kyutai

@kyutai_labs · Paris, France · joined 12 Nov 2023

27 690Followers

14Following

243Posts total

531.5KViews on collected posts

โพสต์ล่าสุด

kyutai@kyutai_labs26 Aug

It should be soon possible to target CPUs, GPUs, TPUs, NPUs and ASICs with a single PyTorch installation, without performance impact! Gabriel de Marmiesse presented his work at Kyutai on integrating Mojo into PyTorch at ModCon last week. Links below https://t.co/0azy44q9EJ 6.5K views · 107 likes · 12 reposts · 5 replies Open on X →

kyutai@kyutai_labs26 Aug

@kyutai_labs PocketTTS --- the fastest model in the audio.cpp Arena! https://t.co/ufiRgHnPj1 397 views · 3 likes · 0 reposts · 0 replies Open on X →

kyutai@kyutai_labs25 Aug

@kyutai_labs Love Pocket, but can you train the English model to say "live" with a hard "i" when appropriate? In my experience, it incorrectly says it with a soft 'i' 90% of the time. 555 views · 2 likes · 0 reposts · 0 replies Open on X →

kyutai@kyutai_labs25 Aug

@kyutai_labs Thanks for releasing the training code it matches what i was doing. only issue was mine kept blabbering and eos token prediction had issues 638 views · 3 likes · 1 reposts · 1 replies Open on X →

kyutai@kyutai_labs25 Aug

If you're looking for a weekend project, how about training your own text-to-speech model from scratch on your own GPU, and then running it on any device's CPU? We just open-sourced the entire Pocket TTS training stack: data pipeline, recipes, and evals. It learns pretty damn h 104.4K views · 1K likes · 118 reposts · 52 replies Open on X →

kyutai@kyutai_labs20 Aug

MuScriptor, our music transcription model, can now make sheet music and tabs! We export PDFs and editable MusicXML. Under the hood, we generate these from MIDI using @musescore + some custom preprocessing. https://t.co/zWHqbR0zmJ 37.3K views · 468 likes · 53 reposts · 21 replies Open on X →

kyutai@kyutai_labs6 Aug

Our audio-to-MIDI model, MuScriptor, now also detects tempo! You can directly drag-and-drop the MIDI into a DAW and it just works, without you having to tempo-match manually. Have fun! https://t.co/zbBNtkknIK 27.6K views · 346 likes · 31 reposts · 8 replies Open on X →

kyutai@kyutai_labs11 Jul

@kyutai_labs @MireloAI Cool, you can just do things now 🤷 https://t.co/E77u6YKzEu 15K views · 63 likes · 3 reposts · 3 replies Open on X →

kyutai@kyutai_labs10 Jul

Transcription has many creative uses, from learning an instrument to music production. We’re excited to see what musicians can do with MuScriptor! Start transcribing: https://t.co/KoPtirqti8 Code: https://t.co/FKFNCoqBe4 Paper: https://t.co/wGGUcxjaN0 10K views · 125 likes · 8 reposts · 7 replies Open on X →

kyutai@kyutai_labs10 Jul

MuScriptor is a decoder-only transformer that autoregressively predicts a stream of MT3-like tokens from a mel-spectrogram of a 5-second audio excerpt. Longer samples are transcribed chunk-by-chunk. The model sizes span from 100M all the way to 1.3B parameters. 7.2K views · 32 likes · 0 reposts · 1 replies Open on X →

kyutai@kyutai_labs10 Jul

We pre-train on synthetic data: 1.5 million MIDI files, synthesized in many variations. Then we fine-tune on real data, the single biggest driver of quality. Finally, we post-train using RL on 300 tracks with manually verified accurate transcriptions. 8.3K views · 38 likes · 0 reposts · 1 replies Open on X →

kyutai@kyutai_labs10 Jul

The main improvement comes from data, the lack of which has bottlenecked automatic music transcription since MT3 (2022). We collected a dataset of 170k music recordings (11k hours), paired with MIDI transcriptions and spanning many genres. 7.9K views · 34 likes · 0 reposts · 1 replies Open on X →

kyutai@kyutai_labs10 Jul

We're releasing MuScriptor, the best open model for multi-instrument transcription to date, created in collaboration with @MireloAI. Give it a recording in any genre: pop, classical, metal, jazz, whatever, and it transcribes the individual instruments into MIDI. Link in 🧵 https:/ 305.6K views · 1.3K likes · 202 reposts · 38 replies Open on X →

เทียบกับบัญชีขนาดเดียวกัน

13 โพสต์จาก 90 วันที่ผ่านมา เทียบกับช่วง 10K–100K ผู้ติดตาม แสดงในวงกว้าง แต่มีผู้ชมตอบสนองน้อย.

ยอดดูมัธยฐาน8 289บัญชีนี้924ค่ามัธยฐานของ 10K–100K

การเข้าถึง, %29.93%บัญชีนี้3.62%ค่ามัธยฐานของ 10K–100K

การมีส่วนร่วม, %0.76%บัญชีนี้1.52%ค่ามัธยฐานของ 10K–100K

ตัวชี้วัด	บัญชีนี้	ค่ามัธยฐานของ 10K–100K	อัตราส่วน
ยอดดูมัธยฐานต่อโพสต์	8 289	924	8.97×
การเข้าถึง (ยอดดู ÷ ผู้ติดตาม)	29.93%	3.62%	8.27×
อัตราการมีส่วนร่วม	0.76%	1.52%	0.50×

บัญชีอื่นในช่วงนี้ → เปรียบเทียบกับบัญชีอื่น → ค่าอ้างอิงเหล่านี้คำนวณอย่างไร →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

305.6K10 Jul

7.9K

8.3K

7.2K

10K

15K11 Jul

27.6K6 Aug

37.3K20 Aug

104.4K25 Aug

638

555

39726 Aug

6.5K

Last 13 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.52%10 Jul

0.45%

0.47%

0.46%

1.41%

0.49%11 Jul

1.41%6 Aug

1.48%20 Aug

1.17%25 Aug

0.78%

0.36%

0.76%26 Aug

1.93%

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.

What the audience does

Likes46.2%3 551 in total

Reposts5.6%428 in total

Replies1.8%138 in total

Quotes1.2%89 in total

Bookmarks45.3%3 486 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

โพสต์ล่าสุด

เทียบกับบัญชีขนาดเดียวกัน

Growth & engagement

Views per post

Engagement rate per post

What the audience does

บัญชีที่คล้ายกัน