Projects

Overview: Physics Master is a model fine-tuned from LLaMA 3-8B-Instruct, designed to answer physics questions. Key features: - Fine-tuned from LLaMA 3-8B-Instruct - Open source on Hugging Face - Python code available Tech stack: Jupyter Notebook Why it matters: A powerful tool for physics enthusiasts and students seeking accurate answers to their questions.

### Overview: Facet AI simplifies the fine-tuning of Gemma models, making it accessible to non-technical users. ### Key Features: - **Dataset Preprocessing & Augmentation**: Handle vision and text data from custom uploads or Hugging Face datasets. - **Fine-tuning with Multiple Frameworks**: Support for Hugging Face Transformers, Unsloth, and more. - **Flexible Training Methods**: Support for SFT, DPO/ORPO, and GRPO techniques. - **Model Export & Deployment**: Export models to various formats and deploy to GCP for inference. ### Tech Stack: Python ### Why it Matters: Facet AI streamlines the fine-tuning process, making it easier for non-technical users to leverage cutting-edge post-training techniques without the need for coding. ### GitHub Stars: 16 ### Primary Language: Python ### Repository: gemma-facet/cloud-services ### GitHub Description: Platform for managing LLM fine-tuning (Google Summer of Code '25) ### Background: Facet AI is an open-sourced backend designed to streamline and centralize fine-tuning workflows, making it easy for non-technical users to deploy on their own cloud without writing code. ### Features: - **Dataset**: Process and enhance datasets using custom files, Hugging Face datasets, and NLP techniques. - **Training**: Support for various post-training techniques, including SFT, DPO/ORPO, and GRPO. - **Utilities**: Export models to multiple formats, track train jobs with Weights & Biases, and deploy models to G

### Overview: 本项目是一个专为 sub2api 项目设计的轻量级 AI 图片生成前端页面，通过 iframe 嵌入方式集成到 sub2api 后台，实现无缝使用 OpenAI 图片生成功能。 ### Key features: - 文生图：输入描述性提示词，AI 自动生成对应图片。 - 图生图：上传参考图，基于参考图进行二次创作。 - 多种分辨率：支持 1:1、16:9、4:3、3:4、9:16 等画面比例。 - 多档质量：支持自动、低、中、高质量选择。 - 批量生成：单次可生成 1-5 张图片。 - 历史记录：自动保存最近 30 条生成记录。 - 费用预估：实时显示生成费用。 ### Tech stack: JavaScript ### Why it matters: 无需配置，嵌入即用，完美适配 sub2api 的 iframe 通信协议。支持最新模型，效果最佳，兼容性好，费用预估准确，节省用户时间和资源。

Overview: jsonltui is a fast, interactive TUI application designed to visually navigate and inspect JSON and JSONL data. It supports both single JSON objects and JSONL files, with a focus on localized parse errors and error handling. Key features: - Interactive tree view of JSON and JSONL data in both TUI and Web UI - Supports single JSON objects and JSONL files - Graceful handling of parsing errors, displaying them inline with successfully parsed data - Expandable error nodes showing original unparsable text and error messages - Truncation of long text values with the ability to view full text in a modal - Keyboard navigation for easy exploration of the JSON structure in TUI mode - Web-based interface for a more familiar and accessible visualization Tech stack: HTML Why it matters: jsonltui is optimized for performance and ease of use, making it suitable for developers and data analysts who need to quickly inspect and debug JSON data. The ability to localize parse errors and handle large JSONL files makes it particularly useful for large-scale data analysis. Repository: gr-b/jsonltui Primary language: HTML GitHub stars: 16 GitHub description: A fast TUI application (with optional slow webui) to visually navigate and inspect JSON and JSONL data. Easily localize parse errors in large JSONL files. Made with LLM fine-tuning workflows in mind.

Overview: This is an unofficial demo app for CogView2, allowing users to interact with the model without needing the official app. Key features: - Web demo access - Training and inference capabilities - Optional pretrained model download Tech stack: Python Why it matters: This app provides a convenient way to experiment with and visualize the CogView2 model, without the need for the official app, making it accessible to a broader audience.

Overview: Abigail is a powerful CLI tool and Python library that allows developers to access free web AI services directly from their terminal or Python code, without requiring any API keys. Key features: - CLI tool for streaming AI responses - Image generation using Perchance - Video generation using Upsampler - Interactive terminal chat mode - List of supported providers and models Tech stack: Python Why it matters: Abigail simplifies the integration of AI services into developer workflows, making it easier to prototype and deploy AI solutions without the need for complex API management. Repository: iqbalmh18/abigail Primary language: Python GitHub stars: 16 GitHub description: CLI & async Python library for free AI chat, image & video generation.

