zinccat

38 ms·

1.
by zinccat 2y ago
P40 don't support fp16 well, buy 3090 instead
2.
by zinccat 2y ago
I feel that you are mistaking the two bandwidth numbers
3.
by zinccat 2y ago
That make sense
4.
by zinccat 2y ago
I see, but still there's a lot of reasonings just for counting the letters. And ridiculous reasonings like: FedEx"? Five letters? F-E-D-E-X? No, that's six. Wait, FedEx is two syllables: Fed-Ex. Five letters? F-E-D-E-X? That&
5.
by zinccat 2y ago
My feeling is that a lot of challenge could come from the tokenizer used by the model, similar to r in strawberry problem.
6.
Prompt of SearchGPT includes the user's ip address
2 points by zinccat 2y ago|0 comments
7.
Achieving Human Level Competitive Robot Table Tennis(sites.google.com)
1 points by zinccat 2y ago|1 comments
8.
Mooncake: A KVCache-Centric Disaggregated Architecture for LLM Serving(github.com)
13 points by zinccat 2y ago|0 comments
9.
Best Papers at CVPR 2024(fxguide.com)
1 points by zinccat 2y ago|0 comments
10.
by zinccat 2y ago
previous post of llama3v on hackernews: https://news.ycombinator.com/item?id=40504827
11.
Llama3V is suspected to have been stolen from the MiniCPM-Llama3-v2.5 project(github.com)
30 points by zinccat 2y ago|7 comments
12.
Generative AI in Search: Let Google do the searching for you(blog.google)
6 points by zinccat 2y ago|0 comments
13.
DeepSeek V2, near GPT4 performance with 30% GPT3.5 cost ($0.14M input tokens)(deepseek.com)
16 points by zinccat 2y ago|2 comments
14.
SEQUOIA: Exact Llama2-70B on an RTX4090 with half-second per-token latency(infini-ai-lab.github.io)
131 points by zinccat 2y ago|61 comments
15.
by zinccat 2y ago
The author used the temperature of 0.7, which is wrong
16.
by zinccat 2y ago
I have retried the experiment with temperature=1, the result for 20 left (0.8)/right (0.2) is 17 lefts and 3 rights. I doubt why it is different from the blog.