	open	true	2022/12/22	30,000,000,000
together/redpajama-incite-base-3b-v1	RedPajama-INCITE-Base-v1 (3B)	null	RedPajama-INCITE-Base-v1 (3B parameters) is a 3 billion base model that aims to replicate the LLaMA recipe as closely as possible.	Together	open	false	2023/5/5	3,000,000,000
together/redpajama-incite-base-7b	RedPajama-INCITE-Base (7B)	null	RedPajama-INCITE-Base (7B parameters) is a 7 billion base model that aims to replicate the LLaMA recipe as closely as possible.	Together	open	true	2023/5/5	7,000,000,000
together/redpajama-incite-chat-3b-v1	RedPajama-INCITE-Chat-v1 (3B)	null	RedPajama-INCITE-Chat-v1 (3B parameters) is a model fine-tuned on OASST1 and Dolly2 to enhance chatting ability. It is built from RedPajama-INCITE-Base-v1 (3B), a 3 billion base model that aims to replicate the LLaMA recipe as closely as possible.	Together	open	true	2023/5/5	3,000,000,000
together/redpajama-incite-instruct-3b-v1	RedPajama-INCITE-Instruct-v1 (3B)	null	RedPajama-INCITE-Instruct-v1 (3B parameters) is a model fine-tuned for few-shot applications on the data of GPT-JT. It is built from RedPajama-INCITE-Base-v1 (3B), a 3 billion base model that aims to replicate the LLaMA recipe as closely as possible.	Together	open	true	2023/5/5	3,000,000,000
together/redpajama-incite-instruct-7b	RedPajama-INCITE-Instruct (7B)	null	RedPajama-INCITE-Instruct (7B parameters) is a model fine-tuned for few-shot applications on the data of GPT-JT. It is built from RedPajama-INCITE-Base (7B), a 7 billion base model that aims to replicate the LLaMA recipe as closely as possible.	Together	open	true	2023/5/5	7,000,000,000
together/t0pp	T0pp (11B)	null	T0pp (11B parameters) is an encoder-decoder model trained on a large set of different tasks specified in natural language prompts ([paper](https://arxiv.org/pdf/2110.08207.pdf)).	BigScience	open	false	2021/10/15	11,000,000,000
together/t5-11b	T5 (11B)	null	T5 (11B parameters) is an encoder-decoder model trained on a multi-task mixture, where each task is converted into a text-to-text format ([paper](https://arxiv.org/pdf/1910.10683.pdf)).	Google	open	false	2019/10/23	11,000,000,000
together/Together-gpt-JT-6B-v1	GPT-JT (6B)	null	GPT-JT (6B parameters) is a fork of GPT-J ([blog post](https://www.together.xyz/blog/releasing-v1-of-gpt-jt-powered-by-open-source-ai)).	Together	open	true	2022/11/29	6,700,000,000
together/ul2	UL2 (20B)	null	UL2 (20B parameters) is an encoder-decoder model trained on the C4 corpus. It's similar to T5 but trained with a different objective and slightly different scaling knobs ([paper](https://arxiv.org/pdf/2205.05131.pdf)).	Google	open	false	2022/5/10	20,000,000,000
together/yalm	YaLM (100B)	null	YaLM (100B parameters) is an autoregressive language model trained on English and Russian text ([GitHub](https://github.com/yandex/YaLM-100B)).	Yandex	open	false	2022/6/23	100,000,000,000
Undi95/Toppy-M-7B	Toppy M 7B	Toppy M 7B	Toppy M 7B is a multimodal model trained on 7B parameters with a 32K token sequence length. ([blog](https://undi95.com/toppy-m-7b/))	Undi95	open	false	2024/4/18	7,000,000,000
upstage/SOLAR-10.7B-Instruct-v1.0	SOLAR 10.7B Instruct v1.0	SOLAR 10.7B Instruct v1.0	SOLAR 10.7B Instruct v1.0 is a multimodal model trained on 10.7B parameters with a 32K token sequence length. ([blog](https://upstage.com/solar-10-7b-instruct-v1-0/))	Upstage	open	false	2024/4/18	10,700,000,000
