
Founder @ Pixel ML. Prev @AlibabaGroup. Building video intelligence agents: Composer turns archives into output; Sentinel turns cameras into evidence.
APJ
Joined September 2012
- Tweets2K
- Following1.3K
- Followers974
- Likes7.5K
DeepSeek-V4.1-Flash-1017 at home (4× CMP 170HX + 256GB DDR4 RAM)
- decode 130.2 / 97.2 / 200.8 (code / prose / structured), c8 aggregate 372 / 255 / 671 tok/s pretty decent coding speed at frontier level
- prefill 1181 / 1688 / 1587 tok/s at 2k / 8k / 32k ctx
video sample from api, our quant EXL3 3 bpw
recipe @BBacktesting @Morrowmake
Model, Apache-2.0 → huggingface.co/PixelML/sea-a…
Labels → huggingface.co/datasets/Pixe…
Code → github.com/PixelML/sea-asr
We open source SEA-ASR-MS, a fast speech-to-text for Bahasa Melayu (Apache 2.0 open model + 70 h labelled speech dataset).
- 1.1 s to transcribe 57 minutes of a Malay talk show (3,277× real time, ~24× faster than Whisper large-v3 on 1x H100).
- 8.8% WER with its n-gram LM (on par with Qwen3-ASR-1.7B, 6x larger), 8.4% with a 1B rescoring LM. Whisper large-v3 is still more accurate on FLEURS (7.1%), working on closing the gap.
~ surprisingly with just a little data, we have quite good results and we transfer learning from our ID variant.
Finetune on Google's EmbeddingGemma 2, same recipe as SEA-ASR-ID, inspired by @TslShahir @stablequan
1/ Introducing the Cua Spaces SDK.
Every big lab now ships a bot with its own computer: Grok Bot, Muse, dots. None let you bring your own. With Cua Spaces, now you can.
Put Claude Code, Codex or any agent on compute you choose, and run jobs across all your computers at once.
We open source SEA-ASR-ID a fast speech-to-text for Bahasa Indonesian (Apache 2.0 364M params open model + 2,287.4 h speech dataset)
- 1.5 s to transcript 79 minutes of Indonesian parliament (3,582× real time and 26–42x cheaper than Whisper large-v3 on 1x H100)
- 5.5% WER with a 1B rescoring LM (on par with Whisper large-v3 1.5B), 6.1% WER with its n-gram LM (ties Whisper large-v3-turbo)
Finetune on Google's EmbeddingGemma 2 inspire by @TslShahir @stablequan
Model, Apache-2.0 → huggingface.co/PixelML/sea-a…
Labels → huggingface.co/datasets/Pixe…
Code → github.com/PixelML/sea-asr
blog.cloudflare.com/cloudfla… move your storage to R2
imagine this but replace by Open AI/Anthropic, own your compute
This is a wild approach, @autodesk.
We’re a small business paying $380 per user, per month. We’ve been paying customers for over a decade, and we make a deliberate effort to honor the license agreements for every product we use. Your software already requires authentication and an assigned license. What, specifically, prompted this email?
Earlier today, I had two conversations on two separate podcast episodes about your direction with Fusion and why cloud-based API access makes sense for agentic MEP workflows. I was making the case for Autodesk. I think the product direction is right based on recent posts from @andrew_anagnost.
Our company has been deciding where to invest our development effort: integrate our custom cloud agent harness into Autodesk, build on an open-source platform, or adopt one of the emerging cloud alternatives. That decision will lock in years of internal development and software spending. This decision is also central to our plans to expand @PermitZIP nationwide.
An email like this becomes part of that decision. So does the time we have to spend responding to it.For more than a decade, we’ve paid for anywhere from 5 to 30 licenses each year. Autodesk is our single largest SaaS expense. This is a heck of a way to make our next platform decision easier.
Starting next week, we’ll begin transitioning away from Autodesk products. We have nothing to hide and will cooperate with the audit. We’ll also be directing our future spending toward companies that earn our confidence as partners.
I hope your audit goes well.
Try this again with Deepseek V4.1 Flash 1017 today.
- One shot in 1h50m with Deepseek Harness
- Output: 269,823 tokens at 258 tokens/s (via API)
Initial impression is that this model like to think a lot and write longer prompt, the visuals are definitely up
Still quanting to be able to test on my rig
Sean Phan retweeted
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
Opus is taking over now, finally i can get rid of ads and bloatware from my TCL TV
Hacked my TV with Claude Code today
No more ads, bloatware, spyware or laggy UI
I also got a beautiful screensaver with my revenue, calendar, weather, and notifications for new Stripe payments
It was super easy! I sent it the listing for my tv, asked "can you hack this?" And it talked me through how




















