helloericsf

6 ms·

31.
by helloericsf 2y ago
True. More benchmark metrics here: https://x.com/deepseek_ai/status/1872242657348710721/photo/2
32.
by helloericsf 2y ago
HF link: https://huggingface.co/deepseek-ai/DeepSeek-V3 Aider link: https://aider.chat/docs/leaderboards/ Pricing($0.14/$0.28 per 1M tokens) reference: https://x.com/xingy
33.
DeepSeek v3 beats Claude sonnet 3.5 and way cheaper(huggingface.co)
48 points by helloericsf 2y ago|9 comments
34.
by helloericsf 2y ago
Tweets from Chinese Scholars call for investigation: - https://x.com/xiaoyongzhu001/status/1867724288369004716 - https://x.com/tydsh/status/1867965300706255060 - https://x.com
35.
by helloericsf 2y ago
NeurIPS: - https://x.com/NeurIPSConf/status/1867759121023336464 - https://neurips.cc/Conferences/2024/StatementOnInclusivity Picard: - https://www.media.mit.edu/posts&#
36.
NeurIPS and Dr. Picard released statement for singling out Chinese scholars(twitter.com)
2 points by helloericsf 2y ago|2 comments
37.
by helloericsf 2y ago
Call Jensen and Lisa!lol
38.
Tencent Hunyuan-Large(github.com)
148 points by helloericsf 2y ago|103 comments
39.
by helloericsf 2y ago
- 389 billion parameters and 52 billion activation parameters, capable of handling up to 256K tokens. - outperforms LLama3.1-70B and exhibits comparable performance when compared to the significantly larger LLama3.1-405B model.
40.
by helloericsf 2y ago
Thank you! Great insight.
41.
by helloericsf 2y ago
Congrats on the launch! Curious to know, which OSS models you see works best at the moment?
42.
by helloericsf 2y ago
Credit to https://x.com/calebfahlgren https://huggingface.co/spaces/cfahlgren1/model-release-heatm...
43.
Chinese AI Community: open-source Heatmap(huggingface.co)
1 points by helloericsf 2y ago|1 comments
44.
by helloericsf 2y ago
OpenCL was discussed more frequently in classes about a decade ago. However, I haven't heard it mentioned in the last five years or so.
45.
by helloericsf 2y ago
How many coding co-pilot we have on the market?
46.
Poolside is raising $400M+ at a $2B valuation to build a coding co-pilot(techcrunch.com)
3 points by helloericsf 2y ago|1 comments
47.
by helloericsf 2y ago
I don't think github is using cloudflare.
48.
by helloericsf 2y ago
4-bit quantization tends to come at the cost of output quality losses. https://github.com/ggerganov/llama.cpp/issues/9
49.
by helloericsf 2y ago
Personally, I never seen onnx used for LLM.
50.
by helloericsf 2y ago
Seems interesting! https://github.com/turboderp/exllama "A more memory-efficient rewrite of the HF transformers implementation of Llama for use with quantized weights."
51.
Is LMDeploy the Ultimate Solution? Why It Outshines VLLM, TRT-LLM, TGI, and MLC(bentoml.com)
16 points by helloericsf 2y ago|8 comments
52.
by helloericsf 2y ago
Seems like is not another model company. https://x.com/ssi/status/1803472825476587910
53.
by helloericsf 2y ago
Agreed. custom models could be a hit or a miss.
54.
by helloericsf 2y ago
Truly mind-blowing to see Mixtral 47B running on a smartphone.
55.
by helloericsf 2y ago
Project Github link: https://github.com/SJTU-IPADS/PowerInfer
56.
21.2× faster than llama.cpp? plus 40% memory usage reduction(arxiv.org)
43 points by helloericsf 2y ago|14 comments
57.
by helloericsf 2y ago
Also, their AI/ML side of the house is a mess. :(
58.
by helloericsf 2y ago
Snowflake unveiled Polaris yesterday, Databricks acquires Tabular today. What's coming tmr?
59.
Databricks acquires Tabular, Snowflake fork Iceberg?(datagravity.dev)
2 points by helloericsf 2y ago|3 comments
60.
New Yi 1.5 models under Apache 2.0(huggingface.co)
2 points by helloericsf 2y ago|0 comments

More ›