Timestral
Paper • 2610.37585 • Published • 15
Social Post Explorers
community
AI & ML interests
None defined yet.
Recent Activity
View all activity
1aurent
authored a
paper 2 days ago
Post
170
SlopSoup TV. A live, never-ending 24/7 pixel-art TV network inspired by old-school late night Adult Swim. No human makes any of it.
Tune in anytime: https://huggingface.co/proxy/severian-slopsoup.hf.space or https://www.youtube.com/live/Bikb9JGy6vg?si=YPpIqLjW6y9yML6m
20+ shows in the spirit of late-night Adult Swim:
- an hourly news desk with exhausted overnight anchors
- a game show with a reigning champion
- small-claims court and a cursed cooking show
- a talk show hosted by a houseplant or a microwave
- a 1 a.m. call-in hotline
- an ’80s sitcom on a decaying VHS tape
- 4 a.m. infomercials that slide into existential dread
- wordless generative music videos
- many more shows and content added daily...
Fully AI-run: AI writes the scripts, invents new characters every episode, designs each one a unique voice, directs the camera, cuts the edit, schedules the day and runs the broadcast.
Nothing ever airs twice: a memory of everything that’s aired blocks repeated premises, jokes and characters, so every minute is new.
Real facts, delivered weird: the news, trivia and explainers are sourced and cited; the comedy is the chaos around them.
Its own engine: a custom pixel renderer with a virtual camera, lip-synced close-ups and 50+ animated sets, all running in a browser on one cloud box.
It’s weird, it’s for adults, and it never stops. Happy to answer questions about how it works.
Tune in anytime: https://huggingface.co/proxy/severian-slopsoup.hf.space or https://www.youtube.com/live/Bikb9JGy6vg?si=YPpIqLjW6y9yML6m
20+ shows in the spirit of late-night Adult Swim:
- an hourly news desk with exhausted overnight anchors
- a game show with a reigning champion
- small-claims court and a cursed cooking show
- a talk show hosted by a houseplant or a microwave
- a 1 a.m. call-in hotline
- an ’80s sitcom on a decaying VHS tape
- 4 a.m. infomercials that slide into existential dread
- wordless generative music videos
- many more shows and content added daily...
Fully AI-run: AI writes the scripts, invents new characters every episode, designs each one a unique voice, directs the camera, cuts the edit, schedules the day and runs the broadcast.
Nothing ever airs twice: a memory of everything that’s aired blocks repeated premises, jokes and characters, so every minute is new.
Real facts, delivered weird: the news, trivia and explainers are sourced and cited; the comedy is the chaos around them.
Its own engine: a custom pixel renderer with a virtual camera, lip-synced close-ups and 50+ animated sets, all running in a browser on one cloud box.
It’s weird, it’s for adults, and it never stops. Happy to answer questions about how it works.
- 2 replies
Post
385
tokenizers 0.23.3 is out!https://github.com/huggingface/tokenizers/releases/tag/v0.23.3
The release is python-only and fixes a few very annoying issues
aotrih
authored a
paper about 1 month ago
Post
141
𝗬𝘂𝗘𝟮 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝘀𝗮𝗺𝗽𝗹𝗲𝘀
Some m-a-p/YuE2-3B samples from my experiments.
All songs are: JA or EN/JA
5️⃣ 𝐎𝐑𝐈𝐆𝐈𝐍𝐀𝐋 (synth-pop)
This song started as a melody SheetSage2 cover of M83 - Midnight City.
Extended for my lyrics, and chords added.
𝘧𝘶𝘭𝘭 𝘴𝘰𝘯𝘨: https://youtu.be/9RX3OLFqV5M
1️⃣ 𝑪𝑶𝑽𝑬𝑹 : BTS - FAKE LOVE (ja version)
𝙛𝙪𝙡𝙡 𝙨𝙤𝙣𝙜: https://youtu.be/fBVmTeSScNg
cleanest sheetsage2-derived song I've gotten.
something that's quite rare about this one is that it's mixed EN and JA. (this has not worked for me in other songs)
4️⃣ 𝐎𝐑𝐈𝐆𝐈𝐍𝐀𝐋 (alt Rock)
𝘧𝘶𝘭𝘭 𝘴𝘰𝘯𝘨: https://youtu.be/D-HXWWCdqNs
Generated from style + lyrics (no user-defined ABC).
Good result, but it is afflicted with the whizz, buzzing artifacting/degeneration on the singing voice, which I get not infrequently.
the singer's pitch and timbre change in a few spots too.