upstage/solar-pro-241126	Solar Pro	Solar Pro	Solar Pro is a LLM designed for instruction-following and processing structured formats like HTML and Markdown. It supports English, Korean, and Japanese and has domain expertise in Finance, Healthcare, and Legal. ([blog](https://www.upstage.ai/blog/press/solar-pro-aws)).	Upstage	limited	false	2024/11/26	22,000,000,000
uw-madison/llava-v1.6-vicuna-13b-hf	LLaVA 1.6 (13B)	LLaVA 1.6 (13B)	LLaVa is an open-source chatbot trained by fine-tuning LlamA/Vicuna on GPT-generated multimodal instruction-following data. ([paper](https://arxiv.org/abs/2304.08485))	Microsoft	open	false	2024/1/1	13,000,000,000
WizardLM/WizardLM-13B-V1.2	WizardLM 13B V1.2	WizardLM 13B V1.2	WizardLM 13B V1.2 is a multimodal model trained on 13B parameters with a 32K token sequence length. ([blog](https://wizardlm.com/wizardlm-13b-v1-2/))	WizardLM	open	false	2024/4/18	13,000,000,000
writer/palmyra-base	Palmyra Base (5B)	null	Palmyra Base (5B)	Writer	limited	false	2022/10/13	5,000,000,000
writer/palmyra-e	Palmyra E (30B)	null	Palmyra E (30B)	Writer	limited	false	2023/3/3	30,000,000,000
writer/palmyra-instruct-30	InstructPalmyra (30B)	null	InstructPalmyra (30B parameters) is trained using reinforcement learning techniques based on feedback from humans.	Writer	limited	false	2023/2/16	30,000,000,000
writer/palmyra-large	Palmyra Large (20B)	null	Palmyra Large (20B)	Writer	limited	false	2022/12/23	20,000,000,000
writer/palmyra-vision-003	Palmyra Vision 003	Palmyra Vision 003	Palmyra Vision 003 (internal only)	Writer	limited	false	2024/5/24	5,000,000,000
writer/palmyra-x	Palmyra X (43B)	null	Palmyra-X (43B parameters) is trained to adhere to instructions using human feedback and utilizes a technique called multiquery attention. Furthermore, a new feature called 'self-instruct' has been introduced, which includes the implementation of an early stopping criteria specifically designed for minimal instruction tuning ([paper](https://dev.writer.com/docs/becoming-self-instruct-introducing-early-stopping-criteria-for-minimal-instruct-tuning)).	Writer	limited	false	2023/6/11	43,000,000,000
writer/palmyra-x-004	Palmyra-X-004	Palmyra-X-004	Palmyra-X-004 language model with a large context window of up to 128,000 tokens that excels in processing and understanding complex tasks.	Writer	limited	false	2024/9/12	null
writer/palmyra-x-32k	Palmyra X-32K (33B)	null	Palmyra-X-32K (33B parameters) is a Transformer-based model, which is trained on large-scale pre-training data. The pre-training data types are diverse and cover a wide range of areas. These data types are used in conjunction and the alignment mechanism to extend context window.	Writer	limited	false	2023/12/1	33,000,000,000
writer/palmyra-x-v2	Palmyra X V2 (33B)	null	Palmyra-X V2 (33B parameters) is a Transformer-based model, which is trained on extremely large-scale pre-training data. The pre-training data more than 2 trillion tokens types are diverse and cover a wide range of areas, used FlashAttention-2.	Writer	limited	false	2023/12/1	33,000,000,000
writer/palmyra-x-v3	Palmyra X V3 (72B)	null	Palmyra-X V3 (72B parameters) is a Transformer-based model, which is trained on extremely large-scale pre-training data. It is trained via unsupervised learning and DPO and use multiquery attention.	Writer	limited	false	2023/12/1	72,000,000,000
writer/silk-road	Silk Road (35B)	null	Silk Road (35B)	Writer	limited	false	2023/4/13	35,000,000,000

