grumpycat (Leslie-Alexandre D.)
@foobar.eagleusb.com
Engineering, science and personal bits.
https://eagleusb.github.io
܉Platform x Senior SRE x MLOps, one system at the time
https://eagleusb.github.io
܉Platform x Senior SRE x MLOps, one system at the time
Reposted by grumpycat (Leslie-Alexandre D.)
what if somebody made a code forge that does *not* charge you to use your own computer
GitHub Actions charging per build minute for *self-hosted-runners*? Shit's about to hit the fan lol
December 17, 2025 at 2:50 AM
what if somebody made a code forge that does *not* charge you to use your own computer
First time I'm paying a small fee for a frontier hosted LLM
- GLM benchmarks are good, but not as great as Gemini Pro 2 or GPT 5.2
- BUT the cost per M tokens is a LOT less
- OpenAI API compatible
- cost less than **30$** for a year
- auto-renewal disabled day 1
- GLM benchmarks are good, but not as great as Gemini Pro 2 or GPT 5.2
- BUT the cost per M tokens is a LOT less
- OpenAI API compatible
- cost less than **30$** for a year
- auto-renewal disabled day 1
December 16, 2025 at 6:20 PM
First time I'm paying a small fee for a frontier hosted LLM
- GLM benchmarks are good, but not as great as Gemini Pro 2 or GPT 5.2
- BUT the cost per M tokens is a LOT less
- OpenAI API compatible
- cost less than **30$** for a year
- auto-renewal disabled day 1
- GLM benchmarks are good, but not as great as Gemini Pro 2 or GPT 5.2
- BUT the cost per M tokens is a LOT less
- OpenAI API compatible
- cost less than **30$** for a year
- auto-renewal disabled day 1
Taking the tome to rediscover the Atmosphere ecosystem is so refreshing.
In 2025 it's hard to find a piece of Internet which is not cannibalized by corporation, marketing or bots.
In 2025 it's hard to find a piece of Internet which is not cannibalized by corporation, marketing or bots.
December 16, 2025 at 12:14 PM
Taking the tome to rediscover the Atmosphere ecosystem is so refreshing.
In 2025 it's hard to find a piece of Internet which is not cannibalized by corporation, marketing or bots.
In 2025 it's hard to find a piece of Internet which is not cannibalized by corporation, marketing or bots.
New 30B MoE model from Nvidia:
- 23 Mamba-2 and MoE layers, along with 6 Attention layers
- each MoE layer includes 128 experts plus 1 shared expert, with 5 experts activated per token
- 3.5B active parameters
huggingface.co/unsloth/Nemo...
#llm #ai
- 23 Mamba-2 and MoE layers, along with 6 Attention layers
- each MoE layer includes 128 experts plus 1 shared expert, with 5 experts activated per token
- 3.5B active parameters
huggingface.co/unsloth/Nemo...
#llm #ai
unsloth/Nemotron-3-Nano-30B-A3B-GGUF · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
December 15, 2025 at 4:50 PM
New 30B MoE model from Nvidia:
- 23 Mamba-2 and MoE layers, along with 6 Attention layers
- each MoE layer includes 128 experts plus 1 shared expert, with 5 experts activated per token
- 3.5B active parameters
huggingface.co/unsloth/Nemo...
#llm #ai
- 23 Mamba-2 and MoE layers, along with 6 Attention layers
- each MoE layer includes 128 experts plus 1 shared expert, with 5 experts activated per token
- 3.5B active parameters
huggingface.co/unsloth/Nemo...
#llm #ai
Models are now hot-swappable in llama.cpp, directly through API
huggingface.co/blog/ggml-or...
#llm #llama #ai
huggingface.co/blog/ggml-or...
#llm #llama #ai
New in llama.cpp: Model Management
A Blog post by ggml.ai on Hugging Face
huggingface.co
December 14, 2025 at 10:30 AM
Models are now hot-swappable in llama.cpp, directly through API
huggingface.co/blog/ggml-or...
#llm #llama #ai
huggingface.co/blog/ggml-or...
