Tech Blog
🌖
NJX-njx PRO
NJX-njx
AI & ML interests
AI infra, large model architecture, intelligent agent,evaluation
Recent Activity
new activity
7 days ago
bigscience/bloom:pretokenizer Regex issues? new activity
7 days ago
bigscience/bloom:training BLOOM/BLOOMZ for text summarization Organizations
pretokenizer Regex issues?
8
#278 opened over 1 year ago
by
hpcpony
Is is feasible to use this checkpoint for multi node inference via deepspeed Zero stage-3
3
#275 opened almost 2 years ago
by
YuTian8328
training BLOOM/BLOOMZ for text summarization
3
#277 opened almost 2 years ago
by
almonzer
Mutli turn support
3
#282 opened over 1 year ago
by
ansumanbehera
Let's talk about the model
4
#284 opened 7 months ago
by
kalashshah19
updated a Space 7 days ago
published a Space 7 days ago
reacted to their post with 🔥 8 days ago
Post
930
I feel that I have become more and more obsessed with studying some "primitive" CLI operations recently.
Compared to the so-called MCP and Skill, enabling AI to understand and use CLI is actually more feasible, explainable, and powerful in terms of code.
I recently deployed a website for my OpenSoul on Vercel. In the past, I might have needed to spend a lot of cognitive effort or time to understand how to operate on the Vercel page, and I would have had to spend a great deal of time reading documents (smarter people might directly feed the documents to AI and let AI summarize feasible and reliable steps).
But in fact, after ChatGPT told me that Vercel actually has a CLI, I directly asked my Copilot in VS Code to download this command line, clearly stated my needs, and it quickly solved everything else. The only thing I actually needed to do was log in to Vercel and create a key.
This suddenly reminds me of a blog post I read earlier that interviewed the father of Claude Code. The reason why Claude Code did not develop front-end pages and the like is precisely because he believes that we should focus most of our energy on the most meaningful interaction logic.
So, in an era where AI capabilities are becoming increasingly strong, perhaps what we really need is to pick up those tools that we used with the sole goal of achieving functionality when computing power was tight. What do you think?
Compared to the so-called MCP and Skill, enabling AI to understand and use CLI is actually more feasible, explainable, and powerful in terms of code.
I recently deployed a website for my OpenSoul on Vercel. In the past, I might have needed to spend a lot of cognitive effort or time to understand how to operate on the Vercel page, and I would have had to spend a great deal of time reading documents (smarter people might directly feed the documents to AI and let AI summarize feasible and reliable steps).
But in fact, after ChatGPT told me that Vercel actually has a CLI, I directly asked my Copilot in VS Code to download this command line, clearly stated my needs, and it quickly solved everything else. The only thing I actually needed to do was log in to Vercel and create a key.
This suddenly reminds me of a blog post I read earlier that interviewed the father of Claude Code. The reason why Claude Code did not develop front-end pages and the like is precisely because he believes that we should focus most of our energy on the most meaningful interaction logic.
So, in an era where AI capabilities are becoming increasingly strong, perhaps what we really need is to pick up those tools that we used with the sole goal of achieving functionality when computing power was tight. What do you think?
- 1 reply
reacted to ajibawa-2023's post with 🔥 8 days ago
Post
3700
Cpp-Code-Large
Dataset: ajibawa-2023/Cpp-Code-Large
Cpp-Code-Large is a large-scale corpus of C++ source code comprising more than 5 million lines of C++ code. The dataset is designed to support research in large language model (LLM) pretraining, code intelligence, software engineering automation, and static program analysis for the C++ ecosystem.
By providing a high-volume, language-specific corpus, Cpp-Code-Large enables systematic experimentation in C++-focused model training, domain adaptation, and downstream code understanding tasks.
Cpp-Code-Large addresses the need for a dedicated C++-only dataset at substantial scale, enabling focused research across systems programming, performance-critical applications, embedded systems, game engines, and large-scale native software projects.
Dataset: ajibawa-2023/Cpp-Code-Large
Cpp-Code-Large is a large-scale corpus of C++ source code comprising more than 5 million lines of C++ code. The dataset is designed to support research in large language model (LLM) pretraining, code intelligence, software engineering automation, and static program analysis for the C++ ecosystem.
By providing a high-volume, language-specific corpus, Cpp-Code-Large enables systematic experimentation in C++-focused model training, domain adaptation, and downstream code understanding tasks.
Cpp-Code-Large addresses the need for a dedicated C++-only dataset at substantial scale, enabling focused research across systems programming, performance-critical applications, embedded systems, game engines, and large-scale native software projects.
-
- 3 replies
replied to ajibawa-2023's post 8 days ago
I'm curious if there is any part of the dataset that involves AI infra.
replied to AbstractPhil's post 8 days ago
听起来你们的工作很有意思，加油
posted an
update 8 days ago
Post
930
I feel that I have become more and more obsessed with studying some "primitive" CLI operations recently.
Compared to the so-called MCP and Skill, enabling AI to understand and use CLI is actually more feasible, explainable, and powerful in terms of code.
I recently deployed a website for my OpenSoul on Vercel. In the past, I might have needed to spend a lot of cognitive effort or time to understand how to operate on the Vercel page, and I would have had to spend a great deal of time reading documents (smarter people might directly feed the documents to AI and let AI summarize feasible and reliable steps).