openthaigpt/openthaigpt-1.0.0-7b-chat

OpenThaiGPT v1.0.0 (7B)

OpenThaiGPT v1.0.0 (7B) is a Thai language chat model based on Llama 2 that has been specifically fine-tuned for Thai instructions and enhanced by incorporating over 10,000 of the most commonly used Thai words into the dictionary. ([blog post](https://openthaigpt.aieat.or.th/openthaigpt-1.0.0-less-than-8-apr-2024-greater-than))

OpenThaiGPT

open

false

2024/4/8

7,000,000,000

qwen/qwen-vl-chat

Qwen-VL Chat

Chat version of Qwen-VL ([paper](https://arxiv.org/abs/2308.12966)).

Alibaba Cloud

open

false

2023/8/24

null

qwen/qwen1.5-0.5b-chat

Qwen1.5 Chat (0.5B)

0.5B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))

Qwen

open

false

2024/2/5

null

qwen/qwen1.5-1.8b-chat

Qwen1.5 Chat (1.8B)

1.8B-parameter chat version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))

Qwen

open

false

2024/2/5

null

qwen/qwen1.5-110b-chat

Qwen1.5 Chat (110B)

110B-parameter chat version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. The 110B version also includes grouped query attention (GQA). ([blog](https://qwenlm.github.io/blog/qwen1.5-110b/))

Qwen

open

false

2024/4/25

null

qwen/qwen1.5-14b

Qwen1.5 (14B)

14B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))

Qwen

open

false

2024/2/5

null

qwen/qwen1.5-32b

Qwen1.5 (32B)

32B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. The 32B version also includes grouped query attention (GQA). ([blog](https://qwenlm.github.io/blog/qwen1.5-32b/))

Qwen

open

false

2024/4/2

null

qwen/qwen1.5-4b-chat

Qwen1.5 Chat (4B)

4B-parameter chat version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))

Qwen

open

false

2024/2/5

null

qwen/qwen1.5-72b

Qwen1.5 (72B)

72B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))

Qwen

open

false

2024/2/5

null

qwen/qwen1.5-72b-chat

Qwen1.5 Chat (72B)

72B-parameter chat version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))

Qwen

open

false

2024/2/5

null

qwen/qwen1.5-7b

Qwen1.5 (7B)

7B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))

Qwen

open

false

2024/2/5

null

qwen/qwen1.5-7b-chat

Qwen1.5 Chat (7B)

7B-parameter version of the large language model series, Qwen 1.5 (abbr. Tongyi Qianwen), proposed by Aibaba Cloud. Qwen is a family of transformer models with SwiGLU activation, RoPE, and multi-head attention. ([blog](https://qwenlm.github.io/blog/qwen1.5/))

Qwen

open

false

2024/2/5

null

qwen/qwen2-72b-instruct

Qwen2 Instruct (72B)

72B-parameter chat version of the large language model series, Qwen2. Qwen2 uses Group Query Attention (GQA) and has extended context length support up to 128K tokens. ([blog](https://qwenlm.github.io/blog/qwen2/))

Qwen

open

false

2024/6/7

null

qwen/qwen2.5-72b-instruct-turbo

Qwen2.5 Instruct Turbo (72B)

Qwen2.5 Instruct Turbo (72B) was trained on 18 trillion tokens and supports 29 languages, and shows improvements over Qwen2 in knowledge, coding, mathematics, instruction following, generating long texts, and processing structure data. ([blog](https://qwenlm.github.io/blog/qwen2.5/)) Turbo is Together's cost-efficient implementation, providing fast FP8 performance while maintaining quality, closely matching FP16 reference models. ([blog](https://www.together.ai/blog/together-inference-engine-2))

Qwen

open

false

2024/9/19

null

qwen/qwen2.5-7b-instruct-turbo

Qwen2.5 Instruct Turbo (7B)

Qwen2.5 Instruct Turbo (7B) was trained on 18 trillion tokens and supports 29 languages, and shows improvements over Qwen2 in knowledge, coding, mathematics, instruction following, generating long texts, and processing structure data. ([blog](https://qwenlm.github.io/blog/qwen2.5/)) Turbo is Together's cost-efficient implementation, providing fast FP8 performance while maintaining quality, closely matching FP16 reference models. ([blog](https://www.together.ai/blog/together-inference-engine-2))

Qwen

open

false

2024/9/19

null

sail/sailor-14b-chat

Sailor Chat (14B)

Sailor is a suite of Open Language Models tailored for South-East Asia, focusing on languages such as Indonesian, Thai, Vietnamese, Malay, and Lao. These models were continually pre-trained from Qwen1.5. ([paper](https://arxiv.org/abs/2404.03608))

