Microsoft killed die GPU-Mafia.
Microsoft hat gerade sein 1-Bit-LLM-Inference-Framework bitnet.cpp als Open Source veröffentlicht - und das ist ein ziemlicher Gamechanger.
Damit lassen sich Modelle mit bis zu 100B Parametern lokal auf der CPU ausführen - ohne GPUs.
Microsoft hat gerade sein 1-Bit-LLM-Inference-Framework bitnet.cpp als Open Source veröffentlicht - und das ist ein ziemlicher Gamechanger.
Damit lassen sich Modelle mit bis zu 100B Parametern lokal auf der CPU ausführen - ohne GPUs.
GitHub - microsoft/BitNet: Official inference framework for 1-bit LLMs
Official inference framework for 1-bit LLMs. Contribute to microsoft/BitNet development by creating an account on GitHub.
github.com
January 29, 2026 at 6:25 PM
Microsoft killed die GPU-Mafia.
Microsoft hat gerade sein 1-Bit-LLM-Inference-Framework bitnet.cpp als Open Source veröffentlicht - und das ist ein ziemlicher Gamechanger.
Damit lassen sich Modelle mit bis zu 100B Parametern lokal auf der CPU ausführen - ohne GPUs.
Microsoft hat gerade sein 1-Bit-LLM-Inference-Framework bitnet.cpp als Open Source veröffentlicht - und das ist ein ziemlicher Gamechanger.
Damit lassen sich Modelle mit bis zu 100B Parametern lokal auf der CPU ausführen - ohne GPUs.
BitNet v2, native 4-bit activations for 1-bit LLMs
BitNet v2 trained from scratch with 8-bit activations matches BitNet b1.58 while BitNet v2 achieves minimal degradation when trained with native 4-bit activations, significantly reducing memory footprint and computational cost for batched inference
BitNet v2 trained from scratch with 8-bit activations matches BitNet b1.58 while BitNet v2 achieves minimal degradation when trained with native 4-bit activations, significantly reducing memory footprint and computational cost for batched inference
April 29, 2025 at 3:27 AM
BitNet v2, native 4-bit activations for 1-bit LLMs
BitNet v2 trained from scratch with 8-bit activations matches BitNet b1.58 while BitNet v2 achieves minimal degradation when trained with native 4-bit activations, significantly reducing memory footprint and computational cost for batched inference
BitNet v2 trained from scratch with 8-bit activations matches BitNet b1.58 while BitNet v2 achieves minimal degradation when trained with native 4-bit activations, significantly reducing memory footprint and computational cost for batched inference
Microsoft's BitNet b1.58 2B4T — the first large-scale, native 1-bit LLM🚀🚀
BitNet achieves performance on par with leading full-precision LLMs — and it’s blazingly fast⚡️⚡️uses much lower memory🎉
Everything is open-sourced, per them.
BitNet achieves performance on par with leading full-precision LLMs — and it’s blazingly fast⚡️⚡️uses much lower memory🎉
Everything is open-sourced, per them.
April 16, 2025 at 4:18 AM
Microsoft's BitNet b1.58 2B4T — the first large-scale, native 1-bit LLM🚀🚀
BitNet achieves performance on par with leading full-precision LLMs — and it’s blazingly fast⚡️⚡️uses much lower memory🎉
Everything is open-sourced, per them.
BitNet achieves performance on par with leading full-precision LLMs — and it’s blazingly fast⚡️⚡️uses much lower memory🎉
Everything is open-sourced, per them.
I like that Microsoft is continuing its research on BitNet, this time on BitNet Distillation.
A lightweight pipeline that fine-tunes off-the-shelf full-precision LLMs (e.g., Qwen) into 1.58-bit precision (i.e., ternary weights {-1, 0, 1}) for specific downstream tasks,
A lightweight pipeline that fine-tunes off-the-shelf full-precision LLMs (e.g., Qwen) into 1.58-bit precision (i.e., ternary weights {-1, 0, 1}) for specific downstream tasks,
October 19, 2025 at 12:10 PM
I like that Microsoft is continuing its research on BitNet, this time on BitNet Distillation.
A lightweight pipeline that fine-tunes off-the-shelf full-precision LLMs (e.g., Qwen) into 1.58-bit precision (i.e., ternary weights {-1, 0, 1}) for specific downstream tasks,
A lightweight pipeline that fine-tunes off-the-shelf full-precision LLMs (e.g., Qwen) into 1.58-bit precision (i.e., ternary weights {-1, 0, 1}) for specific downstream tasks,
🔥 Inference weights: huggingface.co/microsoft/bi...
