Emilian Bold (@emilianbold) header
Emilian Bold (@emilianbold) profile picture

I worked on large programs before we had large language models. Now I make them run on small machines: DwarfStar, ds4j (pure Java). Join me!

Romania
Joined May 2009
Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)Photo by Emilian Bold (@emilianbold)
My DwarfStar (ds4) plugin is in the @NousResearch Hermes catalog! 🎊 Run DeepSeek, GLM and Qwen in Hermes with @antirez's ds4. Use GGUFs you already have, or let it fetch one. The model loads when you chat and unloads when idle. hermes plugins install ds4
Trying to get the DwarfStar Hermes Plugin into their catalog: github.com/NousResearch/herm… A vote there much appreciated. I find this quite handy because ds4 is started automatically. Previously I forgot to start ds4-server then just picked a cloud model. Now it just works.
1
41
Actually, it looks like it's going to take 18 days when you only have 16 cores available. Anyone have a spare 128 cores server? I need it for 2-3 days to check this Riemann Hypothesis.
You can actually just... run the Lean proof for Riemann. Only 1GB and about 1 day needed. See you tomorrow.
1
45
You can actually just... run the Lean proof for Riemann. Only 1GB and about 1 day needed. See you tomorrow.
Statement by the Lean and Mathlib communities Congratulations @__alpoge__ 🦑🦑🦑🦑🦑🦑🦑
1
73
So my broken HP printer has a 600 MHz processor and 128 MB of internal memory. Which of course are e-waste since everything is locked down, lest the owners of the printer get more out of it.
1
1
20
I would be hilarious if the 250,000 pages Lean proof for Riemann turn out to be a bug in Lean itself :D
If it turns out Anthropic did not solve the Riemann hypothesis, is that now securities fraud?
20
Tech job screening is entirely broken. If I may suggest a solution: duels. Any person voting no for a candidate is officially throwing down the gauntlet. The candidate must be allowed to accept the challenge. At the end only one has a job.
45
Trying to get the DwarfStar Hermes Plugin into their catalog: github.com/NousResearch/herm… A vote there much appreciated. I find this quite handy because ds4 is started automatically. Previously I forgot to start ds4-server then just picked a cloud model. Now it just works.
The Hermes DwarfStar plugin now lists GLM 5.3 and Qwen 3.8. I actually like GLM 5.3 a whole lot but somehow locally I was using DeepSeek 4.1, with SSD streaming, most of the time. I'll switch to GLM next week and see how it feels. Interestingly, Qwen has an odd license.
1
76
Starting to like my Hermes Desktop setup: - I sandbox agents in Apple Container github.com/emilianbold/herme… and I need to start using profiles so some agents have GPU access - I run Local AI via DwarfStar engine github.com/emilianbold/herme… Local DeepSeek imported my OpenCode sessions.
1
51
Here is something AI grew: a Hermes plugin for Apple Container. Now you can have each Hermes Agent session in individual Apple Container instances! Just share which folders are read-write or read-only and you have a nice sandbox. github.com/emilianbold/herme…
1
2
66
The Hermes settings dropdown that gaslit me: refused to list my Apple Container plugin backend, then showed it checked after I picked it elsewhere. It was a hardcoded list, fixed upstream: github.com/NousResearch/herm…
1
57
So Fable just goes into babysitter mode if you ask too many questions. At that point might as well switch model since the session is toast. I asked it to think of a naming for some architectural piece. It proceeded to rename "API key", "calls" and everything to simpleton names.
1
58
Moore's Law could have ended decades ago and we could still mine the software problem until the end of time. There are more possible strings of 32 bytes than atoms in the universe. A billion years from now there will still be undiscovered 1000 line programs with revolutionary value. We often like to think we're limited by hardware, but only because that limit is easy to quantify. It's far harder to quantify what's in the infinite unmapped regions of the Library of Babel, so we vastly underestimate it. Our biggest limits all lie in our knowledge and imagination of software and information.
Fable 5.2 wrote the transformer. Opus 5.5 generated the training data. The result: a 367K parameter language model that fits on a floppy disk and runs on a Pentium. AI is now building AI for hardware from 1995.
Readers added context they thought people might want to know
indian engagement account making it seem like this is his when he just took someone else’s post from reddit without credit (for x payout money) reddit.com/r/ClaudeAI/s/m…
7
60
3
1,057
62,995
heres one where the CEO wouldnt rather die than launch it
13
26
5
955
45,625
Emilian Bold retweeted
Sometimes AI makes me feel like Ulysses: the cost of starting a side project is so low that is like hearing the siren song all the time. I'd be working on my task and I get all these inspirations for other things that are so easy to side quest on. It does deplete my willpower at times, but the best way I’ve found to "tie myself to the mast" is to keep my physical notebook open and jot down these ideas. At some later time, I revisit them to see which ones survive the scrutiny of time.
15
7
59
4,824
What the Chinese don't get is the proper way to do this is public ask for sessions! Just make it a proper crowd sharing project for betterment of a free model and people will just... submit their sessions.