SAIL

open

false

2024/4/4

14,000,000,000

sail/sailor-7b-chat

Sailor Chat (7B)

Sailor is a suite of Open Language Models tailored for South-East Asia, focusing on languages such as Indonesian, Thai, Vietnamese, Malay, and Lao. These models were continually pre-trained from Qwen1.5. ([paper](https://arxiv.org/abs/2404.03608))

SAIL

open

false

2024/4/4

7,000,000,000

sambanova/sambalingo-thai-chat

SambaLingo-Thai-Chat

SambaLingo-Thai-Chat is a chat model trained using direct preference optimization on SambaLingo-Thai-Base. SambaLingo-Thai-Base adapts Llama 2 (7B) to Thai by training on 38 billion tokens from the Thai split of the Cultura-X dataset. ([paper](https://arxiv.org/abs/2404.05829))

SambaLingo

open

false

2024/4/8

7,000,000,000

sambanova/sambalingo-thai-chat-70b

SambaLingo-Thai-Chat-70B

SambaLingo-Thai-Chat-70B is a chat model trained using direct preference optimization on SambaLingo-Thai-Base-70B. SambaLingo-Thai-Base-70B adapts Llama 2 (7B) to Thai by training on 26 billion tokens from the Thai split of the Cultura-X dataset. ([paper](https://arxiv.org/abs/2404.05829))

SambaLingo

open

false

2024/4/8

70,000,000,000

scb10x/llama-3-typhoon-v1.5x-70b-instruct

Typhoon 1.5X instruct (70B)

Llama-3-Typhoon-1.5X-70B-instruct is a 70 billion parameter instruct model designed for the Thai language based on Llama 3 Instruct. It utilizes the task-arithmetic model editing technique. ([blog](https://blog.opentyphoon.ai/typhoon-1-5x-our-experiment-designed-for-application-use-cases-7b85d9e9845c))

SCB10X

open

false

2024/5/29

70,000,000,000

scb10x/llama-3-typhoon-v1.5x-8b-instruct

Typhoon 1.5X instruct (8B)

Llama-3-Typhoon-1.5X-8B-instruct is a 8 billion parameter instruct model designed for the Thai language based on Llama 3 Instruct. It utilizes the task-arithmetic model editing technique. ([blog](https://blog.opentyphoon.ai/typhoon-1-5x-our-experiment-designed-for-application-use-cases-7b85d9e9845c))

SCB10X

open

false

2024/5/29

8,000,000,000

scb10x/typhoon-7b

Typhoon (7B)

Typhoon (7B) is pretrained Thai large language model with 7 billion parameters based on Mistral 7B. ([paper](https://arxiv.org/abs/2312.13951))

SCB10X

open

false

2023/12/21

7,000,000,000

scb10x/typhoon-v1.5-72b-instruct

Typhoon v1.5 Instruct (72B)

Typhoon v1.5 Instruct (72B) is a pretrained Thai large language model with 72 billion parameters based on Qwen1.5-72B. ([blog](https://blog.opentyphoon.ai/typhoon-1-5-release-a9364cb8e8d7))

SCB10X

open

false

2024/5/8

72,000,000,000

scb10x/typhoon-v1.5-8b-instruct

Typhoon v1.5 Instruct (8B)

Typhoon v1.5 Instruct (8B) is a pretrained Thai large language model with 8 billion parameters based on Llama 3 8B. ([blog](https://blog.opentyphoon.ai/typhoon-1-5-release-a9364cb8e8d7))

SCB10X

open

false

2024/5/8

8,000,000,000

snorkelai/Snorkel-Mistral-PairRM-DPO

Snorkel Mistral PairRM DPO

Snorkel Mistral PairRM DPO is a multimodal model trained on 7B parameters with a 32K token sequence length. ([blog](https://snorkelai.com/snorkel-mistral-pairrm-dpo/))

Snorkel AI

open

false

2024/4/18

7,000,000,000

snowflake/snowflake-arctic-instruct

Arctic Instruct

Arctic combines a 10B dense transformer model with a residual 128x3.66B MoE MLP resulting in 480B total and 17B active parameters chosen using a top-2 gating.

