Karsten Kuhnke PRO
mindchain
AI & ML interests
Industry Grade Humanoid Synthetic Motion Data Generation, Mechanistic Interpretability Data Generation, Sparse Autoencoders, Edge IOT, Gemma Scope 2, RLHF, Edge AI, Alpa SIM, Alpamayo-R1, Cosmos, Isaac SIM, Isaac LAB, GR00T N1.6, Unreal Engine
Recent Activity
upvoted
a
collection
about 7 hours ago
Qwen3
updated
a collection
about 10 hours ago
Nvidia - Alpamayo-R1
upvoted
a
collection
about 10 hours ago
Physical AI
Organizations
-
nvidia/Alpamayo-R1-10B
Robotics • 11B • Updated • 15.7k • 234 -
nvidia/PhysicalAI-Autonomous-Vehicles
Updated • 153k • 650 -
nvidia/PhysicalAI-Autonomous-Vehicles-NuRec
Updated • 4.93k • 71 -
Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail
Paper • 2511.00088 • Published • 2
-
nvidia/nemotron-speech-streaming-en-0.6b
Automatic Speech Recognition • Updated • 1.07k • 257 -
nvidia/parakeet-tdt-0.6b-v3
Automatic Speech Recognition • Updated • 65.9k • 528 -
nvidia/parakeet_realtime_eou_120m-v1
Updated • 835 • 106 -
nvidia/multitalker-parakeet-streaming-0.6b-v1
Automatic Speech Recognition • Updated • 592 • 60
-
nvidia/Nemotron-AIQ-Agentic-Safety-Dataset-1.0
Viewer • Updated • 10.8k • 1.44k • 10 -
nvidia/Nemotron-Content-Safety-Reasoning-Dataset
Preview • Updated • 66 • 4 -
nvidia/Aegis-AI-Content-Safety-Dataset-2.0
Viewer • Updated • 33.4k • 3.18k • 71 -
nvidia/Nemotron-Content-Safety-Audio-Dataset
Viewer • Updated • 1.93k • 961 • 3
Inference Wrapper - Models: Root LLM (The Architect) + Python REPL (The Engine) + Sub LLMs (The Workers) Spaned by querys
: Manifold-Constrained Hyper-Connections
Mamba/Transformers Combo Hybride
Top LLMs 2025: ZAI GLM-4.7 (358B) & Moonshot Kimi-K2-Thinking. Next-gen reasoning, code, multilingual. State-of-the-art performance. Production-ready.
Custom JumpReLU Sparse Autoencoders for mechanistic interpretability. T5Gemma-2 SAEs across all layers. AI safety & interpretability research.
NVIDIA Nemotron RL datasets for AI agent training. Web search, workplace tasks, instruction following, structured outputs. RLHF & alignment research.
-
nvidia/Nemotron-RL-knowledge-web_search-mcqa
Viewer • Updated • 2.93k • 451 • 6 -
nvidia/Nemotron-RL-agent-workplace_assistant
Viewer • Updated • 1.8k • 400 • 9 -
nvidia/Nemotron-RL-instruction_following
Preview • Updated • 156 • 7 -
nvidia/Nemotron-RL-instruction_following-structured_outputs
Viewer • Updated • 9.95k • 372 • 25
Custom-trained models by mindchain: reward models & SAEs. Haddock Reward Mini (8B), function-specific SAEs. AI interpretability & alignment research.
JumpReLU SAEs for Gemma 3 interpretability. EleutherAI models, DeepSeek-R1, Pythia SAEs. Mechanistic interpretability & AI safety research.
Function calling: Google FunctionGemma-270m-IT & Mobile Actions dataset (9.65k). Efficient tool use in small LMs. AI agent development.
