Yanfei Wang @backyes1

AI/ML infra&application Zhejiang Joined February 2012

Tweets

209
Followers

12
Following

777
Likes

51

Vivek Galatage @vivekgalatage

11 months ago

Learning compilers? Check "A Compiler Writing Journey". github.com/DoctorWkt/acwj Building a compiler from the ground up, documented with each step in detail.

7 133 1K 87K 1K

View Details

There are quite a few surprising developments: 1. Intel's Nova Lake CPUs and GPUs are expected to use TSMC’s 2nm process. 2. AWS-related updates - Alchip’s (世芯) T3 chip was taped out in February and has already gone wafer-out as of May. - T4 is expected to be announced this summer, and Alchip is seen as a strong contender to win the project. - Collaborating with Arista Labs on I/O chiplets. Google TPU 3. Ironwood (TPU v7p) is scheduled to enter mass production in the first half of 2025. - Broadcom has taped out a 3nm-based v7e, MediaTek is expected to tape out the 3nm v8p in August 2025, with mass production slated for the second half of next year. 4. Meta - MTIA v3 (an ASIC version equipped with HBM) is expected to have initial shipment forecasts starting in July. - For MTIA v4, it has been confirmed that Meta plans to adopt larger packaging designs to support multiple computing dies.

駿HaYaO @QQ_Timmy

a year ago

6/17 外電整理 - 2330台積電: 美系重申正向，預期明年漲價美系大行考量匯率因素，認為每1%匯率變動，影響毛利率40bps, 所以下修下半年毛利率區間至55-56% from 58-59%,

0 0 58 12K 5

2 2 21 5K 9

View Details

Min Choi @minchoi

a year ago

MCP is going crazy viral right now🤯 AI apps can now instantly connect to any tool or live data. USB-C moment for AI. 10 wild examples:

283 1K 9K 1.4M 13K

View Details

Sam Julien @samjulien

2 years ago

Shopify’s results with HSTU architecture for shop recommendations: • 200-360% lift over baseline for merchant recommendations • 82% improvement in product recommendations • Scales with compute, similar to language models

1 2 3 1K 0

View Details

Shubham Saboo @Saboo_Shubham_

a year ago

OWL ranks 1 on the GAIA Benchmark among open-source projects with a 58.18 avg score. Try it out yourself: github.com/camel-ai/owl

6 21 186 17K 259

View Details

Yanfei Wang @backyes1

a year ago

@SumoTail @sunjiao123sun_ @awadallah R1 focus algorithm design not system design

1 0 2 22 0

View Details

Yanfei Wang @backyes1

a year ago

@AixplorerAU @sunjiao123sun_ It seems its online services got sync attack so that it can not be reachable.

0 0 0 22 0

View Details

Elon Musk @elonmusk

2 years ago

Might be one day

Rui Huang @RuiHuang_art

2 years ago

Starship 2050

1K 4K 42K 129.8M 5K

15K 32K 459K 128.7M 15K

View Details

Boris Dayma 🖍️ @borisdayma

2 years ago

Well I guess not anymore, I was scared for a moment

Boris Dayma 🖍️ @borisdayma

2 years ago

LLM's are plateauing It's been almost a week since we had real progress

3 5 39 4K 0

0 1 6 1K 0

View Details

Justine Moore @venturetwins

2 years ago

First set of demos at the @a16z AI Artist Retreat kicking off with @krea_ai and @udiomusic

11 12 94 18K 12

View Details

Horace He @cHHillee

3 years ago

Happy to OSS gpt-fast, a fast and hackable implementation of transformer inference in <1000 lines of native PyTorch with support for quantization, speculative decoding, TP, Nvidia/AMD support, and more! Code: github.com/pytorch-labs/g… Blog: pytorch.org/blog/accelerat… (1/12)

45 487 2K 477K 1K

View Details

WizardLM @WizardLM_AI

3 years ago

Sure. As we noted in the above tweet, there are two HumanEval results of GPT4 and ChatGPT-3.5: 1. The 67.0 and 48.1 are reported by the official GPT4 Report (2023/03/15) of OpenAI. 2. The 82.0 and 72.5 are tested by ourselves with the latest API (2023/08/26). We also label these scores on the Figure and Table.

5 3 75 14K 5

View Details

Yanfei Wang @backyes1

3 years ago

@omarsar0 More details on data、training method、model size、limitations？

0 0 0 91 0

View Details

elvis @omarsar0

3 years ago

This is one of the most comprehensive survey papers on LLMs! They recently updated it. It includes over 600 references, LLM collection, useful prompting tips, evaluation of capabilities, a GitHub repo, and more. paper: arxiv.org/abs/2303.18223 repo: github.com/RUCAIBox/LLMSu…

6 363 1K 163K 968

View Details

Yuandong Tian @tydsh

3 years ago

Out of 5 ICML works that I contribute substantially, 4 are accepted😃including one oral that accelerate LLM inference via contextual sparsity (thanks @BeidiChen!) and 3 on AI-guided optimization. Reviewers also like elegant solutions in rejected one, but think does not fit ICML😅

2 5 51 9K 4

View Details

Alex J. Champandard 🌻 @alexjc

4 years ago

You then have to do a couple things: - Sign-up to get the full GLM weights, 260Gb downloads quite fast. - Convert the model to int4 with convert(.)py, check it's actually 4x smaller. - Edit config/model_glm_130b_int4.sh to change path & add --from-quantized-checkpoint.

1 2 13 3K 1

View Details

Yanfei Wang @backyes1

3 years ago

@joshuabrowder Exciting idea. However if for consumer user， the financial privacy could be critical problem that must be considerated carefully.

0 0 0 162 0

View Details

Omar Pera @ompemi

3 years ago

The May 2023 modern AI tech stack. Hosting: @vercel @replit Frontend: @nextjs Backend: @fastAPI DB: @supabase Vector DB: @pinecone @trychroma LLM: @OpenAI @replicatehq @huggingface LLM dev: @langchain @fixieai LLMOps: ¯\_(ツ)_/¯ AGI: autoGPT @babyAGI_ Suggestions for June?

68 147 924 265K 1K

View Details

Geoffrey Hinton @geoffreyhinton

3 years ago

In the NYT today, Cade Metz implies that I left Google so that I could criticize Google. Actually, I left so that I could talk about the dangers of AI without considering how this impacts Google. Google has acted very responsibly.