@seapy

Co-founder & CTO at ๋‹น๊ทผ(Karrot)

seapy.com
Joined February 2009
  • Tweets5.9K
  • Following648
  • Followers1.9K
500Media
Seapy retweeted
0xSero
@0xSero20h
Stable 650+ tok/s on deepseek-v4.1-flash-1017 I am in heaven here
51
10
1
671
28,409
Seapy retweeted
vLLM
@vllm_projectOct 23
๐ŸŽ‰ MiniMax M3.1 Flash from @MiniMax_AI has Day-0 support in vLLM! Built for multi-step automation, real-codebase engineering and visual reasoningโ€”with native NVFP4 KV, NVFP4 MoE weights and DSpark speculative decoding in the serving stack. What the model can do, and the vLLM engineering that makes it practical to serve. ๐Ÿ‘‡ Model: huggingface.co/MiniMaxAI/Minโ€ฆ 1/5
MiniMax (official)
@MiniMax_AIOct 23
MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token. Agentic performance: A generational leap in agentic tasks, coding, and cybersecurity โ€” 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82.2 on CyberGym. Efficient by design: MiniMax Sparse Attention and DSpark speculative decoding improve inference efficiency. Hugging Face: huggingface.co/MiniMaxAI/Minโ€ฆ ModelScope: modelscope.cn/models/MiniMaxโ€ฆ API: platform.minimax.io M Plan: minimax.io/m-planMiniMax MCode for desktop: agent.minimax.io/download

MiniMaxAI/MiniMax-M3.1-Flash ยท Hugging Face

Weโ€™re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co
15
34
5
191
10,511
As of today, during the Python Core Sprint, I have made 500 commits to CPython. It's been 910 days since my 400th commit. See you at 600 commits! Thanks to all the core devs working with me. It's always such a pleasure to collaborate with you all!
Donghee Na@dongheena9225 Apr 2024
As of today, I have committed 400 commits to the CPython project. It's been 244 days since the last 300 commits. See you on the next 500 commits! Thanks to all the core devs working with me and see you soon at US PyCon :)
2
1
4
631
Seapy retweeted
tobi lutke
@tobiOct 22
Craziest day in cyber security in a long time
Tony Ma
@t1o6n3yOct 22
๐Ÿšจ Security vulnerabilities reported in a SINGLE day: PHP โ€” 2 CVEs Node.js โ€” 10 CVEs NGINX โ€” 5 CVEs Go โ€” 1 CVE GitLab โ€” 12 CVEs eMudhra โ€” CA certificate compromise
81
156
16
3,411
279,009
Seapy retweeted
DHH
@dhhOct 22
I remember when 200tps felt impossibly fast. This is going to be crazy.
Fireworks
@FireworksAI_HQOct 21
DeepSeek V4.1 Flash 1017 runs at 1,521+ tok/s on Fireworks Ultrafast. Fast enough to ship Tetris in 11 seconds, 7x faster than Claude Opus 5.5. Join the Ultrafast waitlist: bit.ly/3UA0wKq
148
150
22
4,983
250,618
Seapy retweeted
DHH
@dhhOct 19
I left an Omarchy terminal in Chinese, and I'm not going to lie, this looks pretty bad ass when doing omarchy update!
211
72
19
2,772
211,057
๋“œ๋ŽŒ 10์ธ์น˜ ์„œ๋ฒ„๋ž™์„ ์™„์„ฑํ–ˆ๋‹ค. ๋‚ด 3D ํ”„๋ฆฐํŒ… ํ”„๋กœ์ ํŠธ ์ค‘ ๊ฐ€์žฅ ํฐ ํ”„๋กœ์ ํŠธ๋กœ ์™ธ๋ถ€ ํ”„๋ ˆ์ž„์€ ๋ฉ”์ด์ปค์›”๋“œ์—์„œ ๊ฐ€์Ÿˆ์™”์ง€๋งŒ ๋‚ด๋ถ€ ํ”„๋ ˆ์ž„์€ ๋‚ด ์žฅ๋น„์— ๋งž๊ฒŒ ChatGPT๋กœ ๋‹ค ์„ค๊ณ„ํ–ˆ๋‹ค. ์ค‘๊ฐ„์ค‘๊ฐ„ ์•„์‰ฌ์šด ๋ถ€๋ถ„๋„ ์žˆ์ง€๋งŒ ์ „์ฒด์Ÿ‰์œผ๋ก  ๋งŒ์กฑ์Šค๋Ÿฝ๋‹ค ใ…Žใ…Ž
1
1
28
1,006
Seapy retweeted
Mia
@MiaAI_labOct 17
Rejoice single DGX Spark owners! You can now run Qwen3.8 Flash Next on TensorFold with the new Nvidia FP4 QAD quant ๐Ÿ”ฅ - 1.1M total FP8 KV cache - 256k context, 5 concurrent users - 46-55 tok/s on prose, single stream - 132 tok/s on prose, 5 concurrent steams - Prefill 2000-2300 tok/s Qwen3.8 Flash Next is still the BEST model to run on one DGX Spark, bare none! You can choose what to run: FP4 QAD for quality, or int4 for speed. Details on what is QAD below. Get it here: mia-ai.net/models/Qwen3.8-Flโ€ฆ

