正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Got all 12 student plus 1 instructor Raspberry Pi Pico W with pimoroni GFX pack working and running my #redis streams #micropython workshop. Is this the most GFX Packs on a single project Hel Goblins ⚡🐍 vaporweave ? 😀 event is Thursday Oxford Python

Simon Prickett 🇪🇺| simonprickett.dev

1,630 subscribers

22,128 次观看 • 3 年前 •via X (Twitter)

#redis #micropython

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

got cloud inference working! here i have smolvla running on a remote 4090 on prime intellect with a thin client running on my local pc sending observations to the remote server and receiving action chunks back. round trip latency is varying between 320-470ms! the 4090 instance I'm using is $0.62/hr and not even one of the cheaper ones, a raspberry pi v5 is < $100 and about $25 for a v3, which works fine as a client

got cloud inference working! here i have smolvla running on a remote 4090 on prime intellect with a thin client running on my local pc sending observations to the remote server and receiving action chunks back. round trip latency is varying between 320-470ms! the 4090 instance I'm using is $0.62/hr and not even one of the cheaper ones, a raspberry pi v5 is < $100 and about $25 for a v3, which works fine as a client

Jack Vial

12,992 次观看 • 7 个月前

On a different level 😤 Christian Anderson is your Big 12 Conference Most Improved Player and an All-Big 12 First Team selection. He set the new Texas Tech single-season assist record with 233 and is the only player in program history with 100-plus 3s and 200 assists. 19.1 PPG. 7.7 APG. 42.4% from three this season.

On a different level 😤 Christian Anderson is your Big 12 Conference Most Improved Player and an All-Big 12 First Team selection. He set the new Texas Tech single-season assist record with 233 and is the only player in program history with 100-plus 3s and 200 assists. 19.1 PPG. 7.7 APG. 42.4% from three this season.

Texas Tech Basketball

19,672 次观看 • 5 个月前

1,000,000 Plus streams on this song got me feeling real good over here this week!!! This world of music is a whole different animal in itself with lots of trial, error, and venerability. Accepting the beauty of the good and the bad is all in the mindset which can take time to get too but the most wholesome feeling in the world is seeing people rock with your creativity. 🥂🔊🎵

1,000,000 Plus streams on this song got me feeling real good over here this week!!! This world of music is a whole different animal in itself with lots of trial, error, and venerability. Accepting the beauty of the good and the bad is all in the mindset which can take time to get too but the most wholesome feeling in the world is seeing people rock with your creativity. 🥂🔊🎵

Marquette King

51,396 次观看 • 1 年前

THAT $70 "RUN YOUR OWN LLMS" PI KIT CAN'T RUN A SINGLE LLM. IT'S A VISION CHIP WITH NO RAM. that clip sells a raspberry pi 5 in a slick case with an ai accelerator and the caption "your own llms." clean build, fun kit. the claim is where it breaks. the fine print: the popular $70 pi ai kit uses a hailo-8l, 13 tops. it's built for vision, object detection and image processing, and it has no memory of its own. so it cannot run large language models. full stop the board that actually can is a different one: the newer ai hat+ 2, hailo-10h, 40 tops, with 8gb of dedicated ram. that's $130, not $70 and even that runs only tiny models. llama 3.2 at 1b, qwen 2.5 at 1.5b, deepseek r1 at 1.5b. edge llms live in the 1-7b range, against cloud models at 500b to 2 trillion so the honest pitch: for $130 you can run a very small language model on a pi, slowly, as a fun learning project. that's real and it's cool. "your own llms" on a $70 vision kit is not. why this keeps happening: "ai kit" and a big "tops" number sell. tops sounds like intelligence. but tops measures vision-style math, not whether the chip has the memory to hold a language model. the spec that matters for llms is ram, and the cheap kit has none. the honest caveats, both ways: the $70 kit is genuinely great, just at vision. cameras, object detection, that's its job the $130 hat really does run small llms locally, which a pi couldn't do at all two years ago. that's progress "small" is the load-bearing word. don't expect gpt at home on a pi the takeaway: before you buy a kit because the caption says llm, check two numbers. not the tops. the ram, and the size of the model it can actually load. no 70-dollar miracle, no gpt in a pi case, no tops number that means what you think. save this before you buy the wrong kit for the word on the box.

THAT $70 "RUN YOUR OWN LLMS" PI KIT CAN'T RUN A SINGLE LLM. IT'S A VISION CHIP WITH NO RAM. that clip sells a raspberry pi 5 in a slick case with an ai accelerator and the caption "your own llms." clean build, fun kit. the claim is where it breaks. the fine print: the popular $70 pi ai kit uses a hailo-8l, 13 tops. it's built for vision, object detection and image processing, and it has no memory of its own. so it cannot run large language models. full stop the board that actually can is a different one: the newer ai hat+ 2, hailo-10h, 40 tops, with 8gb of dedicated ram. that's $130, not $70 and even that runs only tiny models. llama 3.2 at 1b, qwen 2.5 at 1.5b, deepseek r1 at 1.5b. edge llms live in the 1-7b range, against cloud models at 500b to 2 trillion so the honest pitch: for $130 you can run a very small language model on a pi, slowly, as a fun learning project. that's real and it's cool. "your own llms" on a $70 vision kit is not. why this keeps happening: "ai kit" and a big "tops" number sell. tops sounds like intelligence. but tops measures vision-style math, not whether the chip has the memory to hold a language model. the spec that matters for llms is ram, and the cheap kit has none. the honest caveats, both ways: the $70 kit is genuinely great, just at vision. cameras, object detection, that's its job the $130 hat really does run small llms locally, which a pi couldn't do at all two years ago. that's progress "small" is the load-bearing word. don't expect gpt at home on a pi the takeaway: before you buy a kit because the caption says llm, check two numbers. not the tops. the ram, and the size of the model it can actually load. no 70-dollar miracle, no gpt in a pi case, no tops number that means what you think. save this before you buy the wrong kit for the word on the box.

RetroChainer

11,100 次观看 • 1 个月前

This is the PDP-11/83 I've been working on... it's got a lot of pretty das blinkenlights! It's running 211BSD UNIX that I custom compiled to make it all work. I had to tweak the kernel's idle loop to get the display panel to work, as this setup uses fast PMI (private memory interface) between the RAM and CPU, so the panel can't see it happen unless you cheat a little! The dual floppy unit is wild... a single stepper motor moves the heads on BOTH drives at the same time, so they're like conjoined drives. The bus spans three chassis, which are the three grey boxes. So unlike a PC, you can just "add most slots" and keep going to your heart's content!