-
nvidia/llama-nemotron-embed-vl-1b-v2
Feature Extraction • 2B • Updated • 439 • 10 -
nvidia/llama-nemotron-embed-1b-v2
Feature Extraction • 1B • Updated • 14.9k • 28 -
nvidia/llama-embed-nemotron-8b
Feature Extraction • 8B • Updated • 343k • 114 -
nvidia/NV-Embed-v2
Feature Extraction • 8B • Updated • 31.9k • 497
-
Lightricks/LTX-2
Image-to-Video • Updated • 330k • • 712 -
Masked Audio Generation using a Single Non-Autoregressive Transformer
Paper • 2401.04577 • Published • 44 -
LTX-2: Efficient Joint Audio-Visual Foundation Model
Paper • 2601.03233 • Published • 81 -
YOLO-World: Real-Time Open-Vocabulary Object Detection
Paper • 2401.17270 • Published • 43
-
nvidia/Nemotron-Cascade-8B
Text Generation • 8B • Updated • 5.14k • 49 -
nvidia/Nemotron-Cascade-8B-Thinking
Text Generation • 8B • Updated • 1.56k • 31 -
nvidia/Nemotron-Cascade-14B-Thinking
Text Generation • 15B • Updated • 3.03k • 53 -
nvidia/Nemotron-Cascade-8B-Intermediate-ckpts
Text Generation • Updated • 10
-
google/functiongemma-270m-it
Text Generation • 0.3B • Updated • 61.2k • 780 -
litert-community/embeddinggemma-300m
Sentence Similarity • Updated • 1.66k • 31 -
google/gemma-3n-E2B-it-litert-lm
Text Generation • Updated • 19.3k • 261 -
google/gemma-3n-E4B-it-litert-lm
Text Generation • Updated • 20.6k • 276
-
nvidia/Nemotron-RL-knowledge-web_search-mcqa
Viewer • Updated • 2.93k • 451 • 6 -
nvidia/Nemotron-RL-agent-workplace_assistant
Viewer • Updated • 1.8k • 400 • 9 -
nvidia/Nemotron-RL-instruction_following
Preview • Updated • 156 • 7 -
nvidia/Nemotron-RL-instruction_following-structured_outputs
Viewer • Updated • 9.95k • 372 • 25
Self-Correcting Delta Transformer - DDL provides the Hardware mechanism (The Erazor), NL solves the software problem.
LLM as a judge
-
nvidia/Llama-3.3-Nemotron-70B-Reward-Principle
Text Generation • 71B • Updated • 62 • 6 -
nvidia/Qwen3-Nemotron-32B-GenRM-Principle
Text Generation • 33B • Updated • 864 • 11 -
nvidia/Qwen3-Nemotron-32B-RLBFF
Text Generation • 33B • Updated • 67 • 27 -
nvidia/Qwen3-Nemotron-8B-BRRM
Text Generation • Updated • 145 • 8
-
PaddlePaddle/PaddleOCR-VL
Image-Text-to-Text • 1.0B • Updated • 12.5k • 1.47k -
baidu/ERNIE-4.5-0.3B-Paddle
Text Generation • 0.4B • Updated • 92 • 19 -
baidu/ERNIE-4.5-21B-A3B-Paddle
Text Generation • 22B • Updated • 65 • 13 -
ibm-granite/granite-docling-258M
Image-Text-to-Text • 0.3B • Updated • 209k • 1.08k
Mamba/Transformers Combo Hybride
Google Gemma Scope 2: JumpReLU SAEs for Gemma 3 interpretability. 270M PT/IT, 1B PT variants. Neuronpedia integration. Mechanistic analysis.
Quantized Gemma 3 models: QAT for efficient deployment. Gemma-3-27B-IT Q4. Low memory, fast inference. Edge & production-ready LLMs.
NVIDIA Nemotron reward models: 340B, 8B BRRM, 70B/32B principle-based. RLHF training, preference learning, AI alignment research.
T5Gemma-2 encoder-decoder models: 270M, 1B, 4B sizes. Text-to-text, summarization, translation. Google's architecture for structured generation.
-
google/t5gemma-2-270m-270m
Image-Text-to-Text • 0.8B • Updated • 16.3k • 161 -
google/t5gemma-2-4b-4b
Image-Text-to-Text • 9B • Updated • 8.57k • 134 -
google/t5gemma-2-1b-1b
Image-Text-to-Text • 2B • Updated • 12.6k • 64 -
T5Gemma 2: Seeing, Reading, and Understanding Longer
Paper • 2512.14856 • Published • 1
Hybrid Mamba + Transformer architectures. NVIDIA Nemotron-3-Nano-30B-A3B (32B). BF16 & GGUF. Efficient long-context & in-context learning.
