Models
一个大型 LLM 通过将两个微调的 Llama 70B 模型合并成一个 120B 模型而创建。结合了 Xwin 和 Euryale。 致谢@chargoddard 开发了用于合并模型的框架 - mergekit。 [@Undi95](h...
Google的旗舰多模态模型，支持在文本或聊天提示中使用图像和视频，以获得文本或代码响应。 请参阅Deepmind提供的基准和提示指南。 使用Gemini需遵循Google的Gemini使用条款。 #multimodal ...
WizardLM-2 7B 是微软 AI 最新 Wizard 模型的较小变体。它是最快的，并且在性能上与现有的 10 倍大开源领先模型相当。 它是对 Mistral 7B Instruct 的微调，使用与 WizardLM-2 8x22B 相同的技术。 要了解更多 ...
Google最新的多模态模型，支持在文本或聊天提示中使用图像和视频。 针对以下语言任务进行了优化：代码生成 文本生成 文本编辑 问题解决 推荐 信息提取 数据提取或生成 AI代理使用Gemini需遵循Google的Gemin使用条款。 #multimodal ...
command-r-plus-08-2024 是 Command R+ 的更新，与之前的 Command R+ 版本相比，吞吐量提高了大约 50%，延迟降低了 25%，同时硬件占用保持不变。 在此处阅读发布帖子 here。 ...
DBRX 是由 Databricks 开发的新开源大语言模型。在 132B 的参数量下，它在语言理解、编程、数学和逻辑的标准行业基准测试中超越了现有的开源 LLM，如 Llama 2 70B 和 Mixtral-8x7b。 它采用了细粒度的专家混合（MoE）架构。任何输入上都有 36B 参数处于激活状态。它在 12T 的文本和代码数据上进行 ...
The Jamba-Instruct model, introduced by AI21 Labs, is an instruction-tuned variant of their hybrid SSM-Transformer Jamba model, specifically optimized for enterprise applications.256K Context Win...
Euryale 70B v2.1 是一个专注于创意角色扮演的模型，来自 Sao10k。更好的提示遵循性。 更好的解剖学/空间意识。 更好地适应独特和自定义的格式/回复格式。 非常有创意，很多独特的风格。 在角色扮演过程中没有限制。...
一个高性能、行业标准的 7.3B 参数模型，针对速度和上下文长度进行了优化。 Mistral 7B Instruct 有多个版本变体，这里是最新版本。 ...
Phi-3 Mini 是一个强大的 3.8B 参数模型，旨在实现高级语言理解、推理和指令跟随。通过监督微调和偏好调整进行优化，它在涉及常识、数学、逻辑推理和代码处理的任务中表现出色。 在发布时，Phi-3 Medium 在轻量级模型中展示了最先进的性能。该模型是静态的，训练于一个截止日期为 2023 年 10 月的离线数据集。 ...
Phi-3 128K Medium 是一个强大的 140 亿参数模型，旨在实现高级语言理解、推理和指令跟随。通过监督微调和偏好调整进行优化，它在涉及常识、数学、逻辑推理和代码处理的任务中表现出色。 在发布时，Phi-3 Medium 在轻量级模型中展示了最先进的性能。在 MMLU-Pro 评估中，该模型甚至接近 Llama3 70B 的性能水平。 对于 4k 上下文长度，请尝试 [Phi-3 ...
Gemini 1.5 Flash 是一个基础模型，在视觉理解、分类、摘要以及从图像、音频和视频创建内容等多种多模态任务中表现出色。它擅长处理视觉和文本输入，如照片、文档、信息图和截图。 Gemini 1.5 Flash 旨在应对高容量、高频率的任务，在这些任务中，成本和延迟至关重要。在大多数常见任务中，Flash 的质量与其他 Gemini Pro 模型相当，但成本显著降低。Flash 非常适 ...
Command 是一个遵循指令的对话模型，能够以更高的质量、更可靠地执行语言任务，并且比我们的基础生成模型具有更长的上下文。 使用此模型须遵循 Cohere 的 Acceptable Use Policy. ...
Command-R 是一个 35B 参数模型，能够以更高的质量、更可靠的方式以及更长的上下文执行对话语言任务。它可以用于复杂的工作流程，如代码生成、检索增强生成（RAG）、工具使用和代理。 阅读发布帖子 here。 使用此模型须遵循 Cohere 的 [Acceptable Use Policy](https://do ...