This is the PDP-11/83 I've been working on... it's got a lot of pretty das blinkenlights! It's running 211BSD UNIX that I custom compiled to make it all work. I had to tweak the kernel's idle loop to get the display panel to work, as this setup uses fast PMI (private memory interface) between the RAM and CPU, so the panel can't see it happen unless you cheat a little! The dual floppy unit is wild... a single stepper motor moves the heads on BOTH drives at the same time, so they're like conjoined drives. The bus spans three chassis, which are the three grey boxes. So unlike a PC, you can just "add most slots" and keep going to your heart's content!

Dave W Plummer

43,343 次观看 • 1 年前

WINDOWS // NYC is an interwoven city that we will all build out together. There are three collections that all interconnect, windows, buildings, and the city itself. Here is exactly how this will unfold with Transient Labs on Tuesday, June 16th. 🏢🏢🏢🏢🏢 BUILDINGS are generative grids of windows, assembled through an onchain seed derived from Ethereum block data and each work’s token ID, creating distinct sizes and layouts. My collectors, friends and family, and limited partner communities can mint these BUILDINGS before public at 0.015Ξ This presale window is FCFS will last two hours (12 NOON EST to 2 PM EST) All remaining supply will go to public at 0.018Ξ after the presale ends. Supply is 1/1/4000 total buildings, each grid is completely unique, and some have very special qualities :) 🪟🪟🪟🪟🪟 Individual windows are 1/1/1000, and each mint is one single window. This is gated at first to my existing collectors, max mint 50 windows per wallet. Priced at 0.03Ξ, blind mint with instant reveal- this section is packed with gems- cats, plants, neon signs, and a bit of New York City embedded in every frame. Its these single windows that are the foundational layer of this project, since all grids are comprised of these windows. All my existing collectors get these multipliers on mint day, final snapshot is MONDAY the 15th at high noon, EST. For example: Drips // 2x DRIVE Lap 1 // 3x DRIVE Lap 2 // 2x 1/1 // 5x Editions // 1x GRAILS from DRIVE and DRIP DROP have an additional allocation on top. To ask about specific collections of mine, drop a comment here. 🌃🌃🌃🌃🌃 Additionally, the major work of this collection is the CITY, a 1/1 interconnected dynamic galaxy of light composed of all the layers below. This also comes with a large, museum quality physical lightbox of all 1,000 windows illuminated from within. This work will be placed by private sale. The CITY contract will also hold 20 NEIGHBORHOODS, our second largest arrangements of WINDOWS, and can be customized in collaboration with the artist or accepted as random grids. NEIGHBORHOODS will have 15 tokens at 1Ξ for private sale, with 5 held back for top holders of BUILDINGS. I am so excited to build this city with all of you. I've never worked so hard on anything in my life, and I am so grateful to all of you for the support so far. Let's do this thing!!!!

WINDOWS // NYC is an interwoven city that we will all build out together. There are three collections that all interconnect, windows, buildings, and the city itself. Here is exactly how this will unfold with Transient Labs on Tuesday, June 16th. 🏢🏢🏢🏢🏢 BUILDINGS are generative grids of windows, assembled through an onchain seed derived from Ethereum block data and each work’s token ID, creating distinct sizes and layouts. My collectors, friends and family, and limited partner communities can mint these BUILDINGS before public at 0.015Ξ This presale window is FCFS will last two hours (12 NOON EST to 2 PM EST) All remaining supply will go to public at 0.018Ξ after the presale ends. Supply is 1/1/4000 total buildings, each grid is completely unique, and some have very special qualities :) 🪟🪟🪟🪟🪟 Individual windows are 1/1/1000, and each mint is one single window. This is gated at first to my existing collectors, max mint 50 windows per wallet. Priced at 0.03Ξ, blind mint with instant reveal- this section is packed with gems- cats, plants, neon signs, and a bit of New York City embedded in every frame. Its these single windows that are the foundational layer of this project, since all grids are comprised of these windows. All my existing collectors get these multipliers on mint day, final snapshot is MONDAY the 15th at high noon, EST. For example: Drips // 2x DRIVE Lap 1 // 3x DRIVE Lap 2 // 2x 1/1 // 5x Editions // 1x GRAILS from DRIVE and DRIP DROP have an additional allocation on top. To ask about specific collections of mine, drop a comment here. 🌃🌃🌃🌃🌃 Additionally, the major work of this collection is the CITY, a 1/1 interconnected dynamic galaxy of light composed of all the layers below. This also comes with a large, museum quality physical lightbox of all 1,000 windows illuminated from within. This work will be placed by private sale. The CITY contract will also hold 20 NEIGHBORHOODS, our second largest arrangements of WINDOWS, and can be customized in collaboration with the artist or accepted as random grids. NEIGHBORHOODS will have 15 tokens at 1Ξ for private sale, with 5 held back for top holders of BUILDINGS. I am so excited to build this city with all of you. I've never worked so hard on anything in my life, and I am so grateful to all of you for the support so far. Let's do this thing!!!!

