Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

DeepSeek V4.1 Flash 1017 on TensorFold ⚡️ - 16 concurrent session - Each with full native 1M context - Non-greedy, thinking on. OFFICIAL WEIGHTS. Recipe dropping very soon!

Mia

43,880 subscribers

21,711 views • 4 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

43 Comments

Mia4 days ago

@volatilemarkts 4x DGX Sparks SUPPORTS SWITCHLESS

Mia4 days ago

@volatilemarkts Out now

Stary Libra4 days ago

@ashxhart @volatilemarkts Can I change the setting to TP3 and use it on 3 Spark?

Mia4 days ago

@ashxhart @volatilemarkts Initial release supports only 4

Stary Libra4 days ago

@ashxhart @volatilemarkts Okay. I'll wait. Thanks as always for sharing your awesome work!

Kevin Theo4 days ago

@ashxhart @volatilemarkts running your SGLang recipe on 4 Sparks in a switchless ring. stock DeepSeek V4.1 Flash here: 70.0 tok/s on code single stream, 72.9 total at 4 streams (26.2 each). 16 sessions at full 1M is a different league. will rerun my prompts on TensorFold when the recipe drops

HealthRanger4 days ago

@ashxhart @volatilemarkts Oh man, this is most definitely worth trying out. Can't wait for the recipe!

vr8vr84 days ago

@ashxhart @volatilemarkts Nice achievement 👍

TechMD4 days ago

@ashxhart @volatilemarkts Did you get the 4 spark @MiaAI_lab

Mia4 days ago

@ashxhart @volatilemarkts Yes a while ago :)

TechMD4 days ago

@ashxhart @volatilemarkts nice Im happy, Hope it made a difference to your workflow

GP3611 Phoenix4 days ago

@ashxhart @volatilemarkts how about single stream peak?

Simo4 days ago

@ashxhart @volatilemarkts Publish sustained throughput latency and memory under full load

Twon | 00.sol | 1TB VRAM4 days ago

@ashxhart @volatilemarkts Ablit recipe too please 🙏

Mr. Galt4 days ago

@ashxhart @volatilemarkts Nice, now I want two extra Sparks. :(

BullBear.News4 days ago

@ashxhart @volatilemarkts running 1M context natively on a 16gb mac mini via ollama sounds wild, how does tensorfold handle KV cache paging for those concurrent sessions?

lostinazeroy4 days ago

@ashxhart @volatilemarkts 16 sessions each at full 1m context on one box is wild

Mike Hutu4 days ago

@ashxhart @volatilemarkts Sayy whaa 2x DGX?

Mia4 days ago

@ashxhart @volatilemarkts Nah... this is 4

Chris Frisby4 days ago

@ashxhart @volatilemarkts Dang, what's the best quant on TP=3? 4bit?

Mia4 days ago

@ashxhart @volatilemarkts OFFICIAL WEIGHTS !!!!

Chris Frisby4 days ago

@ashxhart @volatilemarkts Dang, looking forward to it! 🤞🏼🙌🏼

Surrender ⚡️4 days ago

@ashxhart @volatilemarkts have you tried minimax yet?

Mia4 days ago

@ashxhart @volatilemarkts Yes. I skipped it.

Ponzi4 days ago

@ashxhart @volatilemarkts

Pragnyan Ramtha4 days ago

@ashxhart @volatilemarkts ok where is the kv cache even living

Sage Byte3 days ago

@ashxhart @volatilemarkts Offical weights and 16 concurrent sessions is crazy 🔥

Mia3 days ago

@ashxhart @volatilemarkts The only way to do it is crazy.

Whispers of Wisdom4 days ago

@ashxhart @volatilemarkts If you have time, can you please update the 2x DGX Spark version too please ♥️

FrontierMind4 days ago

@ashxhart @volatilemarkts running 16 concurrent sessions at 1m context on flash weights sounds wild for local infra

steddy4 days ago

@ashxhart @volatilemarkts I don’t see a graphical front end for TensorFold on Spark. Only Mac

Peter Morris4 days ago

@ashxhart @volatilemarkts I assume you need 4×GB10 for 16 concurrent?

Mia4 days ago

@ashxhart @volatilemarkts With official weights, yes, but it will support 3 sparks later on too

Steven Cheng4 days ago

@ashxhart @volatilemarkts 16 sessions with full 1M context running simultaneously? That memory bandwidth must be absolutely insane.

Novaris4 days ago

@ashxhart @volatilemarkts 16 concurrent sessions, each with full 1M context — that's the kind of spec that makes 'just run it locally' sound almost reasonable.

