Natalia Sokolova

[Natalia Sokolova @medai_li]

@satellitedown @ashxhart Your speed is indeed fast, but we use it in real-world scenarios, so we choose 700 🏆🏆🏆

[Ash Hart @ashxhart]

5090 go burrrr

Qwen 3.8 nvfp4 300tps decode on prose with 4 lanes. Let's widen them 👀

[Natalia Sokolova @medai_li]

@ashxhart We can’t keep up with the updates.🏆🎉😂

[Ash Hart @ashxhart]

TensorFold 1.2.7 is out. 🔥

A big fix for long conversations on Metal: Qwen3.8-27B now reuses its history even when the client drops earlier reasoning.

In the measured M3 Ultra test with 34k history and a 5 GiB prompt cache, follow-up time to first token dropped from about 85s to 0.3s. Same replies in that test.

Also in this release:

• Responses API over WebSocket.
• LightOnOCR-3 0.8B and 4B image chat on CUDA.
• Opt-in GLM conversation reuse at turn boundaries on CUDA.
• Prompt-cache and disk-tier stats in the dashboard.
• Targeted speed and memory improvements across Metal, CUDA and HIP.
• Fixes for GPU builds, malformed GLM tool calls, damaged JPEGs and disk-cache recovery.

Massive thanks to everyone contributing code, testing on their hardware and reporting issues. I really appreciate the time you put into this.

@andersjw_ @Bizuayeu @defilan @devopsfortunato @unigilby @kelchm @teleogen @petruspennanen

And on GitHub: AbrahamCete, BobClawblaw, ciprianveg, garvrao80, jschmied, laidick, millaguie, Mirrdhyn, mrv777, tacosta, volfco and wojo.

Thank you too to the tester who reported the JPEG fault privately.

@volatilemarkts @MiaAI_lab @petruspennanen @plotarmordev

[Natalia Sokolova @medai_li]

@SandhuJonathan @ashxhart We need the videos in the form in which they are. If you need engineering, you can find it on GitHub.😊

[Natalia Sokolova @medai_li]

@ashxhart You’re so lucky, Ashley.🥳 Congratulations🎉☃️🤩

[Ash Hart @ashxhart]

4x Pro 6000's go burrrr

[Natalia Sokolova @medai_li]

@ashxhart It turned out very beautifully 🤩🤩🤩🎉☃️☃️☃️

[Natalia Sokolova @medai_li]

@ashxhart So fast and beautiful.🤩🎉☃️

[Ash Hart @ashxhart]

🤯

[Natalia Sokolova @medai_li]

@ashxhart Today we’ll upload it.🏆

[Ash Hart @ashxhart]

TensorFold 1.2.6 is out. 🚀

The big one here is the prompt cache under multi-agent load.

In the 16-agent Qwen3.8-27B + DFlash2 test, median follow-up time to first token dropped:

• M3 Ultra: 546s → 15.9s
• GB10: 105s → 25.8s

Each agent started with a ~17k-token prompt. Same replies throughout that test.

Also in this release:

• Flash Next cold prompts finish 29 to 41% sooner on HIP, measured on the Radeon 8060S.
• GLM-5.3-Flash single-stream code decoding is 32% faster on two GB10s at TP=2.
• Opt-in DSpine-b16 drafting from Tencent for Qwen3.8-27B on CUDA.
• An image-generation API for image-capable engines, local Unix sockets and tool-call fixes.

One upgrade note: kept prompt states need a fresh prefill after updating.

Massive thanks to everyone contributing code, testing on their hardware and reporting the things that need fixing. I really appreciate the time you all put into this.

@andersjw_ @Bizuayeu @defilan @devopsfortunato @unigilby @jvr0x @teleogen

And on GitHub: abm9111, AbrahamCete, akol1, BobClawblaw, chadhurley25075-png, davidtai, ElderOrb, jschmied, kotevcode, Mirrdhyn, mrv777, ppardus, robertung88 and volfco.

@petruspennanen @volatilemarkts @plotarmordev @MiaAI_lab

https://t.co/H2dBxCSIKt

[Ash Hart @ashxhart]

I haven’t even got my hand on the M5U 512 yet and TF is cooking 🚀💪🏻

[Ash Hart @ashxhart]

I think we can double it to 88 👀🚀

[Natalia Sokolova @medai_li]

@wikiwayne So Wayne, where are your successful projects? 🤔🍿🤡

[Ash Hart @ashxhart]

@HowToPrompt__ Did this in september

https://t.co/ZNOBH22vVD

[Natalia Sokolova @medai_li]

@ashxhart No more than 12 hours.🎉👏🏆

[Natalia Sokolova @medai_li]

@ashxhart Your wish came true in less than 12 hours.🎉💯 🏆

[Natalia Sokolova @medai_li]

🟩🟩🟩🟩🟩🟩🟩⬜⬜⬜ 73% of goal on Ko-fi! https://t.co/BsmcskpyI8
👏👏👏👏👏👏👏 Just a little bit left

[Natalia Sokolova @medai_li]

@ashxhart 📌📌📌 Explanation: I don’t need to transfer anything. I do my job for 50 percent of the actual cost, but the main condition is that all the money goes to Ashley for buying his Mac with 512 GB of memory and 2 TB of free space. For helping Ashley, you get the service at half price