FREE
Qwen2 7B 是一个基于变换器的模型，在语言理解、多语言能力、编码、数学和推理方面表现出色。 它具有 SwiGLU 激活、注意力 QKV 偏置和组查询注意力。它在大量数据上进行了预训练，并进行了监督微调和直接偏好优化。 有关更多详细信息，请参见这篇 博客文章 和 [GitHub 仓库](https://git ...
Categories
Tags
- Multilingual chatbots
- Data classification
- Machine learning
- 01 ai
- Natural language processing
- Programming
- Customer support ai
- Data science
- Chatbots
- Yi large
- Knowledge retrieval
- Generative ai
- Ssm transformer
- Jamba 1 5 large
- Resource efficiency
- Ai21
- Document summarization
- Technology
- Context window
- Mamba based model
- Multilingual analysis
- 256k context window
- Jamba 1 5 mini
- Jamba
- Instruction tuning
- Enterprise optimization
- Large document processing
- Safety features
- Goliath 120b
- Model merging
- Large language model
- Mergekit
- Fine tuned llama
- Alpindale
- Prose quality
- Claude 3 alternative
- Qwen2 based
- Magnum 72b
- Roleplay data
- Roleplay
- Real time interaction
- Visual question answering
- Document analysis
- Computer vision
- Multimodal processing
- Nova lite v1
- Amazon
- Translation
- New
- Interactive chat model
- Nova micro v1
- Low latency text
- Text summarization tool
- Cost effective nlp
- Financial document processing
- Nova pro v1
- Video understanding
- Multimodal analysis
- Claude 3 quality
- Qwen 25 integration
- Magnum v4 72b
- Fine tuned model
- Anthracite org
- Prose generation
- Instant responsiveness
- Compact model
- Targeted performance
- Multimodal
- Anthropic
- Claude 3 haiku
- Deep understanding
- Complex task solving
- Claude 3 opus
- Advanced intelligence
- High fluency
- Claude 3 sonnet
- Scaled deployments
- Enterprise workloads
- Cost effective ai
- Benchmark results
- Real time moderation
- Claude 35 haiku
- Code completion
- Rapid response
- Data extraction
- Predictive analytics
- Agentic tasks
- Autonomous systems
- Autonomous coding
- Claude 35 sonnet
- Visual processing
- Data science expertise
- Hybrid reasoning
- Claude 37 sonnet
- Front end development
- Full stack updates
- Agentic workflows
- Text generation
- Baichuan3 turbo
- Baichuan
- Technologyweb
- Conversational systems
- Multilingual support
- Baichuan4
- Contextual understanding
- Conversational ai
- Ethics
- Multilingual ai
- Free
- Together
- High throughput
- Hardware efficiency
- Cohere
- Command r plus
- Low latency
- Performance upgrade
- Command r
- Complex workflows
- Conversational language tasks
- Retrieval augmented generation
- Code generation
- Instruction following
- Programmingscripting
- Long context
- Language tasks
- Command
- Openai
- Language understanding
- Databricks
- Code pre training
- Dbrx
- Mixture of experts
- Deepseek
- 685b parameter
- Deepseek chat v3 0324
- Advanced chatbot
- Multi token prediction
- Deepseek chat v3
- Hot
- Load balancing strategy
- Multi head latent attention
- Large scale inference
- Distill commercialize
- Open source reasoning
- Deepseek r1
- Mit licensed model
- Advanced distillation
- Deepseek r1 distill llama 70b
- Fine tuning results
- Benchmark performance
- Competitive language model
- Competitive ai model
- Language model distillation
- Fine tuning techniques
- Deepseek r1 distill llama 8b
- Education
- Deepseek r1 distill qwen 15b
- Performance frontier
- Benchmark surpassing
- Fine tuning efficiency
- Math optimization
- Dense model performance
- State of the art language processing
- Fine tuning capabilities
- Language benchmarking model
- Deepseek r1 distill qwen 14b
- Performance benchmarks
- Distilled large language model
- Fine tuned language model
- State of the art dense models
- Deepseek r1 distill qwen 32b
- Discount
- Voice assistants
- Eva unit 01
- Roleplay model
- Eva llama 333 70b
- Creative finetune
- Narrative generation
- Storywriting ai
- Rifxonline
- Open source
- Robotics
- Multimodal understanding
- Coding capabilities
- Robust agentic experiences