-
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
Text Generation • 32B • Updated • 299k • 551 -
bartowski/nvidia_Nemotron-3-Nano-30B-A3B-GGUF
Text Generation • 32B • Updated • 10.1k • 9 -
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16
Text Generation • 32B • Updated • 11.3k • 89 -
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8
Text Generation • 32B • Updated • 650k • • 229
-
nvidia/llama-nemotron-embed-vl-1b-v2
Feature Extraction • 2B • Updated • 439 • 10 -
nvidia/llama-nemotron-embed-1b-v2
Feature Extraction • 1B • Updated • 14.9k • 28 -
nvidia/llama-embed-nemotron-8b
Feature Extraction • 8B • Updated • 343k • 114 -
nvidia/NV-Embed-v2
Feature Extraction • 8B • Updated • 31.9k • 497
-
nvidia/Alpamayo-R1-10B
Robotics • 11B • Updated • 15.7k • 234 -
nvidia/PhysicalAI-Autonomous-Vehicles
Updated • 153k • 650 -
nvidia/PhysicalAI-Autonomous-Vehicles-NuRec
Updated • 4.93k • 71 -
Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail
Paper • 2511.00088 • Published • 2
-
Lightricks/LTX-2
Image-to-Video • Updated • 330k • • 712 -
Masked Audio Generation using a Single Non-Autoregressive Transformer
Paper • 2401.04577 • Published • 44 -
LTX-2: Efficient Joint Audio-Visual Foundation Model
Paper • 2601.03233 • Published • 81 -
YOLO-World: Real-Time Open-Vocabulary Object Detection
Paper • 2401.17270 • Published • 43
-
nvidia/nemotron-speech-streaming-en-0.6b
Automatic Speech Recognition • Updated • 1.07k • 257 -
nvidia/parakeet-tdt-0.6b-v3
Automatic Speech Recognition • Updated • 65.9k • 528 -
nvidia/parakeet_realtime_eou_120m-v1
Updated • 835 • 106 -
nvidia/multitalker-parakeet-streaming-0.6b-v1
Automatic Speech Recognition • Updated • 592 • 60
-
nvidia/Nemotron-Cascade-8B
Text Generation • 8B • Updated • 5.14k • 49 -
nvidia/Nemotron-Cascade-8B-Thinking
Text Generation • 8B • Updated • 1.56k • 31 -
nvidia/Nemotron-Cascade-14B-Thinking
Text Generation • 15B • Updated • 3.03k • 53 -
nvidia/Nemotron-Cascade-8B-Intermediate-ckpts
Text Generation • Updated • 10
-
google/functiongemma-270m-it
Text Generation • 0.3B • Updated • 61.2k • 780 -
litert-community/embeddinggemma-300m
Sentence Similarity • Updated • 1.66k • 31 -
google/gemma-3n-E2B-it-litert-lm
Text Generation • Updated • 19.3k • 261 -
google/gemma-3n-E4B-it-litert-lm
Text Generation • Updated • 20.6k • 276
-
nvidia/Nemotron-RL-knowledge-web_search-mcqa
Viewer • Updated • 2.93k • 451 • 6 -
nvidia/Nemotron-RL-agent-workplace_assistant
Viewer • Updated • 1.8k • 400 • 9 -
nvidia/Nemotron-RL-instruction_following
Preview • Updated • 156 • 7 -
nvidia/Nemotron-RL-instruction_following-structured_outputs
Viewer • Updated • 9.95k • 372 • 25
-
nvidia/Nemotron-AIQ-Agentic-Safety-Dataset-1.0
Viewer • Updated • 10.8k • 1.44k • 10 -
nvidia/Nemotron-Content-Safety-Reasoning-Dataset
Preview • Updated • 66 • 4 -
nvidia/Aegis-AI-Content-Safety-Dataset-2.0
Viewer • Updated • 33.4k • 3.18k • 71 -
nvidia/Nemotron-Content-Safety-Audio-Dataset
Viewer • Updated • 1.93k • 961 • 3
Self-Correcting Delta Transformer - DDL provides the Hardware mechanism (The Erazor), NL solves the software problem.