Snowflake

open

false

2024/4/24

482,000,000,000

stabilityai/stablelm-base-alpha-3b

StableLM-Base-Alpha (3B)

null

StableLM-Base-Alpha is a suite of 3B and 7B parameter decoder-only language models pre-trained on a diverse collection of English datasets with a sequence length of 4096 to push beyond the context window limitations of existing open-source language models.

Stability AI

open

true

2023/4/20

3,000,000,000

stabilityai/stablelm-base-alpha-7b

StableLM-Base-Alpha (7B)

null

StableLM-Base-Alpha is a suite of 3B and 7B parameter decoder-only language models pre-trained on a diverse collection of English datasets with a sequence length of 4096 to push beyond the context window limitations of existing open-source language models.

Stability AI

open

true

2023/4/20

7,000,000,000

stanford/alpaca-7b

Alpaca (7B)

null

Alpaca 7B is a model fine-tuned from the LLaMA 7B model on 52K instruction-following demonstrations

Stanford

open

false

2023/3/13

7,000,000,000

teknium/OpenHermes-2-Mistral-7B

OpenHermes 2 Mistral 7B

OpenHermes 2 Mistral 7B is a multimodal model trained on 7B parameters with a 32K token sequence length. ([blog](https://teknium.com/openhermes-2-mistral-7b/))

Teknium

open

false

2024/4/18

7,000,000,000

teknium/OpenHermes-2.5-Mistral-7B

OpenHermes 2.5 Mistral 7B

OpenHermes 2.5 Mistral 7B is a multimodal model trained on 7B parameters with a 32K token sequence length. ([blog](https://teknium.com/openhermes-2-5-mistral-7b/))

Teknium

open

false

2024/4/18

7,000,000,000

tiiuae/falcon-40b

Falcon (40B)

null

Falcon-40B is a 40B parameters causal decoder-only model built by TII and trained on 1,500B tokens of RefinedWeb enhanced with curated corpora.

TII UAE

open

false

2023/5/25

40,000,000,000

tiiuae/falcon-40b-instruct

Falcon-Instruct (40B)

null

Falcon-40B-Instruct is a 40B parameters causal decoder-only model built by TII based on Falcon-7B and finetuned on a mixture of chat/instruct datasets.

TII UAE

open

false

2023/5/25

40,000,000,000

tiiuae/falcon-7b

Falcon (7B)

null

Falcon-7B is a 7B parameters causal decoder-only model built by TII and trained on 1,500B tokens of RefinedWeb enhanced with curated corpora.

TII UAE

open

false

2023/3/15

7,000,000,000

tiiuae/falcon-7b-instruct

Falcon-Instruct (7B)

null

Falcon-7B-Instruct is a 7B parameters causal decoder-only model built by TII based on Falcon-7B and finetuned on a mixture of chat/instruct datasets.

TII UAE

open

false

2023/3/15

7,000,000,000

together/bloom

BLOOM (176B)

null

BLOOM (176B parameters) is an autoregressive model trained on 46 natural languages and 13 programming languages ([paper](https://arxiv.org/pdf/2211.05100.pdf)).

BigScience

open

false

2022/6/28

176,000,000,000

together/bloomz

BLOOMZ (176B)

null

BLOOMZ (176B parameters) is BLOOM that has been fine-tuned on natural language instructions ([details](https://huggingface.co/bigscience/bloomz)).

BigScience

open

true

2022/11/3

176,000,000,000

together/cerebras-gpt-13b

Cerebras GPT (13B)

null

Cerebras GPT is a family of open compute-optimal language models scaled from 111M to 13B parameters trained on the Eleuther Pile. ([paper](https://arxiv.org/pdf/2304.03208.pdf))

Cerebras

limited

true

2023/4/6

13,000,000,000

together/cerebras-gpt-6.7b

Cerebras GPT (6.7B)

null

Cerebras GPT is a family of open compute-optimal language models scaled from 111M to 13B parameters trained on the Eleuther Pile. ([paper](https://arxiv.org/pdf/2304.03208.pdf))

Cerebras

limited

true

2023/4/6

6,700,000,000

together/codegeex

CodeGeeX (13B)

null

CodeGeeX (13B parameters) is an open dense code model trained on more than 20 programming languages on a corpus of more than 850B tokens ([blog](http://keg.cs.tsinghua.edu.cn/codegeex/)).