Qwen3.8 Flash Next on one DGX Spark with TensorFold ยท Mia's AI Lab

Serve Qwen3.8 Flash Next from one DGX Spark on TensorFold's Zig engine, with NVIDIA's FP4 QAD checkpoint (46-55 tok/s for one request, 132.7 for five at once) or the faster INT4-AutoRound (64.4,...

mia-ai.net
50
56
13
655
61,937
Seapy retweeted
Mia
@MiaAI_labOct 17
The 'old' vLLM recipe for DeepSeek v4.1 Flash *should* work with the new 1017 release. I haven't tried it though. Just try replacing the model. That being said, I'm working on a TensorFold recipe. vLLM recipe here: github.com/MiaAI-Lab/DeepSeeโ€ฆ

GitHub - MiaAI-Lab/DeepSeek-v4.1-Flash-DGX-Sparks: DeepSeek-V4.1-Flash on 3-4x NVIDIA DGX Sparks

DeepSeek-V4.1-Flash on 3-4x NVIDIA DGX Sparks. Contribute to MiaAI-Lab/DeepSeek-v4.1-Flash-DGX-Sparks development by creating an account on GitHub.

github.com
Mia
@MiaAI_labOct 17
NEW DeepSeek v4.1 Flash-1017 is LIVE!! ๐Ÿ”ฅ Looks like it improved on everything! Link below.
25
7
218
13,277
Linus Torvalds โ€œAI ์‚ฌ์šฉ์„ ๊ธˆ์ง€ํ•˜์ง€ ์•Š๊ฒ ๋‹ค. ์ž์‹ ์ด ํ•˜๋Š” ์ผ์„ ์ดํ•ดํ•ด์•ผ ํ•œ๋‹คโ€ Linux ์ปค๋„ ๊ธฐ์—ฌ์ž์—๊ฒŒ ์š”๊ตฌํ•˜๋Š” ๊ฒƒ์€ ์ตœ์„ ์„ ๋‹คํ•˜๋Š” ํƒœ๋„์™€ ์ž‘์—…์— ๋Œ€ํ•œ ์ดํ•ด์ด๋ฉฐ, ์–ด๋–ค ๋„๊ตฌ๋ฅผ ์“ฐ๋Š”์ง€๋Š” ์ดํ•ด ์ˆ˜์ค€๊ณผ ๊ด€๋ จ์ด ๋งค์šฐ ์ ๋‹ค๋Š” ์ž…์žฅ ๋„๊ตฌ ์—†์ด ์—ญ๋Ÿ‰์„ ์ฆ๋ช…ํ•˜๋ผ๋Š” ๋…ผ๋ฆฌ๋ผ๋ฉด, ๊ธฐ๊ณ„์˜ ๋™์ž‘์„ ์ดํ•ดํ•œ๋‹ค๋Š” ์ฆ๊ฑฐ๋กœ ์ปดํŒŒ์ผ๋Ÿฌ ์—†์ด ์–ด์…ˆ๋ธ”๋ฆฌ๋ฅผ ์ง์ ‘ ์ž‘์„ฑํ•˜๋ผ๊ณ  ํ•ด์•ผ ํ•˜๋Š๋ƒ๊ณ  ๋ฐ˜๋ฌธํ•จ. ์ด๋Ÿฐ ์ด์œ ๋กœ ํŠน์ • ์ž‘์—…์— AI๋ฅผ ์“ฐ์ง€ ๋ชปํ•˜๊ฒŒ ํ•˜๋Š” ๊ทœ์น™์„ ๋งŒ๋“ค์ง€ ์•Š๊ฒ ๋‹ค๊ณ  ๋ฐํž˜ news.hada.io/topic?id=35504