Dave

46,501 次观看 • 2 个月前

I am the person practice Law of Attraction many years and I do see a lot of change in my life! I believe everything can change to better after we change our heart and our mindset. On the way to Pi success journey, all our community should work together and help each other. Because we are on the same boat. We use our heart on it and we love it and it is in our life and it is our light and hope! Nobody want it fail. I read a story about a bird who has two heads share one body. But one head jealous another head and hate it. So he seduce the other head to eat a poisoned fruit. He is very happy to see the other head die. But after a few minutes, his head feel very painful and died too. I use the story to describe our poineers community. We should understand how to make Pi succes. If you hurt other poineers community, actually it will hurt your community too. I have no discrimiation on different price and respect all Pi Communities. However I would like all pioneers understand why we call GCV, it is not because ourselves greedy. It is because Pi need to start from high price to protect it on open mainnet. We all know a lot of pioneers come from the bottom of our society. They are in hardship, they even cannot afford food and place to live. And these kinds of pioneers has occupied most community. Nobody can prevent them to sell Pi becuase they are Pi holders. They need to survive, this is also the reason why there are a lot under table exchange to FIAT. Even 0.85$ or maybe 0.01$ they will sell Pi. What will happen, if the price is this low? I don't think this price can make eco develpe well. I talked with a Chinese mall rep, he told me many merchants hesitate to put their high value products on the mall is because the price big difference. If they use a price, then pioneers barter and sell in a lower price mall. It will hurt these merchants. This is the reason most merchant won't put high value products. How about the low consensus price bartered by pioneers? Yes, the merchant get a lot of Pi. However they still cannot survive if close mainnet more than 3 months-6 months. The pioneers will used up all his Pi. Then Pi has no relationship with him/her. For myself I have no Pi to use even if I have all basic coin and security circle migrated. All has been locked for 3 years. Now I changed to 90% so I only have 9 Pi in 2-3 years after almost 1 years mining. All bonus Pi cannot use now. before price united. Another thing I want to address again! No need to worry KYC! If you are real person you can pass in 5 min! Many don't want to do anything is because they don't believe they can pass KYC. You must understand Price is the root of eco. Eco healthy can make all KYC open! This is the only thing you don't need to worry. You don't pass or not get open is Because CT not ready to open to all ! #WhatIDoForPi #PiGCV

I am the person practice Law of Attraction many years and I do see a lot of change in my life! I believe everything can change to better after we change our heart and our mindset. On the way to Pi success journey, all our community should work together and help each other. Because we are on the same boat. We use our heart on it and we love it and it is in our life and it is our light and hope! Nobody want it fail. I read a story about a bird who has two heads share one body. But one head jealous another head and hate it. So he seduce the other head to eat a poisoned fruit. He is very happy to see the other head die. But after a few minutes, his head feel very painful and died too. I use the story to describe our poineers community. We should understand how to make Pi succes. If you hurt other poineers community, actually it will hurt your community too. I have no discrimiation on different price and respect all Pi Communities. However I would like all pioneers understand why we call GCV, it is not because ourselves greedy. It is because Pi need to start from high price to protect it on open mainnet. We all know a lot of pioneers come from the bottom of our society. They are in hardship, they even cannot afford food and place to live. And these kinds of pioneers has occupied most community. Nobody can prevent them to sell Pi becuase they are Pi holders. They need to survive, this is also the reason why there are a lot under table exchange to FIAT. Even 0.85$ or maybe 0.01$ they will sell Pi. What will happen, if the price is this low? I don't think this price can make eco develpe well. I talked with a Chinese mall rep, he told me many merchants hesitate to put their high value products on the mall is because the price big difference. If they use a price, then pioneers barter and sell in a lower price mall. It will hurt these merchants. This is the reason most merchant won't put high value products. How about the low consensus price bartered by pioneers? Yes, the merchant get a lot of Pi. However they still cannot survive if close mainnet more than 3 months-6 months. The pioneers will used up all his Pi. Then Pi has no relationship with him/her. For myself I have no Pi to use even if I have all basic coin and security circle migrated. All has been locked for 3 years. Now I changed to 90% so I only have 9 Pi in 2-3 years after almost 1 years mining. All bonus Pi cannot use now. before price united. Another thing I want to address again! No need to worry KYC! If you are real person you can pass in 5 min! Many don't want to do anything is because they don't believe they can pass KYC. You must understand Price is the root of eco. Eco healthy can make all KYC open! This is the only thing you don't need to worry. You don't pass or not get open is Because CT not ready to open to all ! #WhatIDoForPi #PiGCV

Doris Yin 东方紫莲🪷

44,208 次观看 • 3 年前

🧵THREAD: RICHARD HEART IS WORKING WITH ALL MAJOR PLAYERS IN ETH TO PULL OFF THE BIGGEST STUNT IN CRYPTO SINCE BITCOIN. YOU ARE NOT BULLISH ENOUGH. 1/ Eric Wall is one of the most well connected engineers in blockchain. He’s been dropping subtle clues, and even directly stating, that he’s been working with Richard Heart. In January 2025, he posted this PulseDoge picture of him and Richard imitating a scene from Scarface where Frank asks Tony to work for him. Since then, he has posted more directly about working with Richard, as evidenced below. “What would you do if I told you all the people saying bad things about HEX are actually my friends, and in on it, promoting through drama?” — Richard Heart, 2019

🧵THREAD: RICHARD HEART IS WORKING WITH ALL MAJOR PLAYERS IN ETH TO PULL OFF THE BIGGEST STUNT IN CRYPTO SINCE BITCOIN. YOU ARE NOT BULLISH ENOUGH. 1/ Eric Wall is one of the most well connected engineers in blockchain. He’s been dropping subtle clues, and even directly stating, that he’s been working with Richard Heart. In January 2025, he posted this PulseDoge picture of him and Richard imitating a scene from Scarface where Frank asks Tony to work for him. Since then, he has posted more directly about working with Richard, as evidenced below. “What would you do if I told you all the people saying bad things about HEX are actually my friends, and in on it, promoting through drama?” — Richard Heart, 2019

win

48,286 次观看 • 1 年前

Love and Deepspace | Rerun Event Preview The 5-Star Rate UP Pool [Twilight Serenity] Limited-Time Rerun will start soon! "Then, you're not allowed to change your mind even after a hundred years. Or a thousand." 💫Event Duration: From 05:00 on Dec. 11 to 04:59 on Dec. 18 (Server Time) 💫The 5-Star Rate UP Pool [Twilight Serenity] Limited-Time Rerun Event 1. During the event, make a wish with [Deepspace Wish] or [Time Wish: Limited] to participate in the wish event. The drop rate of the event-limited 5-Star Memory [Rafayel: Fireworks Vow] will go up drastically. 2. After the event ends, this limited 5-Star Memory will not be obtainable through other means and will not enter the permanent Wish Pool: Xspace Echo. 3. All Rerun Wish Pools share one pity system. A 5-Star Memory is guaranteed within a specific attempt of wishes. If the 5-Star Memory you have obtained from the Rerun Wish Pool is not the event-limited Memory, you will obtain the event-limited Memory the next time you obtain a 5-Star Memory. The pity count from the last Rerun Wish Pool can be applied to this Rerun Wish Pool, and the pity count in this Rerun Wish Pool will also be applied to the upcoming Rerun Wish Pool. *You can read more about the event on the in-game rules page. 🎁New Packs During the event, the Rerun event-exclusive [Flamebloom Pack] series, which includes [Time Wish: Limited] and other materials, will be available in Shop. Notes: 1. [Time Wish: Limited] can be used in 5-Star Memory Wish Pool Rerun and will be used first when you make a wish. 2. After the event ends, [Time Wish: Limited] will automatically convert to Empyrean Wish. ——— 🪐Official Discord: #LoveandDeepspace #Rafayel