Inference Wrapper - Models: Root LLM (The Architect) + Python REPL (The Engine) + Sub LLMs (The Workers) Spaned by querys
LLM as a judge
-
nvidia/Llama-3.3-Nemotron-70B-Reward-Principle
Text Generation • 71B • Updated • 62 • 6 -
nvidia/Qwen3-Nemotron-32B-GenRM-Principle
Text Generation • 33B • Updated • 864 • 11 -
nvidia/Qwen3-Nemotron-32B-RLBFF
Text Generation • 33B • Updated • 67 • 27 -
nvidia/Qwen3-Nemotron-8B-BRRM
Text Generation • Updated • 145 • 8
: Manifold-Constrained Hyper-Connections
-
PaddlePaddle/PaddleOCR-VL
Image-Text-to-Text • 1.0B • Updated • 12.5k • 1.47k -
baidu/ERNIE-4.5-0.3B-Paddle
Text Generation • 0.4B • Updated • 92 • 19 -
baidu/ERNIE-4.5-21B-A3B-Paddle
Text Generation • 22B • Updated • 65 • 13 -
ibm-granite/granite-docling-258M
Image-Text-to-Text • 0.3B • Updated • 209k • 1.08k
Mamba/Transformers Combo Hybride
Mamba/Transformers Combo Hybride
Top LLMs 2025: ZAI GLM-4.7 (358B) & Moonshot Kimi-K2-Thinking. Next-gen reasoning, code, multilingual. State-of-the-art performance. Production-ready.
Google Gemma Scope 2: JumpReLU SAEs for Gemma 3 interpretability. 270M PT/IT, 1B PT variants. Neuronpedia integration. Mechanistic analysis.
Custom JumpReLU Sparse Autoencoders for mechanistic interpretability. T5Gemma-2 SAEs across all layers. AI safety & interpretability research.
Quantized Gemma 3 models: QAT for efficient deployment. Gemma-3-27B-IT Q4. Low memory, fast inference. Edge & production-ready LLMs.
NVIDIA Nemotron RL datasets for AI agent training. Web search, workplace tasks, instruction following, structured outputs. RLHF & alignment research.
-
nvidia/Nemotron-RL-knowledge-web_search-mcqa
Viewer • Updated • 2.93k • 451 • 6 -
nvidia/Nemotron-RL-agent-workplace_assistant
Viewer • Updated • 1.8k • 400 • 9 -
nvidia/Nemotron-RL-instruction_following
Preview • Updated • 156 • 7 -
nvidia/Nemotron-RL-instruction_following-structured_outputs
Viewer • Updated • 9.95k • 372 • 25
NVIDIA Nemotron reward models: 340B, 8B BRRM, 70B/32B principle-based. RLHF training, preference learning, AI alignment research.
Custom-trained models by mindchain: reward models & SAEs. Haddock Reward Mini (8B), function-specific SAEs. AI interpretability & alignment research.
T5Gemma-2 encoder-decoder models: 270M, 1B, 4B sizes. Text-to-text, summarization, translation. Google's architecture for structured generation.
-
google/t5gemma-2-270m-270m
Image-Text-to-Text • 0.8B • Updated • 16.3k • 161 -
google/t5gemma-2-4b-4b
Image-Text-to-Text • 9B • Updated • 8.57k • 134 -
google/t5gemma-2-1b-1b
Image-Text-to-Text • 2B • Updated • 12.6k • 64 -
T5Gemma 2: Seeing, Reading, and Understanding Longer
Paper • 2512.14856 • Published • 1
JumpReLU SAEs for Gemma 3 interpretability. EleutherAI models, DeepSeek-R1, Pythia SAEs. Mechanistic interpretability & AI safety research.
Hybrid Mamba + Transformer architectures. NVIDIA Nemotron-3-Nano-30B-A3B (32B). BF16 & GGUF. Efficient long-context & in-context learning.
-
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
Text Generation • 32B • Updated • 299k • 551 -
bartowski/nvidia_Nemotron-3-Nano-30B-A3B-GGUF
Text Generation • 32B • Updated • 10.1k • 9 -
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16
Text Generation • 32B • Updated • 11.3k • 89 -
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8
Text Generation • 32B • Updated • 650k • • 229
Function calling: Google FunctionGemma-270m-IT & Mobile Actions dataset (9.65k). Efficient tool use in small LMs. AI agent development.