Loading video...
Video Failed to Load
Why choose between autoregressive (fast, streaming) and bidirectional (high-quality) video diffusion when a single model can do both even better? Excited to share Flex-Forcing, accepted as spotlight at #ICML2026! 📍Say hi to Chao Liu at our poster: July 9 | 2:30 - 4:15 PM KST | Hall A #3909... show more
10,443 views • 3 months ago •via X (Twitter)
3 Comments
At test time, Flex-Forcing allows you to dynamically change the chunk-size over frames and diffusion time-steps. 1⃣Bidirectional: Maximal context for global coherence 2⃣Autoregressive: Causal generation for fast streaming 3⃣Hybrid: Optimal tradeoff in terms of inference speed and quality
What does this mean in practice? 1⃣Better speed-quality tradeoff: Outperforms baselines in terms of both FPS and VBench score. 2⃣Coherent long video generation: Maintains stability over time while preserving a higher dynamic degree than baselines. 3⃣Any-order, any-timestep editing: Edit arbitrary video segments without triggering cascading errors in later frames.
Check out our paper and project page for more information! 📜 Paper: 💻 Project Page:
![At the News18 Town Hall, I addressed a question often asked by critics: Is the #DravidianModel of high industrial growth and high welfarism sustainable? For us, the answer is not just "yes" it is the only way to build a "Good Society." As I shared, the state has three primary responsibilities: improving the quality of life through public goods, creating an environment where entrepreneurs and innovators can thrive, and keeping the gap between the rich and the poor as low as possible to ensure social cohesion. We have always prioritized the social aspects of governance, nutrition, health, and education, as the essential foundation for all growth. This is not just a welfare choice; it is a predicate for economic success. By building high-quality human capital, we ensure that global investment and jobs find us. The results speak for themselves: an 11.19% growth rate and a 700% increase in electronics exports in just the last few years. When critics highlight "better" fiscal metrics in other states, the trade-off is clear. I will always choose a path that invests in the lives and dignity of our citizens, even if it requires borrowing within manageable limits, rather than accepting models where massive capital expenditure coexists with high levels of multi-dimensional poverty. We do not measure success by the creation of a few super-rich billionaires while the average resident lags behind; we measure it by the quality of life and the capacity to aspire for every single citizen. Watch the video here: [English W\ Tamil CC]](https://image.24vids.com/tw-2027351064329289743/amplify_video_thumb/2027350853259259905/img/qrSjyZeR2y4ShHUq.jpg)
![We made a thing! Very happy to announce sqlcoder-pro and the Defog Alignment Platform. Available to use immediately without a wait-list, weights will be open-sourced very soon. The video does a quick show and tell comparison against ChatGPT (with gpt-4o). Read on for more details! TLDR 💪 equal (or better) performance on text-to-SQL as the most capable Claude-3.5 or GPT-4 models 🤝 You can use it today on a free plan/free trial, without a waitlist 🪽 self-hostable on a single RTX4090, with 2 second median generation times for SQL queries 🔁 exactly the same output every time, give the same prompt 👨🏻🏫 teachable and steerable: show the model what you want it to do 🛞 debuggable – you can understand WTF is going on inside the model, instead of treating it like a black box Let's dig into each of these one-by-one! Performance SQLCoder-8b-pro significantly exceeds the performance of our previous sqlcoder-8b model on Postgres text-to-SQL (from 88.2% to 90.2% accuracy - gpt-4o is at 87.6%, for reference). It is also better at following instructions. This was done via self-merges, hand crafted fine-tuning data, and adapting the training data to fit our tokenizer. Cost You can host this on the model on a single $3,500 RTX4090, and support ~5 requests/second via VLLM. If you're looking to host on the cloud instead, you can run it on a single L4 GPU that costs $300/mo on GCP Repeatability We have a dense 8b model with no MoE shenanigans. For the same prompt with temperature=0, you'll always get the same answer – which is critical in BI. Teachable In our alignment and feedback modes, you can give the model feedback on how it answered certain questions, and it will automatically adapt to the feedback. Debuggable You can use logprobs and attention scores to determine where, exactly is the model paying attention to inside a prompt + what it's getting confused by when generating outputs. Available today You can use Defog on the cloud today by going to docs[dot]defog[dot]ai, and getting an API key. Excited to hear what you think!](https://image.24vids.com/tw-1825943270935113842/ext_tw_video_thumb/1825942610525220864/pu/img/qmpqqvd_3KNdJDGd.jpg)
