Video yükleniyor...

Video Yüklenemedi

Ana Sayfaya Dön

WanVideo sliding context window test on the 480p 14B T2V -model, worked surprisingly well but incredibly slow (50 mins on a 5090, 513 frames at 832x480)

Kijaidesign's profile picture

Jukka Seppänen

6,451 subscribers

13,085 görüntüleme • 1 yıl önce •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Yorum

Yorum bulunmuyor

Orijinal gönderinin yorumları burada görünecek

Benzer Videolar

WanVideo has been lot of fun, currently playing with applying sliding context windowing similar to AnimateDiff, the 1.3B model is especially suitable for this due to it's speed, here's 1025 frames in one go, under 5GB VRAM used, but took about 10 mins on a 5090 with 30 steps.
1:04

WanVideo has been lot of fun, currently playing with applying sliding context windowing similar to AnimateDiff, the 1.3B model is especially suitable for this due to it's speed, here's 1025 frames in one go, under 5GB VRAM used, but took about 10 mins on a 5090 with 30 steps.

Jukka Seppänen

13,808 görüntüleme • 1 yıl önce

Wan 2.1 14B LoRA training working on 24GB! 🥳 Avg 1.7sec/step at 480p on a 4090. LR was too low on this one.
0:33

Wan 2.1 14B LoRA training working on 24GB! 🥳 Avg 1.7sec/step at 480p on a 4090. LR was too low on this one.

Ostris

16,101 görüntüleme • 1 yıl önce

holy moly 🔥 Pusa 1.0 is a new image-to-video finetune based on Wan2.1-T2V-14B that surpasses Wan-I2V-14B and is: > 200x cheaper to train (only cost $500 🤯) > 5x faster inference > supports start-end frames, video extension and video completion
0:35

holy moly 🔥 Pusa 1.0 is a new image-to-video finetune based on Wan2.1-T2V-14B that surpasses Wan-I2V-14B and is: > 200x cheaper to train (only cost $500 🤯) > 5x faster inference > supports start-end frames, video extension and video completion

Dreaming Tulpa 🥓👑

15,688 görüntüleme • 11 ay önce

Today, I went out during a snowstorm to test Tesla's FSD V14.2.1. The car adapted in real time to the poor road conditions and performed incredibly well. My Model Y even counter-steered at one moment to offset some sliding. Mind blown! FSD handles snowy conditions very well.
35:24

Today, I went out during a snowstorm to test Tesla's FSD V14.2.1. The car adapted in real time to the poor road conditions and performed incredibly well. My Model Y even counter-steered at one moment to offset some sliding. Mind blown! FSD handles snowy conditions very well.

Sawyer Merritt

134,487 görüntüleme • 7 ay önce

RTX 5090 50% slower on the go!
1:00

RTX 5090 50% slower on the go!

PowerGPU

10,879 görüntüleme • 1 yıl önce

Everyone's comparing the DGX Spark to a 5090 and calling it slow. I think that's the wrong comparison. I ran Qwen3.6 35B-A3B FP8 with the full 262K context window enabled (~96GB RAM) — something gaming GPUs can't really do. Results: 🟢No context: 51.3 tok/s, TTFT 110ms 🟣200K prefill: 34.6 tok/s, TTFT 85s (~2,341 tok/s prefill) Prefill is way faster than a Mac. And 35 tok/s deep into 200K context, on a model this strong, is genuinely usable. The Spark plays a different game.
1:04

Everyone's comparing the DGX Spark to a 5090 and calling it slow. I think that's the wrong comparison. I ran Qwen3.6 35B-A3B FP8 with the full 262K context window enabled (~96GB RAM) — something gaming GPUs can't really do. Results: 🟢No context: 51.3 tok/s, TTFT 110ms 🟣200K prefill: 34.6 tok/s, TTFT 85s (~2,341 tok/s prefill) Prefill is way faster than a Mac. And 35 tok/s deep into 200K context, on a model this strong, is genuinely usable. The Spark plays a different game.

stevibe

33,301 görüntüleme • 2 ay önce

Just swapping out the model at a specific frame works surprisingly well for a transform animation 😁 #blender3d #gamedev
0:40

Just swapping out the model at a specific frame works surprisingly well for a transform animation 😁 #blender3d #gamedev

Kay 🎨

33,446 görüntüleme • 10 ay önce

🚀🎬We introduce TMD (Transition Matching Distillation): 480p videos generated from text prompts in < 3 NFEs! 1️⃣Main backbone for feature extraction and lightweight head for iterative refinement 2️⃣Distilled from Wan2.1 14B T2V combining MeanFlow & DMD2 🔗
0:26

🚀🎬We introduce TMD (Transition Matching Distillation): 480p videos generated from text prompts in < 3 NFEs! 1️⃣Main backbone for feature extraction and lightweight head for iterative refinement 2️⃣Distilled from Wan2.1 14B T2V combining MeanFlow & DMD2 🔗

Julius Berner

16,841 görüntüleme • 5 ay önce

Translating a full-length news article on iPhone 17 Pro with TranslateGemma from Google DeepMind The model works incredibly well on long text Running fully on-device with MLX
0:46