Love and Deepspace | Rerun Event Preview The 5-Star Rate UP Pool [Twilight Serenity] Limited-Time Rerun will start soon! "Then, you're not allowed to change your mind even after a hundred years. Or a thousand." 💫Event Duration: From 05:00 on Dec. 11 to 04:59 on Dec. 18 (Server Time) 💫The 5-Star Rate UP Pool [Twilight Serenity] Limited-Time Rerun Event 1. During the event, make a wish with [Deepspace Wish] or [Time Wish: Limited] to participate in the wish event. The drop rate of the event-limited 5-Star Memory [Rafayel: Fireworks Vow] will go up drastically. 2. After the event ends, this limited 5-Star Memory will not be obtainable through other means and will not enter the permanent Wish Pool: Xspace Echo. 3. All Rerun Wish Pools share one pity system. A 5-Star Memory is guaranteed within a specific attempt of wishes. If the 5-Star Memory you have obtained from the Rerun Wish Pool is not the event-limited Memory, you will obtain the event-limited Memory the next time you obtain a 5-Star Memory. The pity count from the last Rerun Wish Pool can be applied to this Rerun Wish Pool, and the pity count in this Rerun Wish Pool will also be applied to the upcoming Rerun Wish Pool. *You can read more about the event on the in-game rules page. 🎁New Packs During the event, the Rerun event-exclusive [Flamebloom Pack] series, which includes [Time Wish: Limited] and other materials, will be available in Shop. Notes: 1. [Time Wish: Limited] can be used in 5-Star Memory Wish Pool Rerun and will be used first when you make a wish. 2. After the event ends, [Time Wish: Limited] will automatically convert to Empyrean Wish. ——— 🪐Official Discord: #LoveandDeepspace #Rafayel

Love and Deepspace

304,066 次观看 • 8 个月前

Claude Code Desktop now opens a new window for each session This makes it much easier to visualize multiple Claude Code agents running in parallel My current stack depends on the task: - Ghostty: when starting a project. Bash commands, git, env variables, provider connections. All manual through the terminal with a Claude panel running alongside. - Claude Code Desktop: once everything is configured. GitHub connected, CLAUDE.md, Skills, subagents and Hooks ready. Claude Code runs on its own, no more terminal setup, just panels running and outputs to review. - VSCode: when I need to review code by hand. I use it less and less, but there are moments where I have to confirm Claude got it right. I usually open the Claude extension inside VSCode, but it lacks most of the CLI features so it's limited Solid update. Worth trying once your workflows are already set up 👇

Claude Code Desktop now opens a new window for each session This makes it much easier to visualize multiple Claude Code agents running in parallel My current stack depends on the task: - Ghostty: when starting a project. Bash commands, git, env variables, provider connections. All manual through the terminal with a Claude panel running alongside. - Claude Code Desktop: once everything is configured. GitHub connected, CLAUDE.md, Skills, subagents and Hooks ready. Claude Code runs on its own, no more terminal setup, just panels running and outputs to review. - VSCode: when I need to review code by hand. I use it less and less, but there are moments where I have to confirm Claude got it right. I usually open the Claude extension inside VSCode, but it lacks most of the CLI features so it's limited Solid update. Worth trying once your workflows are already set up 👇

Daniel San

38,806 次观看 • 3 个月前

🚨 Anthropic committed up to 1M TPU chips for Claude. Openai is leasing TPUs for chatgpt inference. Here's How kernels work on TPUs (deep dive 2/6 by emi) pallas is Google's answer to kernel writing. a python kernel SDK built on JAX. still very experimental (jax.experimental.pallas). on TPU it compiles through mosaic; on GPU it lowers to triton. if you know CUDA, the syntax will feel familiar but the execution model is completely different. in CUDA, grid=(4,4) launches 16 blocks running simultaneously across SMs. in pallas, those 16 iterations run one after another in lexicographic order. no threads. no warps. no blocks. no occupancy tuning. a TPU is a sequential machine with a very wide vector register — more like a CPU than a GPU. performance comes from width: a 128x128 systolic array doing matmul and an 8x128 SIMD vector unit doing everything else. maximum parallelism on chip: 2, one per TensorCore in megacore mode. three concepts replace CUDA's thread/block/grid hierarchy. Refs are mutable memory references. because execution is sequential, each iteration safely accumulates without atomics. in CUDA you'd need atomics or a separate reduction pass. the memory model is also very different from NVIDIA's. zero hardware caches. VMEM is 32-128 MiB of software-managed scratchpad — 500-1000x larger than GPU shared memory per SM. all data must be explicitly DMA'd from HBM to VMEM before any computation touches it. four levels: HBM → VMEM → VREGs → MXU/VPU, plus SMEM for scalar control data. every byte of data movement is your responsibility. this is like CUDA shared memory except it's 500x bigger and there's no cache fallback. pipelining is mandatory. without double-buffering HBM→VMEM transfers, the MXU just stalls waiting for data. this is the single most important optimization on TPU. and because grid execution is sequential and deterministic, consecutive iterations that need the same input block skip the redundant HBM transfer automatically, impossible on GPU where block execution order is undefined. the compilation pipeline is unlike anything in this series: python → jaxpr → stableHLO → XLA HLO (71+ optimization passes) → LLO (78+ passes) → 322-bit VLIW bundles. the compiler packs instructions for scalar, vector, matrix, and DMA units into a single 322-bit word. everything in that bundle executes in parallel, with no runtime scheduling.