But in fact, after ChatGPT told me that Vercel actually has a CLI, I directly asked my Copilot in VS Code to download this command line, clearly stated my needs, and it quickly solved everything else. The only thing I actually needed to do was log in to Vercel and create a key.
This suddenly reminds me of a blog post I read earlier that interviewed the father of Claude Code. The reason why Claude Code did not develop front-end pages and the like is precisely because he believes that we should focus most of our energy on the most meaningful interaction logic.
So, in an era where AI capabilities are becoming increasingly strong, perhaps what we really need is to pick up those tools that we used with the sole goal of achieving functionality when computing power was tight. What do you think?
Compared to the so-called MCP and Skill, enabling AI to understand and use CLI is actually more feasible, explainable, and powerful in terms of code.
I recently deployed a website for my OpenSoul on Vercel. In the past, I might have needed to spend a lot of cognitive effort or time to understand how to operate on the Vercel page, and I would have had to spend a great deal of time reading documents (smarter people might directly feed the documents to AI and let AI summarize feasible and reliable steps).
But in fact, after ChatGPT told me that Vercel actually has a CLI, I directly asked my Copilot in VS Code to download this command line, clearly stated my needs, and it quickly solved everything else. The only thing I actually needed to do was log in to Vercel and create a key.
This suddenly reminds me of a blog post I read earlier that interviewed the father of Claude Code. The reason why Claude Code did not develop front-end pages and the like is precisely because he believes that we should focus most of our energy on the most meaningful interaction logic.
So, in an era where AI capabilities are becoming increasingly strong, perhaps what we really need is to pick up those tools that we used with the sole goal of achieving functionality when computing power was tight. What do you think?
-
- 1 reply
replied to their post 9 days ago
我觉得也许我们甚至可以不局限与统一的输入框
replied to unmodeled-tyler's post 9 days ago
to say the truth ,your job is extremely meaningful ,thanks for your explaination
replied to ronantakizawa's post 9 days ago
that's fantastic
posted an
update 10 days ago
Post
196
Recently, I've come across some practices in the community where skills empower intelligent agents, and I'd like to share my thoughts on the future of agents inspired by these practices.
https://huggingface.co/blog/custom-cuda-kernels-agent-skills
Hugging Face (hf) recently created a skill related to kernels and achieved good results in two tests. We know that AI infrastructure is actually a relatively high-threshold task that requires considering many variables and pursuing ultimate performance. However, when we internalize the skills in this field into a single skill, it truly brings about tremendous changes.
Perhaps we need to refocus our attention on this function.
Currently, many of our AI products spend a lot of effort on redundant tasks such as prompt engineering and workflow building. But when these AI products are being developed, they fail to consider that the essential capabilities of large models (context, memory, collaboration, logical reasoning) are actually constantly improving. We don't really need these complex tasks.
I believe that "great truths are simple" is the only solution.
Currently, most tasks can be accomplished with a command-line tool, a skill, and one or more large models. There's no need for any other complex logic—just bash.
Maybe there aren't that many things we need to do right now. Find a sufficiently vertical field, internalize the knowledge within that field into a skill (which can also take other forms), create interfaces for any channels you can think of in the form of command lines, and allow AI to thrive in as many tasks as possible.
Then leave everything to AI. The power of bash is beyond your imagination.
@AdinaY @burtenshaw@clem@evalstate
https://huggingface.co/blog/custom-cuda-kernels-agent-skills
Hugging Face (hf) recently created a skill related to kernels and achieved good results in two tests. We know that AI infrastructure is actually a relatively high-threshold task that requires considering many variables and pursuing ultimate performance. However, when we internalize the skills in this field into a single skill, it truly brings about tremendous changes.
Perhaps we need to refocus our attention on this function.
Currently, many of our AI products spend a lot of effort on redundant tasks such as prompt engineering and workflow building. But when these AI products are being developed, they fail to consider that the essential capabilities of large models (context, memory, collaboration, logical reasoning) are actually constantly improving. We don't really need these complex tasks.
I believe that "great truths are simple" is the only solution.
Currently, most tasks can be accomplished with a command-line tool, a skill, and one or more large models. There's no need for any other complex logic—just bash.
Maybe there aren't that many things we need to do right now. Find a sufficiently vertical field, internalize the knowledge within that field into a skill (which can also take other forms), create interfaces for any channels you can think of in the form of command lines, and allow AI to thrive in as many tasks as possible.
Then leave everything to AI. The power of bash is beyond your imagination.
@AdinaY @burtenshaw@clem@evalstate
-
- 3 replies
upvoted an article 10 days ago
Article
Custom Kernels for All from Codex and Claude
- +2
•
69
replied to marksverdhai's post 10 days ago
hhhhhh
replied to ronantakizawa's post 10 days ago
This dataset is really interesting. I'm curious if it contains clear information about the positions of website components.
So my understanding is that the data used to train this model from the beginning is not English corpus, nor is it text, so its tokenizer is also different from the traditional one. I'm curious about how this part is handled and how the model itself understands things. Is it the same as the traditional one, which is also a one-dimensional token sequence?