#llm #llama #ai
hey firefox, are you becoming agentic? /s
December 13, 2025 at 5:10 PM
hey firefox, are you becoming agentic? /s
I'll curate a semble.so collection on weekly basis, with the latest #ai stuff I found interesting across the world wide web
go.eagleusb.com/semble/3m7nz...
go.eagleusb.com/semble/3m7nz...
skynet-ai (by grumpycat (Leslie-Alexandre D.)) — Semble
View grumpycat (Leslie-Alexandre D.)'s collection on Semble
go.eagleusb.com
December 13, 2025 at 5:08 PM
I'll curate a semble.so collection on weekly basis, with the latest #ai stuff I found interesting across the world wide web
go.eagleusb.com/semble/3m7nz...
go.eagleusb.com/semble/3m7nz...
Next Nvidia's model upload turned to be a whole folder leak, oopsy
(forked at huggingface.co/xeophon/NVID..., will be eventually shutdown for sure)
(forked at huggingface.co/xeophon/NVID..., will be eventually shutdown for sure)
December 12, 2025 at 2:09 PM
Next Nvidia's model upload turned to be a whole folder leak, oopsy
(forked at huggingface.co/xeophon/NVID..., will be eventually shutdown for sure)
(forked at huggingface.co/xeophon/NVID..., will be eventually shutdown for sure)
never stop creativity
December 12, 2025 at 10:45 AM
never stop creativity
Nouveau podcast, A la french: www.youtube.com/@alafrench
Propulse par :
- Mehdi Medjaoui
- JB Kempf
- Steeve Morin
Episode #1 pas mal du tout, cote podcast tech francophone
Propulse par :
- Mehdi Medjaoui
- JB Kempf
- Steeve Morin
Episode #1 pas mal du tout, cote podcast tech francophone
A la french 💻
Prenez le temps de vous installer avec nous. A la French c'est votre Tech Show avec :
- Mehdi Medjaoui, fondateur des conférences Apidays, CEO Refeed News,
- JB Kempf, Président VideoLAN - VLC &
- S...
www.youtube.com
December 12, 2025 at 10:29 AM
Nouveau podcast, A la french: www.youtube.com/@alafrench
Propulse par :
- Mehdi Medjaoui
- JB Kempf
- Steeve Morin
Episode #1 pas mal du tout, cote podcast tech francophone
Propulse par :
- Mehdi Medjaoui
- JB Kempf
- Steeve Morin
Episode #1 pas mal du tout, cote podcast tech francophone
Reposted by grumpycat (Leslie-Alexandre D.)
OpenAI et plusieurs géants du numérique (Google, Microsoft, AWS, Anthropic…) annoncent la création de l’Agentic AI Foundation sous l’égide de la Linux Foundation.
👉 https://aaif.io/
👉 https://aaif.io/
December 11, 2025 at 10:00 AM
OpenAI et plusieurs géants du numérique (Google, Microsoft, AWS, Anthropic…) annoncent la création de l’Agentic AI Foundation sous l’égide de la Linux Foundation.
👉 https://aaif.io/
👉 https://aaif.io/
Mistral dense models optimized for size while maintaining performance (~=DeepSeek v3.2) 🇨🇵
- Devstral 2 (123b)
- Devstral 2 small (24b)
mistral.ai/news/devstra...
- Devstral 2 (123b)
- Devstral 2 small (24b)
mistral.ai/news/devstra...
Introducing: Devstral 2 and Mistral Vibe CLI. | Mistral AI
State-of-the-art, open-source agentic coding models and CLI agent.
mistral.ai
December 10, 2025 at 12:47 PM
Mistral dense models optimized for size while maintaining performance (~=DeepSeek v3.2) 🇨🇵
- Devstral 2 (123b)
- Devstral 2 small (24b)
mistral.ai/news/devstra...
- Devstral 2 (123b)
- Devstral 2 small (24b)
mistral.ai/news/devstra...
nginx 1.29.4, got HTTP/2 for backends, finally
Feature: the ngx_http_proxy_module supports HTTP/2.
nginx.org/en/CHANGES
Feature: the ngx_http_proxy_module supports HTTP/2.
nginx.org/en/CHANGES
https://nginx.org/en/CHANGES
Changes with nginx 1.29.4 09 Dec 2025
*) Feature: the ngx_http_proxy_module supports HTTP/2.