2️⃣ 𝐎𝐑𝐈𝐆𝐈𝐍𝐀𝐋 (synthwave J-Pop)
𝘧𝘶𝘭𝘭 𝘴𝘰𝘯𝘨: https://youtu.be/PYGtp0VLjS8
This one was made without ABC, just letting YuE2 compose a plan before generation.
a few dropped words, odd emphases in certain places, but mostly coherent
3️⃣ 𝑪𝑶𝑽𝑬𝑹 (*attempted*) German
進撃の巨人, vogel im kamfig YouSeeBigGirl:TT
(Reference: https://youtu.be/2o5H8wIR8Go)
I didn't get a decent generation for the song.
but, there was *one* this one section of the song 👏🏻
This happens; you'll have a killer track and *suddenly* some dude starts whispering pidgin language in a weird hermaphroditic AI voice for 2-3 sec, and then the song continues fine, no more cursed artifacts 😕
or, gen is a total trainwreck, but suddenly becomes coherent and makes something cool, like this thing 😌
m-a-p/YuE2-3B
Samples from other models:
- MiniMaxAI/MiniMax-Music3 : https://huggingface.co/posts/evdcush/304645800764284
- ACE-Step/acestep-v15-xl-sft : https://huggingface.co/posts/evdcush/646521189747912
Some m-a-p/YuE2-3B samples from my experiments.
All songs are: JA or EN/JA
5️⃣ 𝐎𝐑𝐈𝐆𝐈𝐍𝐀𝐋 (synth-pop)
This song started as a melody SheetSage2 cover of M83 - Midnight City.
Extended for my lyrics, and chords added.
𝘧𝘶𝘭𝘭 𝘴𝘰𝘯𝘨: https://youtu.be/9RX3OLFqV5M
1️⃣ 𝑪𝑶𝑽𝑬𝑹 : BTS - FAKE LOVE (ja version)
𝙛𝙪𝙡𝙡 𝙨𝙤𝙣𝙜: https://youtu.be/fBVmTeSScNg
cleanest sheetsage2-derived song I've gotten.
something that's quite rare about this one is that it's mixed EN and JA. (this has not worked for me in other songs)
4️⃣ 𝐎𝐑𝐈𝐆𝐈𝐍𝐀𝐋 (alt Rock)
𝘧𝘶𝘭𝘭 𝘴𝘰𝘯𝘨: https://youtu.be/D-HXWWCdqNs
Generated from style + lyrics (no user-defined ABC).
Good result, but it is afflicted with the whizz, buzzing artifacting/degeneration on the singing voice, which I get not infrequently.
the singer's pitch and timbre change in a few spots too.
2️⃣ 𝐎𝐑𝐈𝐆𝐈𝐍𝐀𝐋 (synthwave J-Pop)
𝘧𝘶𝘭𝘭 𝘴𝘰𝘯𝘨: https://youtu.be/PYGtp0VLjS8
This one was made without ABC, just letting YuE2 compose a plan before generation.
a few dropped words, odd emphases in certain places, but mostly coherent
3️⃣ 𝑪𝑶𝑽𝑬𝑹 (*attempted*) German
進撃の巨人, vogel im kamfig YouSeeBigGirl:TT
(Reference: https://youtu.be/2o5H8wIR8Go)
I didn't get a decent generation for the song.
but, there was *one* this one section of the song 👏🏻
This happens; you'll have a killer track and *suddenly* some dude starts whispering pidgin language in a weird hermaphroditic AI voice for 2-3 sec, and then the song continues fine, no more cursed artifacts 😕
or, gen is a total trainwreck, but suddenly becomes coherent and makes something cool, like this thing 😌
m-a-p/YuE2-3B
Samples from other models:
- MiniMaxAI/MiniMax-Music3 : https://huggingface.co/posts/evdcush/304645800764284
- ACE-Step/acestep-v15-xl-sft : https://huggingface.co/posts/evdcush/646521189747912
Zhehui-Huang
authored 2
papers about 1 month ago
wenbopan
authored a
paper about 2 months ago
wenbopan
submitted a
paper to Daily Papers about 2 months ago
Post
333
I put together a Space to showcase tokenization-in-the-browser libs we maintain at HF
Let me know what you think!
sbrandeis/tokenizers-wasm-demo
Let me know what you think!
sbrandeis/tokenizers-wasm-demo
aotrih
authored a
paper about 2 months ago
mrfakename
posted an update about 2 months ago
Post
2,979
We’ve been working with LAION on a voice acting arena. You listen to two models doing the same scene and compare how well they pull it off.