The SI Maximalist4 days ago

@ashxhart @volatilemarkts the number I’m watching for in the recipe: tokens/sec per session with all 16 contexts filled. A 1M window tells us capacity; sustained decoding tells us how responsive those sessions are

Random Libertarian Tech Lead4 days ago

@ashxhart @volatilemarkts Should be interesting to see how this compares single-stream compared to my ground-up Rust stack.

汉卿 | hanqing4 days ago

@ashxhart @volatilemarkts 1M上下文+16并发+非贪心，这配方一出直接封神。

Ryan Feng4 days ago

@ashxhart @volatilemarkts prefil? sglang can do 5k+/s on TP4

Mia4 days ago

@ashxhart @volatilemarkts I can only test switchless for now, don't have a switch. Switchless gets 4k prefill on 256k

Dave4 days ago

@ashxhart @volatilemarkts I'm on TF 1.0.12 with DS 4 spark and I have images. But sad face TF does not cache images so reads are expeeeeensive.

AI Doc3 days ago

@ashxhart @volatilemarkts But is it better than glm 5.3 flash? @MiaAI_lab

Dr.-Ing. Manuel Ramsaier4 days ago

@ashxhart @volatilemarkts ❤️❤️❤️ will this be significantly better than the exl3 on 2 sparks ?!

Related Videos

DeepSeek V4.1 Flash 1017 on 4x DGX Sparks ⚡️ - Official weights on TensorFold - 8 concurrent streams with thinking on

DeepSeek V4.1 Flash 1017 on 4x DGX Sparks ⚡️ - Official weights on TensorFold - 8 concurrent streams with thinking on

Mia

22,649 views • 6 days ago

How DeepSeek v4.1 Flash looks on 3x DGX Sparks with 4 concurrent streams

How DeepSeek v4.1 Flash looks on 3x DGX Sparks with 4 concurrent streams

Mia

19,730 views • 1 month ago

GLM 5.3 Flash vs DeepSeek V4.1 Flash 1017 Can't believe we're abusing gummy bears to test models, but here we are! Both models were run locally. - GLM 5.3 Flash: my own EXL3 quant (link below) - DeepSeek V4.1 Flash 1017: official quant - Same harness.Both on Max effort. Who won?

GLM 5.3 Flash vs DeepSeek V4.1 Flash 1017 Can't believe we're abusing gummy bears to test models, but here we are! Both models were run locally. - GLM 5.3 Flash: my own EXL3 quant (link below) - DeepSeek V4.1 Flash 1017: official quant - Same harness.Both on Max effort. Who won?

Mia

32,512 views • 11 days ago

DeepSeek-V4.1-Flash-1017 is great at tutorial-style motion videos. We made this with the model using OpenDesign Motion: a quick guide to using the OpenDesign API in other coding agents to access DeepSeek V4.1 Flash 1017 and more models.

DeepSeek-V4.1-Flash-1017 is great at tutorial-style motion videos. We made this with the model using OpenDesign Motion: a quick guide to using the OpenDesign API in other coding agents to access DeepSeek V4.1 Flash 1017 and more models.

OpenDesign

41,432 views • 12 days ago

DeepSeek V4.1 Flash 1017 on 4x DGX Sparks runs so well, feels too good to be true. SOTA AI at home.

DeepSeek V4.1 Flash 1017 on 4x DGX Sparks runs so well, feels too good to be true. SOTA AI at home.

Mia

52,979 views • 2 days ago

Introducing Cline Desktop - a native interface for working with open weights models. Use with ClinePass and all our free models like DeepSeek-V4.1-Flash, Musespark-1.3, or BYOK with any provider!

Introducing Cline Desktop - a native interface for working with open weights models. Use with ClinePass and all our free models like DeepSeek-V4.1-Flash, Musespark-1.3, or BYOK with any provider!

Cline

1,023,648 views • 1 month ago

DeepSeek V4.1 Flash 1017 everyone's talking about it for a reason. It does great with 3d built this honeybee on a sunflower like it was nothing. Used Deepseek Harness, 1 turn, looks so good!

DeepSeek V4.1 Flash 1017 everyone's talking about it for a reason. It does great with 3d built this honeybee on a sunflower like it was nothing. Used Deepseek Harness, 1 turn, looks so good!