*) Feature: Encrypted ClientHello TLS extension support w...
nginx.org
December 10, 2025 at 12:34 PM
nginx 1.29.4, got HTTP/2 for backends, finally
Feature: the ngx_http_proxy_module supports HTTP/2.
nginx.org/en/CHANGES
Feature: the ngx_http_proxy_module supports HTTP/2.
nginx.org/en/CHANGES
Am I thinking about creating a sovereign CloudFlare alternative just doing what it was supposed to be, like WAF and distributed anti-DDoS **only**?
December 5, 2025 at 9:46 PM
Am I thinking about creating a sovereign CloudFlare alternative just doing what it was supposed to be, like WAF and distributed anti-DDoS **only**?
Reposted by grumpycat (Leslie-Alexandre D.)
Session persistence for terminal processes, pretty awesome (instead of tmux or screen)
github.com/neurosnap/zmx
#linux #terminal #zig
github.com/neurosnap/zmx
#linux #terminal #zig
GitHub - neurosnap/zmx: Session persistence for terminal processes
Session persistence for terminal processes. Contribute to neurosnap/zmx development by creating an account on GitHub.
github.com
December 4, 2025 at 9:41 AM
Session persistence for terminal processes, pretty awesome (instead of tmux or screen)
github.com/neurosnap/zmx
#linux #terminal #zig
github.com/neurosnap/zmx
#linux #terminal #zig
The CUDA librairies overrides (in Omni) to manage GPU memory allocation, is interesting, though the implementation isn't OSS.
blog.cloudflare.com/how-cloudfla...
blog.cloudflare.com/how-cloudfla...
How Cloudflare runs more AI models on fewer GPUs: A technical deep-dive
In order to support a growing catalog of AI models while maximizing GPU utilization, Cloudflare built an internal platform called Omni. This platform uses lightweight isolation and memory over-commitm...
blog.cloudflare.com
December 3, 2025 at 11:05 PM
The CUDA librairies overrides (in Omni) to manage GPU memory allocation, is interesting, though the implementation isn't OSS.
blog.cloudflare.com/how-cloudfla...
blog.cloudflare.com/how-cloudfla...
Reposted by grumpycat (Leslie-Alexandre D.)
I'm happy to announce that my PDS, selfhosted.social, is officially open to the public now!
December 3, 2025 at 6:10 PM
I'm happy to announce that my PDS, selfhosted.social, is officially open to the public now!
New pruned (compressed) Mistral 3 models, intended to run on smaller VRAM rigs.
Let's try the 8B, how it will compare to Qwen 🫠
mistral.ai/news/mistral-3
Let's try the 8B, how it will compare to Qwen 🫠
mistral.ai/news/mistral-3
Introducing Mistral 3 | Mistral AI
A family of frontier open-source multimodal models
mistral.ai
December 3, 2025 at 6:40 PM
New pruned (compressed) Mistral 3 models, intended to run on smaller VRAM rigs.
Let's try the 8B, how it will compare to Qwen 🫠
mistral.ai/news/mistral-3
Let's try the 8B, how it will compare to Qwen 🫠
mistral.ai/news/mistral-3
Google Antigravity
Google Antigravity - Build the new way
antigravity.google
November 20, 2025 at 8:35 PM
codewiki by Google, automated documentation generated by Gemini LLM.
I've mixed feelings about the quality of the generated output, a lot of words to say nothing valuable.
Same for deep wiki.
codewiki.google
deepwiki.com
I've mixed feelings about the quality of the generated output, a lot of words to say nothing valuable.
Same for deep wiki.
codewiki.google
deepwiki.com
Code Wiki
codewiki.google
November 16, 2025 at 1:30 PM
codewiki by Google, automated documentation generated by Gemini LLM.
I've mixed feelings about the quality of the generated output, a lot of words to say nothing valuable.
Same for deep wiki.
codewiki.google
deepwiki.com
I've mixed feelings about the quality of the generated output, a lot of words to say nothing valuable.
Same for deep wiki.
codewiki.google
deepwiki.com
flox is the best from nixpkgs without the headache: flox.dev
Flox | Your dev environment, everywhere
Manage and share development environments with all the frameworks and libraries you need, then publish artifacts anywhere. Harness the power of Nix.
flox.dev
November 3, 2025 at 4:14 PM
flox is the best from nixpkgs without the headache: flox.dev