event

DeepSeek v4.1 Release

22 items · started 2026-11-07 · ongoing (last activity 2026-11-16)

  1. Editing one row of an LLM's output layer to stop it overthinking A linear probe folded into the </think> row lets the model's own hidden state decide when its thinking is done. Tested on Qwen3.5-4B and on DeepSeek V4.1 Flash (763B, running…

  2. Lawyer upgrades eight-V100 private AI server to run 80% of DeepSeek V4.1 Flash at 4-bit NVFP4 precision — adds 1TB of RAM, an Nvidia RTX 3090, and fixes crashes caused by a bad PSU The upgrade costs speed: code output falls to 150–180 toke…

  3. Disclosure: I'm the author. Both tools are free, GPL-3.0 (it covers the tools, not your code), no paid tier.

  4. 50X tunes custom and open-source model inference to your workload: faster responses, higher limits, more tokens per dollar.

  5. could not extract summary

  6. DeepSeek Ultrafast An ultra-fast inference engine for DeepSeek-V4.1-Flash for sm100a and sm103a. It supports 1M context length and multi-server KV cache sharing.

  7. DeepSeek V4.1 Flash is now available on Verda Inference, at $0.15 per million input tokens, $0.006 for cached input and $0.60 for output. That’s about 50% less than most providers charge.* It’s DeepSeek's capable, efficient model for every…

  8. On 13 September 2026, an engineer at DeepSeek published a post on WeChat titled “I had to bury my talent in yesterday”.1 This engineer wrote the main attention kernel2 of the company’s DeepSeek v4.1 model. Their post is about what is happe…

  9. I built a TUI called clingon (an opinionated terminal cheatsheet runner) that helps you to stop googling commands. It has a menu of commands beside your real shell, and now with an AI that can drive it.

  10. genuinely, deepseek v4.1 flash is better than this bullshit its always repetetive, its thoughts make no sense, its like fucking worse than gpt 3.5 like genuinely either improve this shit or juts get the fuck off of ai industry elon your ai…

  11. Some large open-weight models require more memory than a consumer computer provides. DeepSeek-V4.1-Flash-1017 and Muse-Spark-Open are 763B- and 816B-class releases.

  12. Also, you can share your simulation with query parameters, ex) https://cro.sh/specmaxx/?m=deepseek-ai/DeepSeek-V4.1-Flash&d...

  13. Blog · Oct 2026 LTX-2.5 at 4K, in half the time Maxwell achieves 1.87x faster 4k video generation in 134s per 4K clip on two B200s, down from 250s on a pipeline built and optimized by top human experts Maxwell made DeepSeek-V4.1-Flash on B…

  14. If you had to pick a model to coordinate a team of agents, would you just choose the one that performs best on the task? Our team at Einsia tested this in WorkflowArena, and the results suggest that’s not enough.

  15. DeepSeek-V4.1's 203 GB of Engram n-gram tables sit in host DRAM today, taking 190 of the 223 GiB a 4-GPU serving process uses. Row indices depend only on input tokens, so reads can start with each step and overlap earlier layers.

  16. FrontierHarness v1.0 compared 9 harnesses on a single model. Its obvious limitation was that we couldn’t say whether a harness that wins on one model wins on another.

  17. English | 日本語 | 简体中文 | Português Project Maya GLM-5.3-Flash (321 B), DeepSeek-V4.1-Flash-1017 (748 B) and MiniMax-M3.1-Flash (428 B) on your own GPU private and fast · one NVIDIA GPU or up to 16 · AMD and Windows · chat, pictures, OpenAI-…

  18. In less than $1, I just built myself a Mac App to route audio output to different devices using Deepseek 4.1 Flash 1017 via @nebiustf My use-case was that i had two audio playing - one on VLC and another on Brave and I wanted one to be out…

  19. inspired by Bijan Bowen's video - Subway FPS test. Hardware details: RTX 5090, 256GB DDR4 3200Mhz 8 channel, AMD EPYC 7532 CPU (32 core).

  20. Hey all, I wanted to share that we're working on zml/llmd, a fast (we believe the fastest) inference engine that currently runs on NVIDIA, AMD, Intel, TPU and Metal. It has all the features of a modern LLM server: continuous batching, pref…

  21. Run the full 510 GB DeepSeek-V4.1-Flash on a five-year-old gaming PC The original model, not a smaller copy · 64 GB of RAM and one 12 GB NVIDIA card · free and open source PROMETHEVS runs DeepSeek-V4.1-Flash-1017, one of the strongest open…

← all threads