🚨 Anthropic committed up to 1M TPU chips for Claude. Openai is leasing TPUs for chatgpt inference. Here's How kernels work on TPUs (deep dive 2/6 by emi) pallas is Google's answer to kernel writing. a python kernel SDK built on JAX. still very experimental (jax.experimental.pallas). on TPU it compiles through mosaic; on GPU it lowers to triton. if you know CUDA, the syntax will feel familiar but the execution model is completely different. in CUDA, grid=(4,4) launches 16 blocks running simultaneously across SMs. in pallas, those 16 iterations run one after another in lexicographic order. no threads. no warps. no blocks. no occupancy tuning. a TPU is a sequential machine with a very wide vector register — more like a CPU than a GPU. performance comes from width: a 128x128 systolic array doing matmul and an 8x128 SIMD vector unit doing everything else. maximum parallelism on chip: 2, one per TensorCore in megacore mode. three concepts replace CUDA's thread/block/grid hierarchy. Refs are mutable memory references. because execution is sequential, each iteration safely accumulates without atomics. in CUDA you'd need atomics or a separate reduction pass. the memory model is also very different from NVIDIA's. zero hardware caches. VMEM is 32-128 MiB of software-managed scratchpad — 500-1000x larger than GPU shared memory per SM. all data must be explicitly DMA'd from HBM to VMEM before any computation touches it. four levels: HBM → VMEM → VREGs → MXU/VPU, plus SMEM for scalar control data. every byte of data movement is your responsibility. this is like CUDA shared memory except it's 500x bigger and there's no cache fallback. pipelining is mandatory. without double-buffering HBM→VMEM transfers, the MXU just stalls waiting for data. this is the single most important optimization on TPU. and because grid execution is sequential and deterministic, consecutive iterations that need the same input block skip the redundant HBM transfer automatically, impossible on GPU where block execution order is undefined. the compilation pipeline is unlike anything in this series: python → jaxpr → stableHLO → XLA HLO (71+ optimization passes) → LLO (78+ passes) → 322-bit VLIW bundles. the compiler packs instructions for scalar, vector, matrix, and DMA units into a single 322-bit word. everything in that bundle executes in parallel, with no runtime scheduling.

wafer

33,134 次观看 • 1 个月前

#LindaReynolds "It's all about me" followed by stomp, yell, sob, squeal. Linda 'I sue ' Reynolds is all pissed that she didn't get a mention in the film 'Silenced' From The Australian: In a statement to the Australian on Thursday, Ms Reynolds criticised Silenced for erasing a “critical piece” of her history. "The filmmakers of Silenced said they “respectfully disagree with the claims made by Ms Reynolds about our film”. “Silenced is a film that demonstrates how the law has been used by alleged perpetrators to silence survivors of gender-based violence and the journalists who report on these cases,” they said." Linda, you got a verdict, you bankrupted Brittany Higgins, and you STILL want it ALL TO BE ABOUT YOU. Just fuck off. This is the most incredible part: Reynolds “It cost me my job, my health, my reputation and the Liberal Party the election.” Linda, it didn't cost you your job; you cost yourself your job. You're just using that as an excuse. Furthermore, this didn't lose liberals the election; the fucking liberals did because they're all just like you. JUST. GO. AWAY. Off you fuck.

#LindaReynolds "It's all about me" followed by stomp, yell, sob, squeal. Linda 'I sue ' Reynolds is all pissed that she didn't get a mention in the film 'Silenced' From The Australian: In a statement to the Australian on Thursday, Ms Reynolds criticised Silenced for erasing a “critical piece” of her history. "The filmmakers of Silenced said they “respectfully disagree with the claims made by Ms Reynolds about our film”. “Silenced is a film that demonstrates how the law has been used by alleged perpetrators to silence survivors of gender-based violence and the journalists who report on these cases,” they said." Linda, you got a verdict, you bankrupted Brittany Higgins, and you STILL want it ALL TO BE ABOUT YOU. Just fuck off. This is the most incredible part: Reynolds “It cost me my job, my health, my reputation and the Liberal Party the election.” Linda, it didn't cost you your job; you cost yourself your job. You're just using that as an excuse. Furthermore, this didn't lose liberals the election; the fucking liberals did because they're all just like you. JUST. GO. AWAY. Off you fuck.

The Noisy Elephant

20,458 次观看 • 2 个月前

The ATG School One-Pager I’m not trying to reinvent schooling. There are just a handful of things I believe in which I haven’t seen in any school I’ve been around as a student or parent. Policy #1: Each student gets to be responsible for growing some of their own food, no matter how small, and THROUGHOUT schooling (not just a quickie project here or there). Policy #2: Minimum 1:1 ratio of time NOT SITTING IN THE CLASSROOM. What you do with this is up to you. There are so many real world skills, sports, gardening, music, etc. The strict ratio in the school day is the key for me. Common sense and personal interests can take it from there. Policy #3: Daily time to read whatever you want to read about. The biggest barrier for my reading was INTEREST. Be there to ensure the book is at their level, and to help them if they don’t understand something. Other than that, LET THEM ENJOY READING, ALL THE WAY THROUGH SCHOOL, not just in early years. Policy #4: (This is the most unusual yet the biggest reason I’m in education.) High school is a 50/50 bridge to winning in real life. Mornings are for actual work, making and SAVING UP MONEY. Afternoons are for learning finances and professional skills of YOUR INTEREST. With average work, you’ll finish school with $50,000-$100,000 in the bank, more skills than the norm, and a greater chance of creating your life and work from there on out, rather than conforming to make a paycheck. Policy #5: As part of the high school 50/50 system, ensure each student learns the adult financial red tape in your state/country before you’ve got bills, kids, etc.

The ATG School One-Pager I’m not trying to reinvent schooling. There are just a handful of things I believe in which I haven’t seen in any school I’ve been around as a student or parent. Policy #1: Each student gets to be responsible for growing some of their own food, no matter how small, and THROUGHOUT schooling (not just a quickie project here or there). Policy #2: Minimum 1:1 ratio of time NOT SITTING IN THE CLASSROOM. What you do with this is up to you. There are so many real world skills, sports, gardening, music, etc. The strict ratio in the school day is the key for me. Common sense and personal interests can take it from there. Policy #3: Daily time to read whatever you want to read about. The biggest barrier for my reading was INTEREST. Be there to ensure the book is at their level, and to help them if they don’t understand something. Other than that, LET THEM ENJOY READING, ALL THE WAY THROUGH SCHOOL, not just in early years. Policy #4: (This is the most unusual yet the biggest reason I’m in education.) High school is a 50/50 bridge to winning in real life. Mornings are for actual work, making and SAVING UP MONEY. Afternoons are for learning finances and professional skills of YOUR INTEREST. With average work, you’ll finish school with $50,000-$100,000 in the bank, more skills than the norm, and a greater chance of creating your life and work from there on out, rather than conforming to make a paycheck. Policy #5: As part of the high school 50/50 system, ensure each student learns the adult financial red tape in your state/country before you’ve got bills, kids, etc.