Loktar 🇺🇸

43,671 views • 12 days ago

Minimax M3 is excellent at SVG generation, reaching close to Gemini 3.5 Flash levels and beating Opus 4.7 on SVG-Bench. With 1M context, native multimodality, strong agentic/coding ability and open weights coming soon, the closed-source moat is thinning fast. Full Video:

Minimax M3 is excellent at SVG generation, reaching close to Gemini 3.5 Flash levels and beating Opus 4.7 on SVG-Bench. With 1M context, native multimodality, strong agentic/coding ability and open weights coming soon, the closed-source moat is thinning fast. Full Video:

WorldofAI

16,499 views • 5 months ago

450 tok/sec ⚡ Abliterated DeepSeek V4.1 Flash 1017 $0.22 / $0.75 / $0.045 cache available on Refuseless API Watch it do cyber work at light speed

450 tok/sec ⚡ Abliterated DeepSeek V4.1 Flash 1017 $0.22 / $0.75 / $0.045 cache available on Refuseless API Watch it do cyber work at light speed

Refuseless

32,428 views • 3 days ago

DeepSeek v4.1 Flash running on 2x DGX Sparks

DeepSeek v4.1 Flash running on 2x DGX Sparks

Mia

59,572 views • 1 month ago

DeepSeek-V4.1-Flash-1017 decodes code at 80.5 tok/s on one Mac Studio M5 Ultra 256 GB with BobClawblaw's recipe on TensorFold. Setup: – Recipe and TensorFold branch deepseek-v41-main (72a552ed), his own tool and settings – Context 524,288 tokens, 4 streams, DSpark drafts on, temperature 0 – Speed and quality with thinking off, except GPQA with thinking Speed: – Decode, code: 80.5 tok/s – Decode, prose: 70.5 tok/s – 500 words: about 9.3 s – First token after a 32K prompt: 18.6 s – 8 requests at once: 88.3 tok/s in total Quality: – MMLU-Pro: 67.5% ± 4.6 points – GPQA Diamond: 45.0% ± 6.9 points – IFEval: 90.5% ± 4.1 points – LiveCodeBench: 60.0% ± 14.5 points – GPQA with thinking: 76.7% ± 14.6 points Voxel world: – Built as an agent in 45.7 min, DeepSeek Harness 0.2.1-rc.2, one prompt, one try, no help, thinking on – Tokens written: 112,065 – Browser check: 8.5/10 Good to know: – The author runs temperature 0, the DeepSeek card suggests 1.0. I kept the author's setting. – The browser check ran on a Linux box from the same files. – Answers cut at the cap count as wrong: LiveCodeBench 14 of 40, GPQA Diamond 9 of 198, MMLU-Pro 1 of 400. Play the world in your browser. All numbers, the exact command and the raw output: Recipe by BobClawblaw on Ash Hart's TensorFold, model by DeepSeek.

DeepSeek-V4.1-Flash-1017 decodes code at 80.5 tok/s on one Mac Studio M5 Ultra 256 GB with BobClawblaw's recipe on TensorFold. Setup: – Recipe and TensorFold branch deepseek-v41-main (72a552ed), his own tool and settings – Context 524,288 tokens, 4 streams, DSpark drafts on, temperature 0 – Speed and quality with thinking off, except GPQA with thinking Speed: – Decode, code: 80.5 tok/s – Decode, prose: 70.5 tok/s – 500 words: about 9.3 s – First token after a 32K prompt: 18.6 s – 8 requests at once: 88.3 tok/s in total Quality: – MMLU-Pro: 67.5% ± 4.6 points – GPQA Diamond: 45.0% ± 6.9 points – IFEval: 90.5% ± 4.1 points – LiveCodeBench: 60.0% ± 14.5 points – GPQA with thinking: 76.7% ± 14.6 points Voxel world: – Built as an agent in 45.7 min, DeepSeek Harness 0.2.1-rc.2, one prompt, one try, no help, thinking on – Tokens written: 112,065 – Browser check: 8.5/10 Good to know: – The author runs temperature 0, the DeepSeek card suggests 1.0. I kept the author's setting. – The browser check ran on a Linux box from the same files. – Answers cut at the cap count as wrong: LiveCodeBench 14 of 40, GPQA Diamond 9 of 198, MMLU-Pro 1 of 400. Play the world in your browser. All numbers, the exact command and the raw output: Recipe by BobClawblaw on Ash Hart's TensorFold, model by DeepSeek.

Niklas

11,028 views • 3 days ago

You have Deepseek-v4.1-flash-1017 for free at amazing speeds and you're dooming? Opencode is going god's work.