Translating a full-length news article on iPhone 17 Pro with TranslateGemma from Google DeepMind The model works incredibly well on long text Running fully on-device with MLX

Adrien Grondin

15,493 görüntüleme • 4 ay önce

Introducing Large Video Planner (LVP-14B) — a robot foundation model that actually generalizes. LVP is built on video gen, not VLA. As my final work at Massachusetts Institute of Technology (MIT), LVP has all its eval tasks proposed by third parties as a maximum stress test, but it excels!🤗
0:57

Introducing Large Video Planner (LVP-14B) — a robot foundation model that actually generalizes. LVP is built on video gen, not VLA. As my final work at Massachusetts Institute of Technology (MIT), LVP has all its eval tasks proposed by third parties as a maximum stress test, but it excels!🤗

Boyuan Chen

98,873 görüntüleme • 6 ay önce

Reform UK's Nigel Farage is optimistic on the party's chances against PM Keir Starmer, saying 'We're doing incredibly well in many, many traditional Labour areas'. Polls predict Reform UK could make huge gains in the election, a key test for Starmer's sliding popularity 🎥🇬🇧
1:46

Reform UK's Nigel Farage is optimistic on the party's chances against PM Keir Starmer, saying 'We're doing incredibly well in many, many traditional Labour areas'. Polls predict Reform UK could make huge gains in the election, a key test for Starmer's sliding popularity 🎥🇬🇧

AFP News Agency

10,377 görüntüleme • 2 ay önce

TIPPERARY GOAL - 64 mins Darragh Stakelum at the end of a well-worked move to raise a green flag for the hosts #rtegaa #allianzleagues 📺Tipperary GAA v Galway GAA Official on RTÉ2 and RTÉ Player
0:32

TIPPERARY GOAL - 64 mins Darragh Stakelum at the end of a well-worked move to raise a green flag for the hosts #rtegaa #allianzleagues 📺Tipperary GAA v Galway GAA Official on RTÉ2 and RTÉ Player

The Sunday Game

20,835 görüntüleme • 5 ay önce

ran Manaforge Omega on the alpha to test out the new raid frames. I don't think they're quite there yet, but it's a good first step
3:38

ran Manaforge Omega on the alpha to test out the new raid frames. I don't think they're quite there yet, but it's a good first step

Naomi 🐾

185,664 görüntüleme • 8 ay önce

The new Qwen 3.5 4B runs incredibly well on M5. The model is close to GPT-4o in benchmarks. Running fully on-device with MLX.
0:20

The new Qwen 3.5 4B runs incredibly well on M5. The model is close to GPT-4o in benchmarks. Running fully on-device with MLX.

Adrien Grondin

230,413 görüntüleme • 4 ay önce

china just dropped Step-Video-T2V, a new high-quality video model 🔥 > 30B parameters > text-to-video > 204 frames >quality & turbo mode
1:13

china just dropped Step-Video-T2V, a new high-quality video model 🔥 > 30B parameters > text-to-video > 204 frames >quality & turbo mode

Dreaming Tulpa 🥓👑

107,892 görüntüleme • 1 yıl önce

DeepSeek-V4 dropped. 1M context. 10x smaller KV cache. First open model where the context window and the agentic post-training meet.
0:36

DeepSeek-V4 dropped. 1M context. 10x smaller KV cache. First open model where the context window and the agentic post-training meet.

Ben Burtenshaw

49,900 görüntüleme • 2 ay önce

Idefics3-Llama is out! 💥 It's a multimodal model based on Llama 3.1 that accepts arbitrary number of interleaved images with text with a huge context window (10k tokens!) 😍 Link to demo and model in the next one 😏
0:37

Idefics3-Llama is out! 💥 It's a multimodal model based on Llama 3.1 that accepts arbitrary number of interleaved images with text with a huge context window (10k tokens!) 😍 Link to demo and model in the next one 😏

merve

28,014 görüntüleme • 1 yıl önce

Qwen3.5 runs quite well in mlx-lm. Awesome that we have a frontier-level hybrid model. The context gets longer but the inference speed and memory use barely change. Here's the Q4 generating a space invaders game on an M3 Ultra. Generated 4,120 tokens at 37.6 tok/s.
0:28

Qwen3.5 runs quite well in mlx-lm. Awesome that we have a frontier-level hybrid model. The context gets longer but the inference speed and memory use barely change. Here's the Q4 generating a space invaders game on an M3 Ultra. Generated 4,120 tokens at 37.6 tok/s.

Awni Hannun

48,673 görüntüleme • 4 ay önce

FlashPortrait based on Wan2.1 14B: faster infinite-length portrait animation; - no identity drift; - tops Wan-Animate in FID/FVD scores over 5400+ frames.
0:37

FlashPortrait based on Wan2.1 14B: faster infinite-length portrait animation; - no identity drift; - tops Wan-Animate in FID/FVD scores over 5400+ frames.

Wildminder

10,433 görüntüleme • 6 ay önce