Overview: FLUX 2 Pro API, also known as `flux-2-pro`, is a model id, pricing, and parameterized OpenAI-compatible API endpoint provided by APIMart. It offers a single key for all catalog models, including 300+ image, video, and language models, with a flat rate per image and no monthly subscription. Key features: - Single key for all models, including 300+ image, video, and language models. - Flat rate pricing per image, with tiered resolutions. - Asynchronous processing with real-time task tracking. - Billing is reported in the response, and tasks can be retried if failed. Tech stack: Python. Why it matters: This API offers a cost-effective solution for generating images from text or references, with a simple and efficient API endpoint that integrates seamlessly with the APIMart platform.

Overview: Qwen Image 2.0 Pro is an advanced API endpoint for generating high-quality images using Qwen's powerful AI capabilities. It offers flexible pricing and easy integration with OpenAI-compatible endpoints, making it ideal for developers and creators looking to enhance their image generation workflows. Key features: - Supports multiple models and formats, including text-to-image and image-to-image transformations. - Offers a per-image pricing model, with a base price of $0.05 per image. - Provides detailed API documentation and examples for seamless integration. - Asynchronous processing with real-time progress updates. Tech stack: Python Why it matters: Qwen Image 2.0 Pro simplifies the process of generating high-quality images, making it an essential tool for developers and creators looking to enhance their projects with AI-generated visuals.

### Overview: Nano Banana 2 Lite API (`gemini-3.1-flash-lite-image-ext`) offers a cost-effective alternative to Gemini 3.1 Flash Lite Image, providing per-image pricing. This API supports a wide range of models and is compatible with OpenAI endpoints, making it ideal for developers looking for a reliable and efficient image generation service. ### Key Features: - **Per-Image Pricing**: Nano Banana 2 Lite is priced based on the number of images generated, offering a more economical alternative to the Gemini 3.1 Flash Lite Image route. - **OpenAI-Compatible Endpoint**: Supports the OpenAI-compatible API, allowing seamless integration with existing OpenAI models. - **Model ID and Routes**: Provides a list of supported model IDs and their respective pricing tiers. - **Pricing Model**: Offers a list price and an effective price, with real-time updates available through the API. - **Request Parameters**: Accepts various parameters such as `model`, `prompt`, `size`, `resolution`, `n`, and `image_urls`, enabling flexible and efficient image generation. ### Tech Stack: - Python ### Why It Matters: - **Cost-Effective**: Provides a more affordable alternative to Gemini 3.1 Flash Lite Image, making it ideal for developers on a budget. - **OpenAI Compatibility**: Ensures seamless integration with OpenAI models, allowing developers to leverage their existing infrastructure. - **Model ID and Routes**: Offers a comprehensive list of supported model IDs and their pricing tiers, making it easy to choose the right model for your needs. - **Real

### Overview: This repository contains Wan 2.7 R2V prompts tested with example renders, aspect ratios, the model ID (`wan2.7-r2v`), and the cost per image. The prompts are designed to be used with the APIMart API route, providing a cost-effective way to generate images without the need for a monthly subscription. ### Key Features: - **Model ID**: `wan2.7-r2v` - **Cost per Image**: $0.0664 per second - **Duration**: 5 seconds - **Prompt Structure**: - **Subject**: Concrete object or person with material - **Light**: Direction and quality - **Framing**: Ratio and lens or distance - **Setting**: Surface, backdrop, environment - **Finish**: Render or film character ### Tech Stack: Python ### Why It Matters: - **Cost-Effective**: Generate images without the need for a monthly subscription. - **Versatile**: Supports multiple image formats and dimensions. - **Easy to Use**: Simple prompt structure and straightforward API integration. ### Repository: - **Repository Name**: j4mcxueyfsi/wan-2-7-r2v-prompts-es - **GitHub Stars**: 16 - **GitHub Description**: Wan 2.7 R2V prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`wan2.7-r2v`) and the cost per image.

Overview: This plugin allows developers to search their dataset for images similar to a given image, using a similarity index for efficient matching. It supports searching by URL or uploading an image, and displays the image found in the panel. Key features: - Supports searching by URL or drag-and-drop - Displays the image found in the panel - Requires a similarity index to be present on the dataset Tech stack: TypeScript Why it matters: This plugin simplifies the process of finding similar images in a dataset, making it easier for developers to integrate image similarity matching into their projects.

### Overview: This project aims to implement a Brain Computer Interface (BCI) model using Kubernetes for efficient and scalable deployment of machine learning (ML) projects. The use of Kubernetes provides a flexible and automated approach to managing containerized applications, including data collection, preprocessing, feature extraction, model training, and deployment. ### Key Features: - **Kubernetes Cluster Setup:** Deploy a Kubernetes cluster using Google Kubernetes Engine (GKE). - **ML Pipeline:** Build and deploy an ML pipeline using Kubeflow, enabling hyperparameter tuning and model training. - **Data Storage:** Manage data using Google Cloud Storage (GCS). - **Experiment Tracking & AutoML:** Use Weight & Biases to track experiments and find optimal hyperparameters. - **Model Versioning:** Save and version models using Mlflow. - **Model Serving:** Deploy models using BentoML for API communication with users. - **Monitoring:** Monitor cluster resources with Prometheus and Grafana. ### Tech Stack: Kubernetes ### Why It Matters: Kubernetes simplifies the management of containerized applications, making it ideal for implementing ML pipelines and deploying models. It provides features like automated scaling, load balancing, and scheduling, ensuring efficient and scalable deployment of ML projects.