KneeOverToesGuy

31,525 次观看 • 5 个月前

So why we need open mainnet and when it's the good time to open and can we make it faster? I think Yes! WE build up Pi public blockchain is like we are building a highway. When the highway complete, all drivers need this highway because it can bring them benefit. And the highway can charge them fees every time when they pass. Pi is similar like a highway. The drivers are outside eco and different organization. We all know Pi WEB 3.0 blockchain has great value. This will attract outside eco come. When they come, they don't have Pi to pledge to use our blockchain. So they need to use their FIAT to buy our Pi. However, if our inner community has difference on price, it's like the highway has been controlled by different groups. These groups charge different fees. It will confuse outside eco and make them hesitate to come. That's why I believe open mainnet need united price for outside eco. Of course united price can help inner eco develop grow fast too. There are two options regarding price: 1. Let the pioneers and merchants slowly reach the agreement and we do nothing . This is let it go naturally. I do know some pioneers advocate this way. But I am sure it will take long journey because more than 45 million pioneers or maybe will never reach agreement. They will use from $0.01 to 1 million $ different price. 2. Our pioneers community advocate GCV because it has been accepted by a lot pioneers and merchants already. If all pioneers can understand this GCV price logic and if they hold Pi, they will support it. The reason is so simple. It relate to their self interest. This can be done very fast if we all united together. Another reason is based on a business accounting system. All business has tax reporting system even crypto currency. From 2019, USA IRS has the tax regulation on individual crypto currency capital gain. Currently most merchants are pioneers, they use their individual Pi Wallet. I see recently there is developer/merchant business wallet come out. Then they should use business Pi wallet. So it's easier for them to track crypto business profit and loss. As an accountant I know all capital gain or loss are calculated by using USD in USA. So accountant need a unified price. If not, the company will hesitate to join Pi eco system. Based the above reason, i call on Pi Community to support GCV price so that eco can satisfied prosperous condition to open mainnet. As for KYC and migration, the technology part has been mature or almost mature. And we believe CT and all pioneers can follow very fast. During close mainnet all merchants should do barter according to your availability. Please don't put all your products now! Close mainnet period is for cultivate the Pi Payment environment! If you have inventory and small items, you can put maybe 10% or less of your products to barter. All business owners should realize before opening mainnet there is risk of loss. Why many Chinese merchant do it? Because they are pioneers too ! They want to realize Pi GCV price so they do it volunteering to use inventory and small items with limit supply. Therefore it won't have much risk. And if open mainnet can reach GCV, they will not incur loss. So please all merchants do it wisely. Close mainnet small items and use inventory. Of course if you want to put big items, this is your choice and pioneers will be appreciated. But my suggestion is to be Rationalism. If everyone can do a little, it will create big influence. As a pioneer close mainnet it's not a time to get profit for yourself. So please don't expect cross border or big items because merchant has business risk. I see some cross border barter incur which is based on individual merchant. When open mainnet, you can realize all of it. But now it's time to help GCV price as a trend from my opinion. The above analysis is only my personal analysis. Reference only. It is not Business Advice! #WhatIDoForPi #PiGCV

So why we need open mainnet and when it's the good time to open and can we make it faster? I think Yes! WE build up Pi public blockchain is like we are building a highway. When the highway complete, all drivers need this highway because it can bring them benefit. And the highway can charge them fees every time when they pass. Pi is similar like a highway. The drivers are outside eco and different organization. We all know Pi WEB 3.0 blockchain has great value. This will attract outside eco come. When they come, they don't have Pi to pledge to use our blockchain. So they need to use their FIAT to buy our Pi. However, if our inner community has difference on price, it's like the highway has been controlled by different groups. These groups charge different fees. It will confuse outside eco and make them hesitate to come. That's why I believe open mainnet need united price for outside eco. Of course united price can help inner eco develop grow fast too. There are two options regarding price: 1. Let the pioneers and merchants slowly reach the agreement and we do nothing . This is let it go naturally. I do know some pioneers advocate this way. But I am sure it will take long journey because more than 45 million pioneers or maybe will never reach agreement. They will use from $0.01 to 1 million $ different price. 2. Our pioneers community advocate GCV because it has been accepted by a lot pioneers and merchants already. If all pioneers can understand this GCV price logic and if they hold Pi, they will support it. The reason is so simple. It relate to their self interest. This can be done very fast if we all united together. Another reason is based on a business accounting system. All business has tax reporting system even crypto currency. From 2019, USA IRS has the tax regulation on individual crypto currency capital gain. Currently most merchants are pioneers, they use their individual Pi Wallet. I see recently there is developer/merchant business wallet come out. Then they should use business Pi wallet. So it's easier for them to track crypto business profit and loss. As an accountant I know all capital gain or loss are calculated by using USD in USA. So accountant need a unified price. If not, the company will hesitate to join Pi eco system. Based the above reason, i call on Pi Community to support GCV price so that eco can satisfied prosperous condition to open mainnet. As for KYC and migration, the technology part has been mature or almost mature. And we believe CT and all pioneers can follow very fast. During close mainnet all merchants should do barter according to your availability. Please don't put all your products now! Close mainnet period is for cultivate the Pi Payment environment! If you have inventory and small items, you can put maybe 10% or less of your products to barter. All business owners should realize before opening mainnet there is risk of loss. Why many Chinese merchant do it? Because they are pioneers too ! They want to realize Pi GCV price so they do it volunteering to use inventory and small items with limit supply. Therefore it won't have much risk. And if open mainnet can reach GCV, they will not incur loss. So please all merchants do it wisely. Close mainnet small items and use inventory. Of course if you want to put big items, this is your choice and pioneers will be appreciated. But my suggestion is to be Rationalism. If everyone can do a little, it will create big influence. As a pioneer close mainnet it's not a time to get profit for yourself. So please don't expect cross border or big items because merchant has business risk. I see some cross border barter incur which is based on individual merchant. When open mainnet, you can realize all of it. But now it's time to help GCV price as a trend from my opinion. The above analysis is only my personal analysis. Reference only. It is not Business Advice! #WhatIDoForPi #PiGCV

Doris Yin 东方紫莲🪷

28,979 次观看 • 3 年前

We just launched our biggest creator partnership to date. We're excited about it. It's done really well so far. A few of our big bets for this year and beyond are: - creator obsession - partnerships - In real life community activations - sampling - overall brand building Something that is definitely newer to us, and we're definitely late getting into, but the best place to start is today. We launched a product with a well-known creator last week. That has gone really well and will unfortunately be out of stock on a bunch of shades soon. We did an in-person event with her at, 8 months pregnant today at one of our stores. It was our first time ever hanging blister packs to sample. We had a bunch of blister packs and a bunch of agile trucks, plus some other out-of-home. It's all a big content play, but also building community and introing people to the brand and getting samples in as many people's hands as possible. We just bought a rivian and wrapped it for sampling events. Hoping to give away 7 figures in samples this year and next. We started with 1 but I’m gonna pull the trigger on a few more once we see signs of traction. We’re opening 12 stores this year and doubling down on IRL, community, and sampling. We have another brand partnership in a few weeks with a big IRL component as well. Excited to learn new things and keep pushing into building a brand, community, and meeting people IRL.