Tsinghua

open

true

2022/9/19

13,000,000,000

together/codegen

CodeGen (16B)

null

CodeGen (16B parameters) is an open dense code model trained for multi-turn program synthesis ([blog](https://arxiv.org/pdf/2203.13474.pdf)).

Tsinghua

open

true

2022/3/25

16,000,000,000

together/flan-t5-xxl

Flan-T5 (11B)

null

Flan-T5 (11B parameters) is T5 fine-tuned on 1.8K tasks ([paper](https://arxiv.org/pdf/2210.11416.pdf)).

Google

open

false

null

together/galactica-120b

Galactica (120B)

null

Galactica (120B parameters) is trained on 48 million papers, textbooks, lectures notes, compounds and proteins, scientific websites, etc. ([paper](https://galactica.org/static/paper.pdf)).

Meta

open

true

2022/11/15

120,000,000,000

together/galactica-30b

Galactica (30B)

null

Galactica (30B parameters) is trained on 48 million papers, textbooks, lectures notes, compounds and proteins, scientific websites, etc. ([paper](https://galactica.org/static/paper.pdf)).

Meta

open

true

2022/11/15

30,000,000,000

together/glm

GLM (130B)

null

GLM (130B parameters) is an open bilingual (English & Chinese) bidirectional dense model that was trained using General Language Model (GLM) procedure ([paper](https://arxiv.org/pdf/2210.02414.pdf)).

Tsinghua

open

false

2022/8/4

130,000,000,000

together/gpt-j-6b

GPT-J (6B)

null

GPT-J (6B parameters) autoregressive language model trained on The Pile ([details](https://arankomatsuzaki.wordpress.com/2021/06/04/gpt-j/)).

EleutherAI

open

false

2021/6/4

6,000,000,000

together/gpt-neox-20b

GPT-NeoX (20B)

null

GPT-NeoX (20B parameters) autoregressive language model trained on The Pile ([paper](https://arxiv.org/pdf/2204.06745.pdf)).

EleutherAI

open

false

2022/2/2

20,000,000,000

together/gpt-neoxt-chat-base-20b

GPT-NeoXT-Chat-Base (20B)

null

GPT-NeoXT-Chat-Base (20B) is fine-tuned from GPT-NeoX, serving as a base model for developing open-source chatbots.

Together

open

true

2023/3/8

20,000,000,000

together/h3-2.7b

H3 (2.7B)

null

H3 (2.7B parameters) is a decoder-only language model based on state space models ([paper](https://arxiv.org/abs/2212.14052)).

HazyResearch

open

true

2023/1/23

2,700,000,000

together/koala-13b

Koala (13B)

null

Koala (13B) is a chatbot fine-tuned from Llama (13B) on dialogue data gathered from the web. ([blog post](https://bair.berkeley.edu/blog/2023/04/03/koala/))

UC Berkeley

open

true

2022/4/3

13,000,000,000

together/opt-1.3b

OPT (1.3B)

null

Open Pre-trained Transformers (1.3B parameters) is a suite of decoder-only pre-trained transformers that are fully and responsibly shared with interested researchers ([paper](https://arxiv.org/pdf/2205.01068.pdf)).

Meta

open

false

2022/5/2

1,300,000,000

together/opt-175b

OPT (175B)

null

Open Pre-trained Transformers (175B parameters) is a suite of decoder-only pre-trained transformers that are fully and responsibly shared with interested researchers ([paper](https://arxiv.org/pdf/2205.01068.pdf)).

Meta

open

false

2022/5/2

175,000,000,000

together/opt-6.7b

OPT (6.7B)

null

Open Pre-trained Transformers (6.7B parameters) is a suite of decoder-only pre-trained transformers that are fully and responsibly shared with interested researchers ([paper](https://arxiv.org/pdf/2205.01068.pdf)).

Meta

open

false

2022/5/2

6,700,000,000

together/opt-66b

OPT (66B)

null

Open Pre-trained Transformers (66B parameters) is a suite of decoder-only pre-trained transformers that are fully and responsibly shared with interested researchers ([paper](https://arxiv.org/pdf/2205.01068.pdf)).