It’s ready to try now - would love to hear what you think 🙂
TTS-AGI/voice-acting-arena
It’s ready to try now - would love to hear what you think 🙂
TTS-AGI/voice-acting-arena
Post
84
𝐌𝐢𝐧𝐢𝐌𝐚𝐱 𝐌𝐮𝐬𝐢𝐜 𝟑 𝐬𝐚𝐦𝐩𝐥𝐞𝐬
# ミニマックスミュージック３ サンプル
サンプルの歌は：
- https://youtu.be/krF3NGRmwGY
- https://youtu.be/LoYAkJRcuqU
もっと歌はYTに聞けます。
この音楽は
MiniMaxAI/MiniMax-Music3
のモデルで作ってしました。
歌詞は日本語。
キャプションは英語。
（英語の方がうまく行けました）
ヴォーカルの質はとてもいいと思う。
ミニマックスミュージック３の
一つの注意するところは：
このモデルが中華風の音楽にオーバーフィットされてあるかもしれません。
（二胡が勝手に入っちゃうがちがあります）
Stack:
- interface: https://github.com/evdcush/minimax-music3-jukebox
- caption writer: https://github.com/evdcush/minimax-music3-skill-erhu-purge
試してみて〜
# ミニマックスミュージック３ サンプル
サンプルの歌は：
- https://youtu.be/krF3NGRmwGY
- https://youtu.be/LoYAkJRcuqU
もっと歌はYTに聞けます。
この音楽は
MiniMaxAI/MiniMax-Music3
のモデルで作ってしました。
歌詞は日本語。
キャプションは英語。
（英語の方がうまく行けました）
ヴォーカルの質はとてもいいと思う。
ミニマックスミュージック３の
一つの注意するところは：
このモデルが中華風の音楽にオーバーフィットされてあるかもしれません。
（二胡が勝手に入っちゃうがちがあります）
Stack:
- interface: https://github.com/evdcush/minimax-music3-jukebox
- caption writer: https://github.com/evdcush/minimax-music3-skill-erhu-purge
試してみて〜
Post
101
Sample output from ACE-Step/acestep-v15-xl-sft
I had issues with the upstream ACE-Step1.5 pipeline, and seeing there were other reports of issues with the python pipeline, but not the Comfy implementation, I used the Comfy
Inference params:
Both "lyrics" and "tags" were defined, but the the vocals in the output were not properly generated (perhaps due to the low sampler cfg).
https://youtu.be/lFiOUvSCv3w
I had issues with the upstream ACE-Step1.5 pipeline, and seeing there were other reports of issues with the python pipeline, but not the Comfy implementation, I used the Comfy
audio_ace_step1_5_xl_sft workflow.Inference params:
"TextEncodeAceStepAudio1.5": {
"bpm": 96,
"duration": 260,
"timesignature": "4",
"language": "ja",
"keyscale": "F# minor",
"cfg_scale": 2,
"temperature": 0.85,
"top_p": 1,
"top_k": 0,
"min_p": 0
}
"KSampler": {
"steps": 50,
"cfg": 1,
"denoise": 1
}Both "lyrics" and "tags" were defined, but the the vocals in the output were not properly generated (perhaps due to the low sampler cfg).
https://youtu.be/lFiOUvSCv3w
question on how to join and etc
9
#49 opened 6 months ago
by 
simonko912
Request to join group
7
#55 opened 3 months ago
by 
NILKNARFGonzo
BramVanroy
posted an update 3 months ago
Post
2,892
**I benchmarked HF buckets against https access for Common Crawl.**
Took me a while to get round to do this but I benchmarked access to Common Crawl via https vs hf buckets. Both experiments were run at night in Europe. I do not think other hardware problems were impacting the speeds since CPU processing time of the non-download pipeline components were highly similar (within 2% identical) and below only the WarcReader speeds of datatrove are used.
Experiment: selected 5 disjoint samples of 64 files each (randomly from the latest crawl; 20,499 docs/file). Those five batches were then processed by 32 single-core tasks with 4GB/core (five batches to calculate CIs). Paired experiment between using https and hf bucket.
- https: 40.0 [39.3-40.6] (seconds per WARC file)
- hf bucket: 172.0 [122.7-221.2]
That is a difference of about 4x in streaming speed. You'll see that https is also more stable (smaller CI).
I also ran raw throughput tests to the endpoints to measure rate limiting (64MiB transfer at 8/32/128/256 concurrent readers) and rate limiting seems not an issue for either: at any of those parallel reader numbers, their respective speeds stay about the same.
Note that, given CC scale, this is still a small test. Rate limiting may become more obvious when processing a full crawl. I do not know whether the https endpoint vs HF bucket will shut you out earlier with which limits.
Took me a while to get round to do this but I benchmarked access to Common Crawl via https vs hf buckets. Both experiments were run at night in Europe. I do not think other hardware problems were impacting the speeds since CPU processing time of the non-download pipeline components were highly similar (within 2% identical) and below only the WarcReader speeds of datatrove are used.
Experiment: selected 5 disjoint samples of 64 files each (randomly from the latest crawl; 20,499 docs/file). Those five batches were then processed by 32 single-core tasks with 4GB/core (five batches to calculate CIs). Paired experiment between using https and hf bucket.
- https: 40.0 [39.3-40.6] (seconds per WARC file)
- hf bucket: 172.0 [122.7-221.2]
That is a difference of about 4x in streaming speed. You'll see that https is also more stable (smaller CI).
I also ran raw throughput tests to the endpoints to measure rate limiting (64MiB transfer at 8/32/128/256 concurrent readers) and rate limiting seems not an issue for either: at any of those parallel reader numbers, their respective speeds stay about the same.
Note that, given CC scale, this is still a small test. Rate limiting may become more obvious when processing a full crawl. I do not know whether the https endpoint vs HF bucket will shut you out earlier with which limits.
Question
2
#52 opened 4 months ago
by simonko912
Membership request — stevenpr (open-source local-LLM app & model author)
🔥 2
1
#51 opened 4 months ago
by 
stevenpr
felfri
authored a
paper 4 months ago