很奇怪的一件事 Anthropic 封死了蒸馏大门以后 中国的所有大模型直接举步维艰 Kimi k3.1 pro发版100天毫无技术飞跃 Deepseek 到现在也交不出合格的pro 版 GLM 5.3惊艳以后3个月都还是白卷 而通义千问Qwen 4 Max表现也远低预期 国内大厂最近的动作肉眼可见的变慢了 国产大模型究竟是遇到了瓶颈 还是因为别人卡脖子了
29
Emilian Bold retweeted
DOWNLOAD THESE 24 OPEN MODELS BEFORE MORE REPOS GET GATED OR TAKEN DOWN. You asked for the exact links, so I put together the full local AI pack: LLMs, uncensored quants, image generation, video, voice, music and speech to text. Put them on your own disk while you still can. 2 repos in this list already block anonymous downloads behind a login gate, and Hugging Face has disabled uncensored repos with zero warning before. If a provider bans your account, raises prices or rewrites its rules next month, nobody can take away files that are already on your drive. For all 4 LLMs below I included 3 versions: original weights, standard GGUF and abliterated GGUF with refusals removed. For Qwen3.8 Flash-Next and GLM-5.3-Flash, the uncensored GGUF repos are my own quants. [LLMs: original, GGUF and uncensored] Qwen3.8-27B Original: huggingface.co/Qwen/Qwen3.8-… GGUF: huggingface.co/unsloth/Qwen3… Abliterated GGUF: huggingface.co/huihui-ai/Hui… Qwen3.8 Flash-Next Original: huggingface.co/Qwen/Qwen3.8-… GGUF: huggingface.co/unsloth/Qwen3… Abliterated GGUF (my Strata quants): huggingface.co/alesha-pro/Qw… GLM-5.3-Flash Original: huggingface.co/zai-org/GLM-5… GGUF: huggingface.co/unsloth/GLM-5… Abliterated GGUF (my Maya-M quants): huggingface.co/alesha-pro/GL… DeepSeek V4.1 Flash 1017 Original: huggingface.co/deepseek-ai/D… GGUF: huggingface.co/unsloth/DeepS… Abliterated GGUF: huggingface.co/huihui-ai/Hui… [Image generation] Qwen-Image 2.1: huggingface.co/Qwen/Qwen-Ima… MOSS-Image Base (Apache 2.0): huggingface.co/OpenMOSS-Team… FLUX.3 Klein for ComfyUI (gated, accept terms on HF first): huggingface.co/black-forest-… [Video generation] Wan 2.2 Text-to-Video A14B (Apache 2.0): huggingface.co/Wan-AI/Wan2.2… Wan 2.2 Image-to-Video A14B (Apache 2.0): huggingface.co/Wan-AI/Wan2.2… HunyuanVideo 1.5: huggingface.co/tencent/Hunyu… LTX-2.5 (gated, accept terms on HF first): huggingface.co/Lightricks/LT… [Voice and TTS] Fish Audio S2 Pro (non-commercial research license): huggingface.co/fishaudio/s2-… Qwen3-TTS 12Hz 1.7B Base (Apache 2.0): huggingface.co/Qwen/Qwen3-TT… [Music generation] ACE-Step 1.5 (MIT): huggingface.co/ACE-Step/Ace-… YuE2 ComfyUI bundle (non-commercial CC-BY-NC-4.0): huggingface.co/Comfy-Org/YuE… [Speech recognition] Whisper Large v3 Turbo (MIT): huggingface.co/openai/whispe… [How to download cleanly] Use the official Hugging Face CLI: pip install -U huggingface_hub hf download OWNER/REPO --local-dir ./MODEL Do not pull every quant inside a GGUF repo by mistake. Pick 1 quant (and all of its split shards if it is split) with a file filter: hf download OWNER/REPO --include "*Q4_K_M*" --local-dir ./MODEL For original weights and media models, save the configs, tokenizers, text encoders, VAEs, README and LICENSE files too. For the 2 gated repos, accept the terms on the model card and run hf auth login first. An API account can get cut off in 5 seconds. Weights on your own disk stay yours even if the lab changes its mind tomorrow. Save this post and go fill your drives while every link still works.
I know what you need to do right now. Go download the weights of these 4 open models and put them on your disk. Right now, even if you cannot run the big ones yet. Nobody knows where any of this goes next. Qwen3.8-27B Qwen3.8 Flash-Next GLM-5.3-Flash DeepSeek V4.1 Flash 1017 Grab the abliterated quants too while every link still works. RIGHT NOW!
132
516
43
4,440
313,379
The Hermes DwarfStar plugin now lists GLM 5.3 and Qwen 3.8. I actually like GLM 5.3 a whole lot but somehow locally I was using DeepSeek 4.1, with SSD streaming, most of the time. I'll switch to GLM next week and see how it feels. Interestingly, Qwen has an odd license.
DeepSeek V4 / V4.1 Flash, running locally inside @NousResearch's Hermes. My plugin wraps @antirez's DwarfStar: one button fetches the engine, downloads the model and runs the server. It starts on your first message and stops when idle. Needs a Mac with ~96 GB RAM.
1
2
90
Asked Claude Fable to make a plan about migrating OpenCode sessions to Hermes and it was refused by model's safety filters! First time I saw such a thing.
1
1
29
DeepSeek V4 / V4.1 Flash, running locally inside @NousResearch's Hermes. My plugin wraps @antirez's DwarfStar: one button fetches the engine, downloads the model and runs the server. It starts on your first message and stops when idle. Needs a Mac with ~96 GB RAM.
1
1
1
103
Feedback welcome, this is pretty much an AIpha release. (Got it? alpha, AIpha)
3