Meta

open

false

2022/5/2

66,000,000,000

together/opt-iml-175b

OPT-IML (175B)

null

OPT-IML (175B parameters) is a suite of decoder-only transformer LMs that are multi-task fine-tuned on 2000 datasets ([paper](https://arxiv.org/pdf/2212.12017.pdf)).

Meta

open

true

2022/12/22

175,000,000,000

together/opt-iml-30b

OPT-IML (30B)

null

OPT-IML (30B parameters) is a suite of decoder-only transformer LMs that are multi-task fine-tuned on 2000 datasets ([paper](https://arxiv.org/pdf/2212.12017.pdf)).

Meta

open

true

2022/12/22

30,000,000,000

together/redpajama-incite-base-3b-v1

RedPajama-INCITE-Base-v1 (3B)

null

RedPajama-INCITE-Base-v1 (3B parameters) is a 3 billion base model that aims to replicate the LLaMA recipe as closely as possible.

Together

open

false

2023/5/5

3,000,000,000

together/redpajama-incite-base-7b

RedPajama-INCITE-Base (7B)

null

RedPajama-INCITE-Base (7B parameters) is a 7 billion base model that aims to replicate the LLaMA recipe as closely as possible.

Together

open

true

2023/5/5

7,000,000,000

together/redpajama-incite-chat-3b-v1

RedPajama-INCITE-Chat-v1 (3B)

null

RedPajama-INCITE-Chat-v1 (3B parameters) is a model fine-tuned on OASST1 and Dolly2 to enhance chatting ability. It is built from RedPajama-INCITE-Base-v1 (3B), a 3 billion base model that aims to replicate the LLaMA recipe as closely as possible.

Together

open

true

2023/5/5

3,000,000,000

together/redpajama-incite-instruct-3b-v1

RedPajama-INCITE-Instruct-v1 (3B)

null

RedPajama-INCITE-Instruct-v1 (3B parameters) is a model fine-tuned for few-shot applications on the data of GPT-JT. It is built from RedPajama-INCITE-Base-v1 (3B), a 3 billion base model that aims to replicate the LLaMA recipe as closely as possible.

Together

open

true

2023/5/5

3,000,000,000

together/redpajama-incite-instruct-7b

RedPajama-INCITE-Instruct (7B)

null

RedPajama-INCITE-Instruct (7B parameters) is a model fine-tuned for few-shot applications on the data of GPT-JT. It is built from RedPajama-INCITE-Base (7B), a 7 billion base model that aims to replicate the LLaMA recipe as closely as possible.

Together

open

true

2023/5/5

7,000,000,000

together/t0pp

T0pp (11B)

null

T0pp (11B parameters) is an encoder-decoder model trained on a large set of different tasks specified in natural language prompts ([paper](https://arxiv.org/pdf/2110.08207.pdf)).

BigScience

open

false

2021/10/15

11,000,000,000

together/t5-11b

T5 (11B)

null

T5 (11B parameters) is an encoder-decoder model trained on a multi-task mixture, where each task is converted into a text-to-text format ([paper](https://arxiv.org/pdf/1910.10683.pdf)).

Google

open

false

2019/10/23

11,000,000,000

together/Together-gpt-JT-6B-v1

GPT-JT (6B)

null

GPT-JT (6B parameters) is a fork of GPT-J ([blog post](https://www.together.xyz/blog/releasing-v1-of-gpt-jt-powered-by-open-source-ai)).

Together

open

true

2022/11/29

6,700,000,000

together/ul2

UL2 (20B)

null

UL2 (20B parameters) is an encoder-decoder model trained on the C4 corpus. It's similar to T5 but trained with a different objective and slightly different scaling knobs ([paper](https://arxiv.org/pdf/2205.05131.pdf)).

Google

open

false

2022/5/10

20,000,000,000

together/yalm

YaLM (100B)

null

YaLM (100B parameters) is an autoregressive language model trained on English and Russian text ([GitHub](https://github.com/yandex/YaLM-100B)).