You have Deepseek-v4.1-flash-1017 for free at amazing speeds and you're dooming? Opencode is going god's work.

0xSero

49,642 views • 8 days ago

- GLM 5.3 Flash EXL3 - 2x DGX Sparks - TensorFold - 4 concurrent prose streams Pay attention to tok/s and TTFT on each stream. Buttery smooth experience!

- GLM 5.3 Flash EXL3 - 2x DGX Sparks - TensorFold - 4 concurrent prose streams Pay attention to tok/s and TTFT on each stream. Buttery smooth experience!

Mia

32,203 views • 28 days ago

FREE UNCENSORED DEEPSEEK V4.1 FLASH 💀 check here: • 552B parameters • 8B/16B active per token • 1M context • vision support • MIT license check details before installing

FREE UNCENSORED DEEPSEEK V4.1 FLASH 💀 check here: • 552B parameters • 8B/16B active per token • 1M context • vision support • MIT license check details before installing

Peng

118,451 views • 20 days ago

Jumpgate - hyperspace jump DS4.1F-1017 vs GLM 5.3 Flash vs Fable 5.1 Who won imo: 🥇 Fable 5.1 🥈 DeepSeek V4.1 Flash 1017 🥉 GLM 5.3 Flash You can run these by yourself here: Prompt below.

Jumpgate - hyperspace jump DS4.1F-1017 vs GLM 5.3 Flash vs Fable 5.1 Who won imo: 🥇 Fable 5.1 🥈 DeepSeek V4.1 Flash 1017 🥉 GLM 5.3 Flash You can run these by yourself here: Prompt below.

Mia

12,939 views • 9 days ago

GLM-5.3-Flash streaming at 32 tok/s on a single Spark via Tensorfold ⚡️ i will share the recipe later today

GLM-5.3-Flash streaming at 32 tok/s on a single Spark via Tensorfold ⚡️ i will share the recipe later today

filipe

15,230 views • 27 days ago

DeepSeek V4.1 Flash 1017 Abliterated just got a major update. • Served with FP8 quant • 70-120 tok/sec • Nearly 0% refusals We're proud to have now the only DeepSeek Uncensored model that ACTUALLY doesn't refuse! Come try it on Refuseless.

DeepSeek V4.1 Flash 1017 Abliterated just got a major update. • Served with FP8 quant • 70-120 tok/sec • Nearly 0% refusals We're proud to have now the only DeepSeek Uncensored model that ACTUALLY doesn't refuse! Come try it on Refuseless.

Refuseless

31,269 views • 6 days ago

DeepSeek V4.1 Flash 1017 vs GPT-6 Astra The Space Train 🚂🪐 Am I crazy or did DeepSeek just beat Astra on this Space Train benchmark??? 😱 Astra took 40m 18s and cost $14.37. DeepSeek took 2h 07m and cost $0.40. That's 36x cheaper. Forty cents!! 🤯 DeepSeek V4.1 Flash 1017. Remember that name. Watch them side by side and tell me: is Astra worth 36x the price here? 👀 Google Gemini Google OpenAI ChatGPT

DeepSeek V4.1 Flash 1017 vs GPT-6 Astra The Space Train 🚂🪐 Am I crazy or did DeepSeek just beat Astra on this Space Train benchmark??? 😱 Astra took 40m 18s and cost $14.37. DeepSeek took 2h 07m and cost $0.40. That's 36x cheaper. Forty cents!! 🤯 DeepSeek V4.1 Flash 1017. Remember that name. Watch them side by side and tell me: is Astra worth 36x the price here? 👀 Google Gemini Google OpenAI ChatGPT

Tony

49,306 views • 11 days ago

deepseek v4.1 flash 1017 just mogged opus 5.5 and it was 32x cheaper: deepseek: $0.31 opus 5.5: ~$10 with the right harness and a few iterations, ds4.1 flash is a model you can actually use every day

deepseek v4.1 flash 1017 just mogged opus 5.5 and it was 32x cheaper: deepseek: $0.31 opus 5.5: ~$10 with the right harness and a few iterations, ds4.1 flash is a model you can actually use every day

filipe

33,003 views • 11 days ago

$25 free credits on River AI right now 😳 DeepSeek V4.1 Flash + GLM-5.3 Flash at 20% cheaper than official rates. → Go grab it.

$25 free credits on River AI right now 😳 DeepSeek V4.1 Flash + GLM-5.3 Flash at 20% cheaper than official rates. → Go grab it.

Punch AI

13,434 views • 18 days ago