Hacker News: serialx

Moonshot AI suspends new subscriptions due to Kimi K3 demand

serialx — Sun, 19 Jul 2026 16:02:25 +0000

Article URL: https://twitter.com/kimi_moonshot/status/2078855608565207130

Comments URL: https://news.ycombinator.com/item?id=48969291

Points: 283

# Comments: 113

Beeg float library, a Rust port of Fabrice Bellard's libbf

serialx — Sun, 05 Jul 2026 05:23:24 +0000

Article URL: https://github.com/lifthrasiir/libbeef

Comments URL: https://news.ycombinator.com/item?id=48791440

Points: 19

# Comments: 12

All-smi: Command-line utility for monitoring GPU/TPU/NPU hardwares

serialx — Sun, 05 Apr 2026 11:32:14 +0000

Article URL: https://github.com/lablup/all-smi

Comments URL: https://news.ycombinator.com/item?id=47648325

Points: 4

# Comments: 0

New comment by serialx in "Zml-smi: universal monitoring tool for GPUs, TPUs and NPUs"

serialx — Sun, 05 Apr 2026 11:30:39 +0000

Look into all-smi https://github.com/lablup/all-smi It supports all GPUs thinkable including Apple Silicon and many AI accelerator cards.

Gild Just One Lily

serialx — Mon, 08 Dec 2025 13:51:30 +0000

Article URL: https://www.smashingmagazine.com/2025/04/gild-just-one-lily/

Comments URL: https://news.ycombinator.com/item?id=46192212

Points: 38

# Comments: 7

FSM for Python, Inspired by Gen_fsm

serialx — Thu, 14 Aug 2025 07:46:37 +0000

Article URL: https://github.com/serialx/pygenfsm

Comments URL: https://news.ycombinator.com/item?id=44897827

Points: 2

# Comments: 1

New comment by serialx in "How attention sinks keep language models stable"

serialx — Fri, 08 Aug 2025 16:50:38 +0000

Yeah, attention sinks were applied to gpt-oss

Unsloth Dynamic 2.0 GGUFs

serialx — Sun, 27 Jul 2025 01:43:30 +0000

Article URL: https://docs.unsloth.ai/basics/unsloth-dynamic-2.0-ggufs

Comments URL: https://news.ycombinator.com/item?id=44698237

Points: 3

# Comments: 0

FuriosaAI RNGD – Tensor Contraction Processor

serialx — Mon, 26 May 2025 03:18:20 +0000

Article URL: https://developer.furiosa.ai/latest/en/overview/rngd.html

Comments URL: https://news.ycombinator.com/item?id=44093664

Points: 4

# Comments: 0

New comment by serialx in "TinyZero"

serialx — Sat, 25 Jan 2025 08:35:38 +0000

Ah sorry, you might be right. I meant "sparse reward" as a reward system that is mostly 0 but occasionally 1. Your "sparse reward" means only providing reward at the end of each output.

New comment by serialx in "TinyZero"

serialx — Sat, 25 Jan 2025 08:24:29 +0000

I don't think it's only using sparse rewards because of the format rewards. The training recipe is pretty comprehensive and involves multiple stages.[1] The paper mentions that when only using the RL technique, the output is often not suitable for reading. (Language mixing, etc) That feels like a AlphaZero moment for LLMs?

[1]: https://www.reddit.com/r/LocalLLaMA/comments/1i8rujw/notes_o...

New comment by serialx in "TinyZero: Reproduction of DeepSeek R1 Zero in countdown and multiplication tasks"

serialx — Sat, 25 Jan 2025 07:43:43 +0000

So to my understanding, this work reproduces DeepSeek R1's reinforcement learning mechanism in a very small language model.

The AI gets "rewards" (like points) for doing two things correctly:

Accuracy : Getting the right answer. For example, math answers must be in a specific format (e.g., inside a box) so a computer can easily check them. For coding problems, test cases verify if the code works.

Format : Using the and tags properly. This forces the AI to organize its responses clearly.

So in this case, the training program can extract the model's answer by parsing tag. We can eval the answer and evaluate if it's correct or not. If it's correct give reward, else: no reward.

Create N such answers from a single question, create N reward array. This is enough for the RL algorithm to guide the model to be more smart.

South Korea's impeached president is arrested over a martial law declaration

serialx — Sun, 19 Jan 2025 01:53:05 +0000

Article URL: https://apnews.com/article/south-korea-yoon-martial-law-arrest-rebellion-court-hearing-da68666edc4ac5ad11049362da1a84d7

Comments URL: https://news.ycombinator.com/item?id=42752954

Points: 21

# Comments: 0

PyMC-Marketing – Open-Source Marketing Analytics Solution

serialx — Tue, 03 Dec 2024 08:23:30 +0000

Article URL: https://www.pymc-marketing.io/en/stable/

Comments URL: https://news.ycombinator.com/item?id=42304065

Points: 1

# Comments: 0

New comment by serialx in "OpenWRT One Released: First Router Designed Specifically for OpenWrt"

serialx — Sun, 01 Dec 2024 06:58:53 +0000

Change the currency to USD

New comment by serialx in "The Copper Plate Must Die"

serialx — Fri, 29 Nov 2024 09:25:20 +0000

What are the compute requirements for solving efficient energy grid transmission? Is there a efficient algorithm that is able to solve this?

New comment by serialx in "Model Context Protocol"

serialx — Tue, 26 Nov 2024 02:36:53 +0000

Is there any plans to add Well-known URI[1] as a standard? It would be awesome if we can add services just by inputting domain names of the services.

[1]: https://en.wikipedia.org/wiki/Well-known_URI

Parse JSON Stream Fast for Realtime UI in LLM AI App

serialx — Sun, 10 Nov 2024 03:06:01 +0000

Article URL: https://partial.stream/

Comments URL: https://news.ycombinator.com/item?id=42098357

Points: 1

# Comments: 0

New comment by serialx in "Show HN: Pocache, preemptive optimistic caching for Go"

serialx — Fri, 11 Oct 2024 15:51:49 +0000

PSA: You can also use singleflight[1] to solve the problem. This prevents the thundering herd problem. Pocache is an interesting/alternative way to solve thundering herd indeed!

[1]: https://pkg.go.dev/golang.org/x/sync/singleflight

New comment by serialx in "GPUs Go Brrr"

serialx — Mon, 13 May 2024 12:56:03 +0000

Actually, llama.cpp running on Apple silicon uses GPU(Metal Compute Shader) to inference LLM models. Token generation is also very memory bandwidth bottlenecked. On high end Apple silicon it's about 400MB/s to 800MB/s, comparable to NVIDIA RTX 4090, which has memory bandwidth of 1000MB/s. Not to mention that Apple silicon has unified memory architecture and has high memory models (128GB, up to 192GB), which is necessary to run large LLMs like Llama 3 70B, which roughly takes 40~75GB of RAM to work reasonably.