🔥 Training weights (bf16): huggingface.co/microsoft/bi...
🧰 Inference code: github.com/microsoft/bi...
Demo: bitnet-demo.azurewebsites.net
🔥 Training weights (bf16): huggingface.co/microsoft/bi...
🧰 Inference code: github.com/microsoft/bi...
Demo: bitnet-demo.azurewebsites.net
Bitnet
bitnet-demo.azurewebsites.net
April 16, 2025 at 4:18 AM
🔥 Inference weights: huggingface.co/microsoft/bi...
🔥 Training weights (bf16): huggingface.co/microsoft/bi...
🧰 Inference code: github.com/microsoft/bi...
Demo: bitnet-demo.azurewebsites.net
🔥 Training weights (bf16): huggingface.co/microsoft/bi...
🧰 Inference code: github.com/microsoft/bi...
Demo: bitnet-demo.azurewebsites.net
Paper: BitNet b1.58 2B4T Technical Report ( arxiv.org/abs/2504.12285 )
List of previous papers:
BitNet (b1): arxiv.org/abs/2310.11453
BitNet b1.58: arxiv.org/abs/2402.17764
BitNet b1.58 Tips and Code: github.com/microsoft/un...
BitNet a4.8: arxiv.org/abs/2411.04965
List of previous papers:
BitNet (b1): arxiv.org/abs/2310.11453
BitNet b1.58: arxiv.org/abs/2402.17764
BitNet b1.58 Tips and Code: github.com/microsoft/un...
BitNet a4.8: arxiv.org/abs/2411.04965
April 17, 2025 at 7:33 AM
Paper: BitNet b1.58 2B4T Technical Report ( arxiv.org/abs/2504.12285 )
List of previous papers:
BitNet (b1): arxiv.org/abs/2310.11453
BitNet b1.58: arxiv.org/abs/2402.17764
BitNet b1.58 Tips and Code: github.com/microsoft/un...
BitNet a4.8: arxiv.org/abs/2411.04965
List of previous papers:
BitNet (b1): arxiv.org/abs/2310.11453
BitNet b1.58: arxiv.org/abs/2402.17764
BitNet b1.58 Tips and Code: github.com/microsoft/un...
BitNet a4.8: arxiv.org/abs/2411.04965
Microsoft's BitNet a4.8, enabling 4-bit activations for 1.58-bit LLMs!
2B BitNet a4.8 trained with 2T tokens achieves 50.30% acc on MMLU, almost no degradation to BitNet b1.58.
arxiv.org/abs/2411.04965
2B BitNet a4.8 trained with 2T tokens achieves 50.30% acc on MMLU, almost no degradation to BitNet b1.58.
arxiv.org/abs/2411.04965
November 9, 2024 at 6:36 PM
Microsoft's BitNet a4.8, enabling 4-bit activations for 1.58-bit LLMs!
2B BitNet a4.8 trained with 2T tokens achieves 50.30% acc on MMLU, almost no degradation to BitNet b1.58.
arxiv.org/abs/2411.04965
2B BitNet a4.8 trained with 2T tokens achieves 50.30% acc on MMLU, almost no degradation to BitNet b1.58.
arxiv.org/abs/2411.04965
Siis jokainen jotenkin merkittävä firmahan koodasi sähköpostin lisäksi myös tekstinkäsittelysoftan työntekijöilleen jo 80-luvulla 😊
May 29, 2025 at 10:16 AM
Siis jokainen jotenkin merkittävä firmahan koodasi sähköpostin lisäksi myös tekstinkäsittelysoftan työntekijöilleen jo 80-luvulla 😊
BITNET on a university VAX
August 9, 2024 at 2:52 PM
BITNET on a university VAX
"Bitcoin -- said to be named after Bitnet, a primitive kind of networking used by IBM mainframes --"
April 9, 2026 at 5:02 AM
"Bitcoin -- said to be named after Bitnet, a primitive kind of networking used by IBM mainframes --"
hell fuckin' yeah, great hack mattbeton.com/blog/bitnet-...
Autoregressive Language Model on the 6502 Processor - Matt Beton
A language model on a 1975 MOS 6502 processor, using a tiny Mamba-based language model and custom 8-bit ternary inference engine, used to generate text on a BBC Micro from the 1980s.
mattbeton.com
August 3, 2026 at 4:00 AM
hell fuckin' yeah, great hack mattbeton.com/blog/bitnet-...
