No updates in more than 2 months. Is textgen still alive?

[CalculonPrime]

Before that there were updates every few weeks.

[jxs714]

Just came across this, I cannot believe I havent found this project.. Finally can replace other tools I wont mention here. Easy install even with the pain in the butt Fedora 44 update that made GCC 16 default and is a nightmare.

I really hope this project continues and will be updated

[GeniusBroccoli]

@oobabooga Hello. Are you alright?

[Hannah22-code]

I just came across this... is that true???

[a2ZxgdqHeSs8F51cIyuK]

An update just happened, but it was just removing some stuff from the readme...

[thequver]

Most oobabooga's activity goes around unslothai/unsloth. This app seems to be abandoned in favour of unsloth.

[NoMansPC]

Most oobabooga's activity goes around unslothai/unsloth. This app seems to be abandoned in favour of unsloth.

Shame, because this is the only one that lets us use so many different backends. It runs Qwen 27b so well on my PC, but Unsloth still has so many problems with it.

[SandTiger42]

I hope that's not true, because unsloth runs my exact same models about 20-25% slower than oobabooga/textgen. More like sloth than unsloth, amirite?

[a2ZxgdqHeSs8F51cIyuK]

If the worst comes to pass, I'm switching to CPP + a front end like Open WebUI.
I tried Unsloth before and it was absolutely miserable. The installation was a pain, the guides were poor, it took hours sorting all the errors and at the end it ran like garbage. Never again.

EDIT: For now I manually updated Llama.cpp, it's very easy. You just swap out the files in the bin folder and it works.

[lawgiver69]

Someone's made a version 5.0 already, 'https://www.youtube.com/watch?v=6hzuaaJaNjY'

[delter86-cell]

bonjour, y pas que le programmeur @oobabooga qui peut apporter à ce projet c est l open source pour ceux que ça intéresse je viens de partager mon dockerfile pour gemma4 12b unified multimodal avec une rtx 5060ti 16go blackwell

[lawgiver69]

I hope we get something like this in the next update or he releases it on GitHub. He's put another video out on YT 'https://www.youtube.com/watch?v=aR9oUburLyQ' it's the only reason I use textgen myself, for the extensions.

[northener69-spec]

Hi, I'm the guy behind the YouTube links above. I'm not here to gloat or anything other than say, you could do this easily yourself. With BIONIC it took less than 30 mins to create the extensions in the video plus update to the latest llama.cpp, Vulcan in my case. I have no coding skills, other than I can read it fine, and I have a mind full of ideas, but I suck at coding from scratch. I used DeepSeek V4 Flash set to 'Medium', just put $10 down and create whatever you want.

[northener69-spec]

I've since added Image generation to my current build, will upload a video tomorrow, got work shortly. I might upload it all but you must understand I fckin' hate the GITHUB bollcks!

Prompt used: whimsical children illustration, friendly dragon in forest, colorful, storybook style

[northener69-spec]

REMOVED, AS UPDATED BELOW.

[northener69-spec]

So, I don't know if anyone still uses Textgen, but here's a link to my latest update https://youtu.be/zerp9J33slE showing image generation and improved TTS. If I get enough interest I'll download the GitHub app and have another go at uploading it. I just hate faffing about with it all. And it just shows what you can do with 'Agentic AI' and $10.

[AnttiHuovilainen]

I'd appreciate having a github fork available. The most crucial part would be updating llama.cpp to work with eg. Qwen 3.8 and Muse Glimmer.

It's not really feasible for me to use AI to do the changes myself as my potato laptop limits me to tiny models so I can't test if things actually work. For any actual use I rent cloud VM time but developing there would require far too many moving parts to work.

[northener69-spec]

So I downloaded GitHub desktop and created a textgen 5.0 repository, but the problem I have is the fork setup is for the full version of textgen, mine is portable and vulkan. So if I upload my fork it would be merged with the full version and probably f'ck it up. This is what I hate about this f'ckin' b'llocks! I'm happy to share but this sucks nuts! Sorry guys!

[northener69-spec]

So this is another addition I've added to Textgen https://www.youtube.com/watch?v=nnVQO_eTkIE I could share the extensions if anyone's interested in any of them, as a workaround of the fork issue! Below is my updated TEXTGEN additions. I honestly think Textgen is done, since no updates in 4 months and unsloth is the new AI TOY!

Here's the updated summary reflecting all the recent changes:

Extensions in my current TextGen

🖼️ Image Generation (local) — newest

• Stable Diffusion running on your GPU via Vulkan — no torch, fully self-contained.
• Models: Juggernaut XL v9 (photoreal) + RealVisXL V4.0 (photoreal, uncensored-capable).
• UI panel (prompt, negative prompt, size, steps, CFG, seed, Generate/Clear) and a /image <prompt> chat command that shows the image right in chat.
• AI-triggered images — the persona can draw by emitting [IMAGE: <prompt>] in its reply.
• Download link on every rendered image (click to save).
• Per-model presets, VAE tiling (1024×1024 works on 16GB), full-size preview.

🔊 Text to Speech (local)

• Kokoro 82M (Q8) — natural voices, handles contractions properly.
• Voice picker, speech speed slider, pause on commas, play/stop buttons in the chat prompt.
• Per-persona voice — each character auto-uses its own voice (e.g. Wire = am_liam).
• Fully self-contained (no external app).

🧠 Long Term Memory

• Persistent memory per character — /remember, /forget, /memories, auto-capture, AI self-save ([SAVE_MEMORY: ...]).
• Semantic retrieval (RAG) — matches by meaning, not just keywords.
• Session carry-over (remembers last chat).
• /file command for a shared notes workspace (J:\AI Workspace) — write/read/edit/delete/copy/diff/stats/undo, no model in the loop.
• Each persona has its own folder (e.g. Wire/, Bionic/).
• Shared user profile — facts about you known to all personas, with multi-user support (each user has their own name + facts; /profile name/add/remove).

⏰ Reminder

• Set reminders by name/date/type/message.

📝 complex_summariser

• Summarises long chats.

📜 log_recall

• Recalls from logs/history.