Yandex

open

false

2022/6/23

100,000,000,000

Undi95/Toppy-M-7B

Toppy M 7B

Toppy M 7B is a multimodal model trained on 7B parameters with a 32K token sequence length. ([blog](https://undi95.com/toppy-m-7b/))

Undi95

open

false

2024/4/18

7,000,000,000

upstage/SOLAR-10.7B-Instruct-v1.0

SOLAR 10.7B Instruct v1.0

SOLAR 10.7B Instruct v1.0 is a multimodal model trained on 10.7B parameters with a 32K token sequence length. ([blog](https://upstage.com/solar-10-7b-instruct-v1-0/))

Upstage

open

false

2024/4/18

10,700,000,000

upstage/solar-pro-241126

Solar Pro

Solar Pro is a LLM designed for instruction-following and processing structured formats like HTML and Markdown. It supports English, Korean, and Japanese and has domain expertise in Finance, Healthcare, and Legal. ([blog](https://www.upstage.ai/blog/press/solar-pro-aws)).

Upstage

limited

false

2024/11/26

22,000,000,000

uw-madison/llava-v1.6-vicuna-13b-hf

LLaVA 1.6 (13B)

LLaVa is an open-source chatbot trained by fine-tuning LlamA/Vicuna on GPT-generated multimodal instruction-following data. ([paper](https://arxiv.org/abs/2304.08485))

Microsoft

open

false

2024/1/1

13,000,000,000

WizardLM/WizardLM-13B-V1.2

WizardLM 13B V1.2

WizardLM 13B V1.2 is a multimodal model trained on 13B parameters with a 32K token sequence length. ([blog](https://wizardlm.com/wizardlm-13b-v1-2/))

WizardLM

open

false

2024/4/18

13,000,000,000

writer/palmyra-base

Palmyra Base (5B)

null

Palmyra Base (5B)

Writer

limited

false

2022/10/13

5,000,000,000

writer/palmyra-e

Palmyra E (30B)

null

Palmyra E (30B)

Writer

limited

false

2023/3/3

30,000,000,000

writer/palmyra-instruct-30

InstructPalmyra (30B)

null

InstructPalmyra (30B parameters) is trained using reinforcement learning techniques based on feedback from humans.

Writer

limited

false

2023/2/16

30,000,000,000

writer/palmyra-large

Palmyra Large (20B)

null

Palmyra Large (20B)

Writer

limited

false

2022/12/23

20,000,000,000

writer/palmyra-vision-003

Palmyra Vision 003

Palmyra Vision 003 (internal only)

Writer

limited

false

2024/5/24

5,000,000,000

writer/palmyra-x

Palmyra X (43B)

null

Palmyra-X (43B parameters) is trained to adhere to instructions using human feedback and utilizes a technique called multiquery attention. Furthermore, a new feature called 'self-instruct' has been introduced, which includes the implementation of an early stopping criteria specifically designed for minimal instruction tuning ([paper](https://dev.writer.com/docs/becoming-self-instruct-introducing-early-stopping-criteria-for-minimal-instruct-tuning)).

Writer

limited

false

2023/6/11

43,000,000,000

writer/palmyra-x-004

Palmyra-X-004

Palmyra-X-004 language model with a large context window of up to 128,000 tokens that excels in processing and understanding complex tasks.

Writer

limited

false

2024/9/12

null

writer/palmyra-x-32k

Palmyra X-32K (33B)

null

Palmyra-X-32K (33B parameters) is a Transformer-based model, which is trained on large-scale pre-training data. The pre-training data types are diverse and cover a wide range of areas. These data types are used in conjunction and the alignment mechanism to extend context window.

Writer

limited

false

2023/12/1

33,000,000,000

writer/palmyra-x-v2

Palmyra X V2 (33B)

null

Palmyra-X V2 (33B parameters) is a Transformer-based model, which is trained on extremely large-scale pre-training data. The pre-training data more than 2 trillion tokens types are diverse and cover a wide range of areas, used FlashAttention-2.

Writer

limited

false

2023/12/1

33,000,000,000

writer/palmyra-x-v3

Palmyra X V3 (72B)

null

Palmyra-X V3 (72B parameters) is a Transformer-based model, which is trained on extremely large-scale pre-training data. It is trained via unsupervised learning and DPO and use multiquery attention.

Writer

limited

false

2023/12/1

72,000,000,000

writer/silk-road

Silk Road (35B)

null

Silk Road (35B)

Writer

limited

false

2023/4/13

35,000,000,000