- Gemini 20 flash 001
- Complex instruction
- Faster ttft
- Gemini 20 flash exp
- Complex instruction handling
- Gemini 20 flash lite 001
- Faster token generation
- Economical machine learning
- Cost effective ai solutions
- High efficiency nlp
- Fast token generation
- Optimized performance
- Gemini 20 flash lite preview 02 05
- Token pricing
- Rate limited access
- Advanced thinking capabilities
- Experimental reasoning model
- Gemini 20 flash thinking exp 1219
- Thought process generation
- Reasoning enhancement
- Thinking process generation
- Advanced reasoning
- Reasoning capabilities
- Gemini 20 flash thinking exp
- Experimental ai model
- Multimodal application
- Gemini 20 pro exp 02 05
- Rate limited ai
- Experimental model
- Google terms compliance
- Real time processing
- Gemini flash 15 8b
- Chat transcription
- Cost effective translation
- Visual understanding
- Content generation
- Gemini flash 15
- High frequency tasks
- Text editing
- Multimodal model
- Ai agents
- Gemini pro 15
- Text code response
- Gemini pro vision
- Image video processing
- Chat prompts
- Multiturn chat
- Gemini pro
- Gemma 2 27b it
- Reasoning
- Summarization
- Question answering
- Open source nlp
- Efficient ai
- Language model
- Performance optimization
- Gemma 2 9b it
- Code chatbot
- Palm 2 codechat bison 32k
- Developer support
- Coding qa
- Programming assistant
- Vision understanding
- Interactive applications
- Coding performance
- Gpt 41 mini
- Low latency ai
- Advanced small model
- Sota intelligence
- Multimodal inputs
- Gpt 4o mini
- Text and image processing
- Gpt 4o
- Fast ai model
- Llama 2 integration
- Roleplay ai
- Extended context
- Fine tuning
- Gryphe
- Mythomax l2 13b
- Fictional narrative model
- Roleplay storytelling
- Mn mag mell r1
- Creative writing tool
- Mergekit language model
- Emotional intelligence
- Customer support
- Safety
- Inflection
- Chatbot safety
- Emotional intelligence chatbot
- Inflection 3 pi
- Roleplay scenarios
- Task optimization
- Precise guidelines
- Json output
- Inflection 3 productivity
- Multilingual
- Human evaluation performance
- Large scale language
- Meta llama
- Llama 3 70b
- Pre trained nlp
- Open source alternative
- Human evaluations performance
- Llama 3 8b
- Open source language model
- Meta pre trained model
- Model comparison
- Closed source comparison
- Human evaluations
- Pre trained model
- Meta policy
- Llama 31 405b
- Multimodal integration
- Image captioning
- Visual linguistic ai
- Llama 32 11b vision
- Dialogue summarization
- Low resource nlp
- Llama 32 1b
- Efficient language processing
- Multilingual text analysis
- Dialogue generation
- Llama 32 3b
- Complex reasoning
- Text summarization
- Multilingual language model
- Multimodal ai
- Visual reasoning
- Llama 32 90b vision
- Multilingual text generation
- Multilingual dialogue model
- Generative language processing
- Instruction tuned llm
- Llama 33 70b
- Llama guard 2 8b
- Safety classification
- Prompt response analysis
- Content moderation
- Llama 3 family
- Logical reasoning
- Code processing
- Advanced mathematics
- Phi 3 medium 128k
- Microsoft azure
- Mathematics tasks
- Phi 3 mini 128k
- Dense transformer
- High quality datasets
- Phi 35 mini 128k
- Supervised fine tuning
- Y
- L
- E
- U
- 4
- R
- K
- X
- G
- S
- M
- T
- O
- D
- I
- P
- C
- H
- A
- Q
- N
- Fast performance
- Model finetuning
- Mistral 7b
- Wizardlm 2 7b
- Model optimization
- Image understanding
- Minimax 01
- Vit mlp llm
- Text generation model
- Lightning attention
- Mistralai
- Codestral mamba
- Code reasoning model
- Transformer alternative
- Infinite sequence inference
- Large context window
- Parameter model
- Context length
- Industry standard ai
- Speed optimization
- Long context window
- Coding languages
- Mistral large
- Reasoning model
- Multilingual model
- Large context length