My first email address was on Bitnet.
How old are you in Internet years without telling us your age?
August 10, 2024 at 12:26 AM
My first email address was on Bitnet.
Anyone else remember BITNET?
HTTP before Host: header?
Gopher?
HTTP before Host: header?
Gopher?
March 6, 2026 at 6:21 AM
Anyone else remember BITNET?
HTTP before Host: header?
Gopher?
HTTP before Host: header?
Gopher?
We need to go back to this version of the internet
September 11, 2026 at 3:05 PM
We need to go back to this version of the internet
Age verification? DECnet-to-BITNET gateway to dual-homed BITNET/Internet shell
Age verification? I have sweaters old enough to run for president.
Age verification? My first online experience was on local BBSes with a dial-up modem.
July 11, 2025 at 4:19 PM
Age verification? DECnet-to-BITNET gateway to dual-homed BITNET/Internet shell
On en a pas mal parlé ces derniers jours, BitNet est un framework d'inférence pour des LLM avec une quantification à 1-bit, permettant de faire tourner des modèles léger en mémoire et de faire l'inférence relativement rapidement sur CPU ⬇️
github.com/microsoft/Bi...
github.com/microsoft/Bi...
GitHub - microsoft/BitNet: Official inference framework for 1-bit LLMs
Official inference framework for 1-bit LLMs. Contribute to microsoft/BitNet development by creating an account on GitHub.
github.com
April 21, 2025 at 6:13 AM
On en a pas mal parlé ces derniers jours, BitNet est un framework d'inférence pour des LLM avec une quantification à 1-bit, permettant de faire tourner des modèles léger en mémoire et de faire l'inférence relativement rapidement sur CPU ⬇️
github.com/microsoft/Bi...
github.com/microsoft/Bi...
At age 30. No Internet but we had bitnet. Super cool!
September 23, 2026 at 10:19 AM
At age 30. No Internet but we had bitnet. Super cool!
BITNET RELAY
Can I ask y’all a genuine question?
In all your life online, what’s the greatest social media you’ve ever used?
In all your life online, what’s the greatest social media you’ve ever used?
June 17, 2026 at 7:16 PM
BITNET RELAY
BITNET and EARN!
March 6, 2026 at 4:00 AM
BITNET and EARN!
Found in an old Filofax: notes on how to search Bitnet, Gopher, Listserv. Mid or early 1990s?
July 8, 2025 at 10:00 PM
Found in an old Filofax: notes on how to search Bitnet, Gopher, Listserv. Mid or early 1990s?
Microsoft finally open-sourced their 1-bit LLM inference framework called bitnet.cpp:
6.17x faster inference
82.2% less energy on CPUs
Supports Llama3, Falcon3, and BitNet models
You can now run 100B parameter models on your local CPU without GPUs.
6.17x faster inference
82.2% less energy on CPUs
Supports Llama3, Falcon3, and BitNet models
You can now run 100B parameter models on your local CPU without GPUs.
GitHub - microsoft/BitNet: Official inference framework for 1-bit LLMs
Official inference framework for 1-bit LLMs. Contribute to microsoft/BitNet development by creating an account on GitHub.
github.com
April 19, 2025 at 6:59 PM
Microsoft finally open-sourced their 1-bit LLM inference framework called bitnet.cpp:
6.17x faster inference
82.2% less energy on CPUs
Supports Llama3, Falcon3, and BitNet models
You can now run 100B parameter models on your local CPU without GPUs.
6.17x faster inference
82.2% less energy on CPUs
Supports Llama3, Falcon3, and BitNet models
You can now run 100B parameter models on your local CPU without GPUs.
Also in the Ender's Game series, though that is quaintly recognizable as a far-future extrapolation of Bitnet discussion boards
August 21, 2026 at 3:03 PM
Also in the Ender's Game series, though that is quaintly recognizable as a far-future extrapolation of Bitnet discussion boards
i don't understand why people have such a hard time with floating point precision, it's easy:
* BitNet
* mxfp4
* fp8
* bf16
if you forget, just look at the number. it means something, i think, usually
* BitNet
* mxfp4
* fp8
* bf16
if you forget, just look at the number. it means something, i think, usually
October 16, 2025 at 12:38 PM
i don't understand why people have such a hard time with floating point precision, it's easy:
* BitNet
* mxfp4
* fp8
* bf16
if you forget, just look at the number. it means something, i think, usually
* BitNet
* mxfp4
* fp8
* bf16
if you forget, just look at the number. it means something, i think, usually