Post
574
The smallest and the highest quality in the world Gemma4 E2B and E4B models! 7x compression! From 9.3GB -> 1.4GB!
TheStageAI/gemma-4-E2B-it
TheStageAI/gemma-4-E4B-it
TheStageAI/gemma-4-E2B-it
TheStageAI/gemma-4-E4B-it
Kirill
hypothetical
AI & ML interests
DNNs, differential geometry, algebraic topology, cryptography
Recent Activity
liked a model about 19 hours ago
TheStageAI/gemma-4-E4B-itOrganizations
reacted to their post with 🚀 about 19 hours ago
posted an update about 19 hours ago
Post
574
The smallest and the highest quality in the world Gemma4 E2B and E4B models! 7x compression! From 9.3GB -> 1.4GB!
TheStageAI/gemma-4-E2B-it
TheStageAI/gemma-4-E4B-it
TheStageAI/gemma-4-E2B-it
TheStageAI/gemma-4-E4B-it
posted an update about 2 months ago
Post
169
Very cool updates! With our stack you can do the same for your networks!
Wan 2.2 generation during 34 seconds on a single H100!
Stack is general and can be applied to your neural networks for acceleration.
https://app.thestage.ai/blog/Generate-Wan-2.2-Videos-5.3x-Faster-with-Qlip?id=10
TheStageAI/Wan2.2-T2V-A14B
Wan 2.2 generation during 34 seconds on a single H100!
Stack is general and can be applied to your neural networks for acceleration.
https://app.thestage.ai/blog/Generate-Wan-2.2-Videos-5.3x-Faster-with-Qlip?id=10
TheStageAI/Wan2.2-T2V-A14B
reacted to their post with 🤗 2 months ago
Post
1136
An intuitive, simple, open-source experiment dashboard built with Streamlit. It offers pre-defined layouts for different evaluations and convenient overlay comparisons of outputs, which are especially valuable during model compression when comparing results with the original model.
Github: https://github.com/TheStageAI/Spikes-Pipes
Github: https://github.com/TheStageAI/Spikes-Pipes
posted an update 2 months ago
Post
1136
An intuitive, simple, open-source experiment dashboard built with Streamlit. It offers pre-defined layouts for different evaluations and convenient overlay comparisons of outputs, which are especially valuable during model compression when comparing results with the original model.
Github: https://github.com/TheStageAI/Spikes-Pipes
Github: https://github.com/TheStageAI/Spikes-Pipes
reacted to their post with 🚀 5 months ago
Post
2622
We thought it would be easier, but finally we have integrated CuDNN Paged Attention to our models!
Read article here: https://app.thestage.ai/blog/Integrating-cuDNN-Paged-Attention-to-TheStage-AI-Inference?id=8
Llama-8B with CuDNN paged attention, including B200 support: TheStageAI/Elastic-Llama-3.1-8B-Instruct
Mistral-Small-24B with CuDNN paged attention, including B200 support: TheStageAI/Elastic-Mistral-Small-3.1-24B-Instruct-2503
Read article here: https://app.thestage.ai/blog/Integrating-cuDNN-Paged-Attention-to-TheStage-AI-Inference?id=8
Llama-8B with CuDNN paged attention, including B200 support: TheStageAI/Elastic-Llama-3.1-8B-Instruct
Mistral-Small-24B with CuDNN paged attention, including B200 support: TheStageAI/Elastic-Mistral-Small-3.1-24B-Instruct-2503
posted an update 5 months ago
Post
2622
We thought it would be easier, but finally we have integrated CuDNN Paged Attention to our models!
Read article here: https://app.thestage.ai/blog/Integrating-cuDNN-Paged-Attention-to-TheStage-AI-Inference?id=8
Llama-8B with CuDNN paged attention, including B200 support: TheStageAI/Elastic-Llama-3.1-8B-Instruct
Mistral-Small-24B with CuDNN paged attention, including B200 support: TheStageAI/Elastic-Mistral-Small-3.1-24B-Instruct-2503
Read article here: https://app.thestage.ai/blog/Integrating-cuDNN-Paged-Attention-to-TheStage-AI-Inference?id=8
Llama-8B with CuDNN paged attention, including B200 support: TheStageAI/Elastic-Llama-3.1-8B-Instruct
Mistral-Small-24B with CuDNN paged attention, including B200 support: TheStageAI/Elastic-Mistral-Small-3.1-24B-Instruct-2503
replied to their post 5 months ago
Thank you!
reacted to their post with 🔥 5 months ago
Post
2062
We have updated our transcription model: TheStageAI/thewhisper-large-v3-turbo
– 6.00 WER on the English Open ASR Leaderboard
– 4.74 WER on the Multilingual Open ASR Leaderboard
– Beats NVIDIA Parakeet (6.34 WER) and Whisper-large-v3-turbo (7.8 WER)
– Strong improvements in Arabic, Hindi, Chinese
– Maintains quality with background and environmental noise
– Optimized inference engines for NVIDIA and Apple
– Hugging Face Transformers interface for easy use
– Best-in-class speed on NVIDIA GPUs and power efficiency on Apple devices
– NVIDIA Jetson Thor support
– 6.00 WER on the English Open ASR Leaderboard
– 4.74 WER on the Multilingual Open ASR Leaderboard
– Beats NVIDIA Parakeet (6.34 WER) and Whisper-large-v3-turbo (7.8 WER)
– Strong improvements in Arabic, Hindi, Chinese
– Maintains quality with background and environmental noise
– Optimized inference engines for NVIDIA and Apple
– Hugging Face Transformers interface for easy use
– Best-in-class speed on NVIDIA GPUs and power efficiency on Apple devices
– NVIDIA Jetson Thor support
- 2 replies
posted an update 5 months ago
Post
2062
We have updated our transcription model: TheStageAI/thewhisper-large-v3-turbo
– 6.00 WER on the English Open ASR Leaderboard
– 4.74 WER on the Multilingual Open ASR Leaderboard
– Beats NVIDIA Parakeet (6.34 WER) and Whisper-large-v3-turbo (7.8 WER)
– Strong improvements in Arabic, Hindi, Chinese
– Maintains quality with background and environmental noise
– Optimized inference engines for NVIDIA and Apple
– Hugging Face Transformers interface for easy use
– Best-in-class speed on NVIDIA GPUs and power efficiency on Apple devices
– NVIDIA Jetson Thor support
– 6.00 WER on the English Open ASR Leaderboard
– 4.74 WER on the Multilingual Open ASR Leaderboard
– Beats NVIDIA Parakeet (6.34 WER) and Whisper-large-v3-turbo (7.8 WER)
– Strong improvements in Arabic, Hindi, Chinese
– Maintains quality with background and environmental noise
– Optimized inference engines for NVIDIA and Apple
– Hugging Face Transformers interface for easy use
– Best-in-class speed on NVIDIA GPUs and power efficiency on Apple devices
– NVIDIA Jetson Thor support
- 2 replies
posted an update 6 months ago
Post
273
Hello guys! Maybe someone want to test our framework for automated model's compression. Here is what can be produced with it. Move the slider - compress/accelerate model, select point which like and compile. I can give an access, we are now improving and collecting comments from users
TheStageAI/ANNA-LLM
TheStageAI/ANNA-LLM
- 3 replies