KL3M
Click to visit website
About
KL3M is a pioneering family of large language models distinguished by its commitment to "clean" provenance. It's trained on high-quality, ethically sourced data with clear documentation, ensuring no copyright infringements, terms of service violations, or reliance on synthetic data from other LLMs. KL3M also explicitly avoids toxic sources, making it one of the cleanest models available, a claim backed by its Fairly Trained L Certification. Early models, like kl3m-170m and kl3m-1.7b, demonstrate best-in-class perplexity on business content and exceptionally low toxicity rates. KL3M models are already being used for tasks such as drafting invoices, contracts, SEC filings, and patents. Users can further train KL3M on their own content, fine-tune it for safe conversational AI or specific tasks, and even license its vast 2.5 trillion+ token training data. Designed for accessibility, smaller models run efficiently on consumer hardware.
Platform
Task
Features
• fairly trained l certification
• multi-language support for larger models
• available as standard pytorch weights
• supports custom pretraining and fine-tuning
• low toxicity scores
• efficient performance on business/legal content
• no copyright or toxicity issues
• clean provenance training data
FAQs
What kind of hardware do I need to run KL3M?
kl3m-170 runs quickly on a MacBook Air M1, and kl3m-1.7b runs well on a $300 consumer GPU.
What architectures are your models?
Smaller KL3M models use GPT-NeoX; larger models use Mixtral Mixture-of-Experts (trained from scratch).
How can I run KL3M?
KL3M is distributed as standard PyTorch model weights. Architectures are supported for HuggingFace transformers and vllm for inference.
Which languages are supported?
`kl3m-170m` and `kl3m-1.7b` are predominantly English. Larger models include English, Spanish, French, and German.
Do you provide an API?
Not yet. The focus is on small, local LLMs for information security and accessibility, but an API is being evaluated.
Is it easy to fine-tune KL3M?
Yes, excellent results have been seen for drafting, summarization, and classification. `kl3m-170` and `kl3m-1.7b` can be fine-tuned on consumer hardware.
How many tokens do you have?
Over 2.5 trillion tokens of training data (public domain and explicitly licensed), constantly adding more.
How many tokens have your models seen?
`kl3m-170m` and `kl3m-1.7b` trained on ~350B tokens. Larger models on 500B to 1T tokens.
Do you have a conversational chat model?
Not yet. While pretraining data includes conversational sources, a model designed for standard conversational rounds has not yet been trained.
Do you have a general instruction-aligned model?
Base models support tasks like summarization/conversion. An open-ended model has not been trained. The first instruct model supports legal drafting and revision.
Pricing Plans
Open Source
Free Plan• Access to KL3M model weights
• Local deployment
• Supports custom pretraining and fine-tuning
• Fairly Trained L Certification
Job Opportunities
There are currently no job postings for this AI tool.
Ratings & Reviews
No ratings available yet. Be the first to rate this tool!
Alternatives
Google Gemma
Google Gemma is a family of cutting-edge, lightweight open language models developed by Google, available for free and optimized for various devices and platforms.
View Details
Bhabha AI
Bhabha AI is dedicated to advancing AI capabilities, specifically focusing on open-source Large Language Models and datasets for Indic languages, and making AI wisdom accessible globally.
View Details
GEITje
GEITje is an open-source Dutch language model with 7 billion parameters, created by Edwin Rijgersberg.
View Details
Typhoon
Typhoon is an open-source AI research initiative creating advanced language models optimized for the Thai language. It provides open-source models, APIs, datasets, and tools for Thai-specific AI solutions.
View DetailsFeatured Tools
GirlfriendGPT
NSFW AI chat platform with customizable characters, AI image generation, and voice chat. Explore roleplay and intimate interactions with AI companions.
View Details
xMates AI
xMates AI is a next-generation AI chat app powered by large language models, offering human-like interactions and roleplaying with customizable AI characters.
View Details
AI Song Maker
AI Song Maker is an AI music generator that helps users create songs effortlessly. Compose tracks, generate AI songs, and enjoy royalty-free music creation with ease.
View Details
Wan 2.5
Wan 2.5 is a revolutionary native multimodal video generation platform. It features synchronized A/V output, 1080p HD cinematic quality, and precision image editing.
View Details
Sora 2 AI
Sora 2 AI is the next generation AI video generator, creating more realistic, controllable, and immersive videos that understand the laws of physics.
View Details
Sora 2 AI
Sora 2 AI is OpenAI's flagship model for video and audio generation, creating physics-accurate videos with synchronized dialogue, sound effects, and music.
View Details