Linus Torvalds โ€œAI ์‚ฌ์šฉ์„ ๊ธˆ์ง€ํ•˜์ง€ ์•Š๊ฒ ๋‹ค. ์ž์‹ ์ด ํ•˜๋Š” ์ผ์„ ์ดํ•ดํ•ด์•ผ ํ•œ๋‹ค | GeekNews

Linus Torvalds๊ฐ€ Linux ์ปค๋„ ๊ธฐ์—ฌ์—์„œ ํŠน์ • ์ž‘์—…์— AI ์‚ฌ์šฉ์„ ๊ธˆ์ง€ํ•˜๋Š” ๊ทœ์น™์„ ๋งŒ๋“ค์ง€ ์•Š๊ฒ ๋‹ค๊ณ  ๋ฐํž˜. ๊ธฐ์—ฌ์ž์˜ ์—ญ๋Ÿ‰์„ ๋„๊ตฌ ์‚ฌ์šฉ ์—ฌ๋ถ€๋กœ ํŒ๋‹จํ•˜๋Š” ์ ‘๊ทผ์— ๊ฐ•ํ•˜๊ฒŒ ๋ฐ˜๋ฐœ ๊ธฐ์—ฌ์ž์—๊ฒŒ ๊ธฐ๋Œ€ํ•˜๋Š” ๊ฒƒ์€ ์ตœ์„ ์„ ๋‹คํ•˜๊ณ  ์ž์‹ ์ด ํ•˜๋Š” ์ผ์„ ์ดํ•ดํ•˜๋Š” ๊ฒƒ. ์–ด๋–ค ๋„๊ตฌ๋ฅผ ์‚ฌ์šฉํ•˜๋Š๋ƒ์™€ ์‹ค์ œ ์ดโ€ฆ

news.hada.io
19
1
41
4,095
Seapy retweeted
0xSero
@0xSeroOct 17
Deepseek is AGI I canโ€™t use anything else, either too slow or too big, or too stupid.
Teortaxesโ–ถ๏ธ (DeepSeek ๆŽจ็‰น๐Ÿ‹้“็ฒ‰ 2023 โ€“ โˆž)
@teortaxesTexOct 17
DeepSeek V4.1F-1017 has built this in like 5 minutes. (Since then it's autistically rabbitholing into minor details) These are unambiguously better than Opus's. It's a good deal stronger than V4.1.
33
32
2
1,159
44,748
๋”์ด์ƒ glm-5.3์„ ์“ธ ํ•„์š”๊ฐ€ ์—†๋„ค์š”.
DeepSeek
@deepseek_aiOct 17
๐Ÿš€ DeepSeek-V4.1-Flash-1017 is here! Stronger benchmark performance, faster task completion, and shorter outputs than the previous versionโ€”a Pareto improvement in performance and cost. Now live on our website, app, and API. ๐Ÿ› ๏ธ 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
1
5
1
68
8,748
Amused that AWS just unilaterally cancelled Cloudflare's Reinvent booth. There are always signs when someone finally realizes they're losing. Guess we'll just attend and throw the most fun side events without paying Amazon! Reply if you want an invite.
438
150
79
3,861
537,490
Seapy retweeted
Mia
@MiaAI_labOct 16
Oh oh... ๐Ÿ˜ฌ China wants to "tighten" its tech export rules. The plan covers 40 technologies. 23 are AI, included large models. Not sure when, but this could mean future Chinese open models get restricted...
68
35
18
413
31,031
Seapy retweeted
Mia
@MiaAI_labOct 16
So basically GLM 5.3 Flash beats Muse Spark Open, while being MUCH smaller? What's the point??
84
25
6
945
51,333
Seapy retweeted
DHH
@dhhOct 16
Let's fucking go.
222
128
29
8,319
205,721
muse spark open -- weights out today! best US open-source model, and runs on the new mac studio ๐ŸŽ
Mark Zuckerberg
@finkdOct 15
Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.
177
234
43
5,837
256,885
Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.
1,482
1,726
878
31,735
4,599,241
โœจ๐Ÿ‡จ๐Ÿ‡ณShenzhen, China has deployed drone docking stations on traffic light poles. These drones automatically take off during rush hours to monitor traffic, issue warnings, and return to the stations to recharge when their tasks are done.
49
457
63
1,925
1,053,644
DGX Spark owners: Don't forget to clean this occasionally. โœŒ๏ธ๐Ÿ™
34
33
8
309
63,641
Load more