Akjava (Akihito Miyazaki)

posted an update 9 months ago

Post

799

I've shared Hugging Face Spaces for CPU-based RAG and T5/Flan-T5 models. The smolagents-rag space sometimes produces high-quality answers, but it can be slow. Qwen2.5-0.5B is as fast as a CPU implementation and generates answers of acceptable quality. I've found that Gemma3-4B produces significantly more stable answers than the 1B version.

Rag
Akjava/Gemma3-4B-llamacpp-cpu-rag-smolagents
Akjava/Qwen2.5-0.5B-Rag-Thinking-Flan-T5

t5/flan-t5
Akjava/llamacpp-flan-t5-large-grammar-synthesis
Akjava/llamacpp-madlad400-3b-mt-2jp

Huggingface Free CPU Limitations
When duplicating a space, the build process(llama-cpp-python) can occasionally become stuck, requiring a manual restart to finish.
Spaces may unexpectedly stop functioning or even be deleted, leading to the need to rework them. Refer to issue for more information.

posted an update 9 months ago

Post

507

Akjava/Smolagents-ExtraSearchTools
A tool for executing searches using multiple search tools in a prioritized order,
particularly useful for developers who have experienced rate-limiting issues with
DuckDuckGoSearchTool during smolagents development.

GoogleCustomSearchTool: This tool utilizes the Google Custom Search JSON API to perform web searches. It can be configured to search entire websites, but this requires setting up a Custom Search Engine in the Google Cloud Console. The free tier of the Google Custom Search API is limited to 100 queries per day.

BraveSearchTool: This tool uses the Brave Search API to perform web searches. While there is a free tier allowing 2000 queries per month, it requires adding payment information.

PrioritySearchTool: This tool manages multiple search tool instances, executing them in a prioritized order. It returns the result from the first search tool that completes successfully, providing a fallback mechanism if higher-priority tools fail.

save_json_path (optional): This optional parameter specifies the path to a JSON file. If provided, the PrioritySearchTool will cache search queries and their corresponding results to this file. This can improve performance and reduce API usage for repeated queries.

Example:
priority_search = PrioritySearchTool(
[DuckDuckGoSearchTool(), GoogleCustomSearchTool("xxxxxx"),BraveSearchTool()],
save_json_path="history.json",
)

posted an update 9 months ago

Post

2678

Initial API-Based Smolagents and Linear.app Integration Example
Akjava/linear-app-api-smolagents
In short,this example contain get_todo_issue() tool and add_comment(),change_state_reviewing() function to linear.app

Large language models, like 70B parameter models, can often readily utilize tools such as add_comment or change_state, potentially handling multiple issues concurrently.

However, smaller models may require repeated calls to a tool or even fail to utilize it entirely.

Therefore, this initial example focuses on the get_todo_issue() tool.

posted an update 9 months ago

Post

758

A dataset of 50 instrumental music tracks generated with the DiffRhythm model, using 10 CC0-licensed instrument samples from OEPN Game Art.
Akjava/diffrhythm-instrument-cc0-oepngamearg-10x5-generated

I've released the dataset. It's a little skewed towards certain types of music. It might be interesting for people curious about the range of variations it can generate. It could also be a good starting point for experimenting with the Distrill model. I believe the quality is good enough to be used as background music for YouTube videos or probably as reference tracks for YuE or Udio.

posted an update 9 months ago

Post

600

First Example of Direct Webhook-triggered AI Agent
Akjava/linear-app-webhook-smolagents

This space-code might be helpful as a reference if you want to receive issue changes from linea.app via webhook and handle them using Gradio on Hugging Face Spaces or locally with AI.

Imagine an agent, responding instantly.

In short, Huggingface published webhooks_server.py under Apache 2.0,I've adapted it to work with a very small part of linear.app.

https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/_webhooks_server.py

Be caraful

This method can be used as an AI-powered content management system, but I'm not sure if Hugging Face will allow it.

Github
https://github.com/akjava/smolagents-examples

replied to their post 9 months ago

Sorry if there is any misunderstanding, English is not my native langauage.
This is an issue between me and huggingface.co.

posted an update 9 months ago

Post

505

2 replies

·

reacted to m-ric's post with 👍 9 months ago

Post

4919

We now have a Deep Research for academia: SurveyX automatically writes academic surveys nearly indistinguishable from human-written ones 🔥

Researchers from Beijing and Shanghai just published the first application of a deep research system to academia: their algorithm, given a question, can give you a survey of all papers on the subject.

To make a research survey, you generally follow two steps, preparation (collect and organize papers) and writing (outline creation, writing, polishing). Researchers followed the same two steps and automated them.

🎯 For the preparation part, a key part is find all the important references on the given subject.
Researchers first cast a wide net of all relevant papers. But then finding the really important ones is like distilling knowledge from a haystack of information. To solve this challenge, they built an “AttributeTree” object that structures key information from citations. Ablating these AttributeTrees significantly decreased structure and synthesis scores, so they were really useful!

📝 For the writing part, key was to get a synthesis that's both short and true. This is not easy to get with LLMs! So they used methods like LLM-based deduplication to shorten the too verbose listings made by LLMs, and RAG to grab original quotes instead of made-up ones.

As a result, their system outperforms previous approaches by far!

As assessed by LLM-judges, the quality score os SurveyX even approaches this of human experts, with 4.59/5 vs 4.75/5 🏆

I advise you to read the paper, it's a great overview of the kind of assistants that we'll get in the short future! 👉 SurveyX: Academic Survey Automation via Large Language Models (2502.14776)
Their website shows examples of generated surveys 👉 http://www.surveyx.cn/

posted an update 10 months ago

Post

2492

I shared smolagents examples

Akjava/open_Deep-Research-DuckDuckGo
Akjava/open_Deep-Research-DuckDuckGo-Groq

Replacing img-src to "#" in mdconvert.py help reducing tokens
I added translate final answer to my language

reacted to nyuuzyou's post with ❤️ 11 months ago

Post

1719

🎨 Artfol Dataset - nyuuzyou/artfol

A collection of 1,892,816 artwork posts featuring:
- High-quality art pieces with various styles and techniques
- Complete metadata including artist IDs, titles, and moderation flags
- Content from Artfol social media platform

The dataset contains:
- Public domain artwork posts
- Artist attribution and identifiers
- Direct image URLs and web page links
- Content safety flags (NSFW, gore)
- Post titles and descriptions

All content is available under CC0 license, allowing unrestricted use including commercial applications.

posted an update 11 months ago

Post

660

I've released some spaces that demonstrates more advanced uses of MediaPipe-landmarks.

Head-pose-estimate
original mediapipe’s detection is good on short angles,trained-model seems work and there are more improve space
Akjava/mediapipe-head-pose-estimation

generate-3d-head:gltf
this is simple and initial
Akjava/mediapipe-face-mesh-3d
Akjava/mediapipe-head-2d-spinning

prototype-faceswap
color adjust and transform soso work,need find a way to keep face features.
Akjava/mediapipe-face-skin-transform

reacted to davidberenstein1957's post with 🔥 about 1 year ago

Post

1728

Let’s make a generation of amazing image-generation models

The best image generation models are trained on human preference datasets, where annotators have selected the best image from a choice of two. Unfortunately, many of these datasets are closed source so the community cannot train open models on them. Let’s change that!

The community can contribute image preferences for an open-source dataset that could be used for building AI models that convert text to image, like the flux or stable diffusion families. The dataset will be open source so everyone can use it to train models that we can all use.

Blog: https://huggingface.co/blog/burtenshaw/image-preferences