- 12b parameters
- Function calling
- Mistral nemo
- Mistral small
- Fast translation model
- Sentiment analysis ai
- Cost efficient nlp
- Fine tuned ai
- Large scale tasks
- Cost effective processing
- Mistral tiny
- Batch processing model
- Generative model
- Pretrained experts
- Sparse mixture of experts
- Feed forward networks
- Mixtral 8x7b
- Pixtral 12b
- October 2023 data
- Torrent release
- Mistral ai
- Image to text
- Natural image processing
- Pixtral large 2411
- Chart interpretation
- Chatbot integration
- Customer support automation
- Moonshot v1 8k
- Moonshot
- Semantic understanding
- Uncensored
- Intelligent dialogue
- Llama 3 lumimaid 70b
- Curated data
- Uncensored chatbot
- Language processing
- Improved dataset
- Finetune model
- Chat optimization
- Llama 31 lumimaid 8b
- Multi turn conversation
- Structured output
- Code generation skills
- Powerful steering capabilities
- Advanced agentic capabilities
- Nousresearch
- Hermes 3 llama 31 405b
- Hermes 3 llama 31 70b
- Roleplaying enhancement
- Rlhf language model
- Automatic alignment benchmarks
- High accuracy model
- Nvidia
- Llama 31 nemotron 70b
- Precise response generation
- Science
- Chatgpt 4o latest
- Research evaluation tool
- Rate limited chatbot
- Dynamic language model
- Parallel function calling
- Reproducible outputs
- Gpt 35 turbo
- Improved instruction following
- Json mode
- Token context window
- Low latency model
- Mmlu high scores
- Autocompletion model
- Gpt 41 nano
- Advanced instruction following
- Enterprise knowledge retrieval
- Large context reasoning
- Precise code diffs
- Gpt 41
- Rate limited
- O1 mini
- Phd level accuracy
- Stem optimization
- Advanced scientific computing
- O1 preview
- O1
- Chain of thought reasoning
- Reinforcement learning
- Stem reasoning
- Structured outputs
- Reduced errors
- O3 mini high
- Developer tools
- Conditioned reinforcement learning
- Mixed quality data
- Openchat
- Language models library
- Openchat 7b
- Offline ai model
- Cost efficient llm
- Perplexity
- High performance language model
- Llama 31 sonar large 128k chat
- Fast processing llm
- Llama 31 sonar small 128k chat
- Offline chat
- Censorship removal
- Multilingual sensitive topics
- R1 1776
- Qwen
- Multilingual understanding
- Coding and reasoning
- Swiglu activation
- Qwen 2 7b
- Transformer model
- Video question answering
- Multilingual text recognition
- Mobile robot integration
- Qwen 2 vl 72b
- Qwen 2 vl 7b
- Structured data understanding
- Long context processing
- Qwen 25 72b
- Chatbot role play
- Qwen 25 7b
- Text recognition
- Qwen vl plus
- High resolution support
- Visual recognition
- Large aspect ratio handling
- Qwen25 vl 72b
- Visual layout processing
- Image analysis
- Text and chart analysis
- Object recognition
- Ai reasoning
- Recursive loops
- Safety considerations
- Qwq 32b preview
- Language mixing
- 40off
- Unique formatting
- L3 euryale 70b
- Creative roleplay
- Prompt adherence
- Spatial awareness
- Strategic merge
- L3 lunaris 8b
- General knowledge
- Creative logic
- Roleplaying model
- Storytelling ai
- Character interaction
- L31 euryale 70b
- L33 euryale 70b
- Character simulation
- Sophosympatheia
- Natural language creativity
- Frankenmerge architecture
- Scene logic enhancement
- Roleplaying performance
- Storytelling applications
- Rogue rose 103b v02
- Multilingual embeddings
- Semantic search
- Text embedding 3 large
- Similarity matching
- Budget friendly nlp
- Text embedding 3 small
- Cost effective embeddings
- Text similarity matching
- Creative writing model
- Engaging prose
- Rocinante 12b
- Thedrummer
- Remm slerp l2 13b
- Merge model
- Recreation trial
- Updated models
- Mythomax l2 b13
- Undi95
- Task arithmetic
- Parameter blending
- Toppy m 7b
- Uncensored ai
- Visual comprehension
- X ai
- Style analysis
- Grok 2 vision