Cody Plofker

16,457 次观看 • 4 个月前

Had an awesome time with Mark Haywood and the $edm.v team on the site tour! The DFO permit is a near term major catalyst, which upon receiving will be followed by the final production decision. The water at the polishing pond after the tailings pond is astoundingly clean and loaded with fish, and tested as much cleaner than the water of the local streams and rivers. DFO permits are very rarely not granted, and $EDM.v's thoroughness and hard work is sure to be rewarded. The provincial government of Nova Scotia is also very supportive of the project and is eager to see them enter production. EDM's cash flow for next year will likely be greater than the current mcap, only to continue ramping up to double or greater than the current mcap. Very few juniors are this close to cash flow with such a small capex needed for restart, which will be financed with debt and paid back in around half a year. $edm.v should be 3-4x higher now, at 0.5x NPV, and even higher once the DFO permit is received this summer, 0.7-0.8x NPV. Right now we are trading near 0.10-0.15x NPV when you adjust for the DMS update, current metal prices, as well as doubled Gypsum prices, which is crazy cheap for how advanced the project is. Another impression from the tour is how solid and well made the 2700tpd mill is. It was built by Imperial Oil, a multi billion dollar company, who put in a lot of extras. There are about 5-10m CAD in replacement and maintainence parts on site from the previous opperators, still in original packaging. All of this is just the base case projection, it doesn't include the $gold optionality. Once the mine is in production and the main put is drained, they can go back to where the last producers will pulling ore and resuming mining high grade gold, which left off at 20k to 60k ounces a year based on a rough back calculation from the left over concentrates. This gold is proven via production and recovery, its absolutely there, and the team is actively working on defining the extent and quantities of it. I have been adding to my position this week $zinc $lead $gold $silver $gypsum $emo.v $fwz.v $teck $nexa $glen

Had an awesome time with Mark Haywood and the $edm.v team on the site tour! The DFO permit is a near term major catalyst, which upon receiving will be followed by the final production decision. The water at the polishing pond after the tailings pond is astoundingly clean and loaded with fish, and tested as much cleaner than the water of the local streams and rivers. DFO permits are very rarely not granted, and $EDM.v's thoroughness and hard work is sure to be rewarded. The provincial government of Nova Scotia is also very supportive of the project and is eager to see them enter production. EDM's cash flow for next year will likely be greater than the current mcap, only to continue ramping up to double or greater than the current mcap. Very few juniors are this close to cash flow with such a small capex needed for restart, which will be financed with debt and paid back in around half a year. $edm.v should be 3-4x higher now, at 0.5x NPV, and even higher once the DFO permit is received this summer, 0.7-0.8x NPV. Right now we are trading near 0.10-0.15x NPV when you adjust for the DMS update, current metal prices, as well as doubled Gypsum prices, which is crazy cheap for how advanced the project is. Another impression from the tour is how solid and well made the 2700tpd mill is. It was built by Imperial Oil, a multi billion dollar company, who put in a lot of extras. There are about 5-10m CAD in replacement and maintainence parts on site from the previous opperators, still in original packaging. All of this is just the base case projection, it doesn't include the $gold optionality. Once the mine is in production and the main put is drained, they can go back to where the last producers will pulling ore and resuming mining high grade gold, which left off at 20k to 60k ounces a year based on a rough back calculation from the left over concentrates. This gold is proven via production and recovery, its absolutely there, and the team is actively working on defining the extent and quantities of it. I have been adding to my position this week $zinc $lead $gold $silver $gypsum $emo.v $fwz.v $teck $nexa $glen

Zachariah Loney

12,670 次观看 • 2 个月前

hey if you have a 3060, or any GPU with 8GB or more sitting in a drawer right now, that thing can run 9 billion parameters of intelligence autonomously. and you don't know it yet. 2 hours ago i posted that 9B hit a ceiling. 2,699 lines across 11 files. blank screen. said the limit for autonomous multifile coding on 9 billion parameters is real. then i audited every file. found 11 bugs. exact file, exact line, exact fix. duplicate variable declarations killing the script loader. a canvas reference never connected to the DOM. enemies with no movement logic. particle systems called on the class instead of the instance. fed that list as a single prompt to the same Qwen 3.5 9B on the same RTX 3060 through Hermes Agent. it fixed all 11. surgically. patch level edits across 4 files. no rewrites. no hallucinated changes. game boots. enemies spawn, move, collide. background renders. particles fire. and here's what nobody is talking about. this is a 9 billion parameter model running a full agentic framework. Hermes Agent with 31 tools. file operations, terminal, browser, code execution. not a single tool call failed. the agent chain never broke. most people think you need 70B+ for reliable tool use. this is 9B on 12 gigs doing it clean. the model didn't fail. my prompting strategy did. the ceiling is not the parameter count. the ceiling is how you prompt it. this is not done. bullets don't fire yet. boss fights need wiring. but the screen that was black 2 hours ago now has a full game rendering in real time. iterating right now. anyone with a GPU from the last 5 years should be paying attention to what is happening right now.

hey if you have a 3060, or any GPU with 8GB or more sitting in a drawer right now, that thing can run 9 billion parameters of intelligence autonomously. and you don't know it yet. 2 hours ago i posted that 9B hit a ceiling. 2,699 lines across 11 files. blank screen. said the limit for autonomous multifile coding on 9 billion parameters is real. then i audited every file. found 11 bugs. exact file, exact line, exact fix. duplicate variable declarations killing the script loader. a canvas reference never connected to the DOM. enemies with no movement logic. particle systems called on the class instead of the instance. fed that list as a single prompt to the same Qwen 3.5 9B on the same RTX 3060 through Hermes Agent. it fixed all 11. surgically. patch level edits across 4 files. no rewrites. no hallucinated changes. game boots. enemies spawn, move, collide. background renders. particles fire. and here's what nobody is talking about. this is a 9 billion parameter model running a full agentic framework. Hermes Agent with 31 tools. file operations, terminal, browser, code execution. not a single tool call failed. the agent chain never broke. most people think you need 70B+ for reliable tool use. this is 9B on 12 gigs doing it clean. the model didn't fail. my prompting strategy did. the ceiling is not the parameter count. the ceiling is how you prompt it. this is not done. bullets don't fire yet. boss fights need wiring. but the screen that was black 2 hours ago now has a full game rendering in real time. iterating right now. anyone with a GPU from the last 5 years should be paying attention to what is happening right now.

Sudo su

684,336 次观看 • 5 个月前

This is Virgil….. he was sat down, by himself at Anfield. One was filmed by me and the other by my son. 1 of Robbos kids was running up to him and interacting. He looked Somber if anything but then so was Anfield today for the most part . 2 legends leaving the club at the end of the season after losing one totally at the beginning -RIP Jota 🕊️ It’s not all on Virgil, as I’ve read, he’s done the most mins then others his age. He also will likely go through his last season with his mins being managed. They’re a team not singular players. So back them all regardless of favourites. It’s gonna be a bumpy ride….

This is Virgil….. he was sat down, by himself at Anfield. One was filmed by me and the other by my son. 1 of Robbos kids was running up to him and interacting. He looked Somber if anything but then so was Anfield today for the most part . 2 legends leaving the club at the end of the season after losing one totally at the beginning -RIP Jota 🕊️ It’s not all on Virgil, as I’ve read, he’s done the most mins then others his age. He also will likely go through his last season with his mins being managed. They’re a team not singular players. So back them all regardless of favourites. It’s gonna be a bumpy ride….

Sian ﾒ𝟶 ᥫ᭡

28,671 次观看 • 3 个月前

On the leg between Guam (PGUM) and Kalaeloa (PHJR) we experienced a fault with one of our 2 AC packs. This left us a single point of failure away from de-pressurization, which would require a descent to an altitude where we could breathe without supplemental O2. Keep in mind, a lower altitude drastically increases fuel consumption. Of course a single pack can maintain cabin pressure all the way up to the aircraft’s service ceiling, but if that pack fails, the aircraft cannot be pressurized. The situation required that Bob Allen & I make a plan as to how to handle the potential loss of our remaining pack. This plan would be dynamic, as the action taken would depend where we were at the time of de-pressurization. Past the ETP (Equal Time Point) turning back to GUM was not an option; and continuing to JRF at a low cabin altitude was ALSO not an option - as we’d be short on fuel. PKMJ was a bit too far south to be of use, but PWAK (Wake Island) and PMDY (Midway) were valid alternates that we could use for diversion and still land with the engines turning. Even though we did have ETOPS planning info on our flight plan, we can’t always rely on it 100% because ferry flying creates some unique challenges due to the fact that we don’t operate the same aircraft all the time, and we have no operational history to set the burn bias. We need to be proactive in the cockpit so on the secondary FPL page of the MCDU, Bob and I played out a few scenarios at different points and compared fuel burns. We decided on a boundary where once crossed, the diversion alternate switched from WAK to MDY. In either case - we’d land with fuel, but it would be pretty minimal. Passing 170W would be the most critical point. The options would be MDY or LIH from there and both would land with under 800 KG of fuel. Because that didn’t inspire much confidence, we took some additional measures and brought 2 POBs (portable O2 bottles) up to the flight deck. This way, we’d have both the crew O2 from the pressure demand bottle (about 30-40 mins estimated) plus the O2 from POBs available to remain a bit higher longer in the event of a depress. Staying higher meant saving fuel and having more options - always a good thing. Anyway, the operational pack remained working, and we were able to get the aircraft to JRF before getting the bad pack back online for the next leg - so crisis averted, but the moral of the story is: you always need to have a plan! Especially when flying over remote areas of the planet..

On the leg between Guam (PGUM) and Kalaeloa (PHJR) we experienced a fault with one of our 2 AC packs. This left us a single point of failure away from de-pressurization, which would require a descent to an altitude where we could breathe without supplemental O2. Keep in mind, a lower altitude drastically increases fuel consumption. Of course a single pack can maintain cabin pressure all the way up to the aircraft’s service ceiling, but if that pack fails, the aircraft cannot be pressurized. The situation required that Bob Allen & I make a plan as to how to handle the potential loss of our remaining pack. This plan would be dynamic, as the action taken would depend where we were at the time of de-pressurization. Past the ETP (Equal Time Point) turning back to GUM was not an option; and continuing to JRF at a low cabin altitude was ALSO not an option - as we’d be short on fuel. PKMJ was a bit too far south to be of use, but PWAK (Wake Island) and PMDY (Midway) were valid alternates that we could use for diversion and still land with the engines turning. Even though we did have ETOPS planning info on our flight plan, we can’t always rely on it 100% because ferry flying creates some unique challenges due to the fact that we don’t operate the same aircraft all the time, and we have no operational history to set the burn bias. We need to be proactive in the cockpit so on the secondary FPL page of the MCDU, Bob and I played out a few scenarios at different points and compared fuel burns. We decided on a boundary where once crossed, the diversion alternate switched from WAK to MDY. In either case - we’d land with fuel, but it would be pretty minimal. Passing 170W would be the most critical point. The options would be MDY or LIH from there and both would land with under 800 KG of fuel. Because that didn’t inspire much confidence, we took some additional measures and brought 2 POBs (portable O2 bottles) up to the flight deck. This way, we’d have both the crew O2 from the pressure demand bottle (about 30-40 mins estimated) plus the O2 from POBs available to remain a bit higher longer in the event of a depress. Staying higher meant saving fuel and having more options - always a good thing. Anyway, the operational pack remained working, and we were able to get the aircraft to JRF before getting the bad pack back online for the next leg - so crisis averted, but the moral of the story is: you always need to have a plan! Especially when flying over remote areas of the planet..

Steve Giordano

1,466,610 次观看 • 3 年前