Overview: Seedream 5.0 Flash API (`seedream-5-0-flash`) is a model id, pricing, parameters, and copy-paste examples for the OpenAI-compatible APIMart endpoint, offering low-cost, high-quality image generation with text rendering and multi-reference style fusion. Key features: - Per-image pricing at $0.0137 per image - Supports 1K, 1.5K, and 2K resolutions - Model id shown in route table - Supports multi-reference style fusion - Supports batch jobs and rapid iteration Tech stack: Python Why it matters: Seedream 5.0 Flash API provides a cost-effective alternative to traditional image generation tools, offering high-quality images at a lower price point. The API supports text rendering and multi-reference style fusion, making it ideal for creative projects requiring detailed and varied image outputs. Repository: jikbtcp30/seedream-5-0-flash-api-cn Primary language: Python GitHub stars: 16 GitHub description: 即梦图片 — Seedream 5.0 Flash API (`seedream-5-0-flash`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint. README excerpt: # Seedream 5.0 Flash API (`seedream-5-0-flash`) <!-- conv-kit:v1 --> <p align="center"> <img src="assets/badges/price.svg" alt="observed unit price"> <img src="assets/badges/b

### Overview: This repository contains 8 tested Imagen 4.0 prompts, each with an example render, aspect ratio, and the model ID (`imagen-4.0-apimart`) and cost per image. The prompts are designed to be used as a key for generating high-quality images at a cost-effective rate. ### Key Features: - **Model ID:** `imagen-4.0-apimart` - **Prompt Structure:** Each prompt is structured to include the subject, light direction and quality, framing ratio, and finish. This ensures clear and effective image generation. - **Cost:** $0.04 per image, with a minimum of $1 for the entire set. - **Asynchronous by Design:** Tasks can be submitted and tracked using `task_id`, with automatic retries and batching. ### Tech Stack: The project uses Python for its implementation. ### Why It Matters: Using Imagen 4.0 with APIMart offers a single, OpenAI-compatible endpoint for generating images at a cost-effective rate. This makes it ideal for teams looking to leverage Imagen's capabilities without the need for a monthly subscription or prepaid plans. The consistent base URL and shared `Authorization` header across multiple models ensures flexibility and ease of use.

### Overview: This repository contains 8 tested Nano Banana 2 Lite prompts, each with an example render, aspect ratio, and the cost on the APIMart image API. The prompts are designed to help developers create high-quality images efficiently and cost-effectively. ### Key Features: - **8 Prompts**: Each prompt is designed to achieve a specific look, such as commercial product photography, editorial food photography, and cinematic stills. - **Aspect Ratios**: The prompts cover various aspect ratios, including 1:1, 4:3, and 16:9, ensuring versatility in different use cases. - **Cost Calculation**: The cost is provided on the same API endpoint as the image, making it easy to track and manage expenses. ### Tech Stack: Python ### Why It Matters: - **Efficiency**: The prompts are designed to be used with a single API key, allowing for seamless integration with existing workflows. - **Cost Management**: The cost is automatically calculated and reported, eliminating the need for manual billing and prepayment. - **Customization**: The prompts can be easily swapped and modified without altering the client-side code, ensuring flexibility and ease of use. The repository provides a comprehensive guide on how to structure prompts effectively, including examples and cheat sheets for quick reference.

Overview: Seedream 4.0 API is an OpenAI-compatible endpoint that provides access to Seedream 4.0, a unified text-to-image and image editing model. It offers a single endpoint compatible with OpenAI, allowing for unsubscription-based pricing and batch processing with no prepayment required. Key features: - Unsubscription-based pricing - Batch processing with no prepayment - Real-time billing and cost reporting - Asynchronous processing Tech stack: Python Why it matters: Seedream 4.0 API simplifies the integration of Seedream 4.0 with other applications, offering a single endpoint for text-to-image and image editing tasks, making it easier for developers to incorporate Seedream 4.0 into their projects without the need for additional subscriptions or upfront payments. Repository: johv6uyc/seedream-4-0-api-fr Primary language: Python GitHub stars: 8 GitHub description: Seedream 4.0 API (`seedream-4-0`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint. README excerpt: # Seedream 4.0 API (`seedream-4-0`) <!-- conv-kit:v1 --> <p align="center"> <img src="assets/badges/price.svg" alt="observed unit price"> <img src="assets/badges/billing.svg" alt="billing model"> <img src="assets/badges/compat.svg" alt="OpenAI-compatible endpoint"> </p>

Overview: Seedream 5.0 Lite API, also known as `seedream-5-0-lite`, is an OpenAI-compatible API endpoint that provides access to a wide range of image generation models, including Seedream 5.0 Lite, for a flat rate of $0.028 per image. This API supports batch processing of up to 15 images and offers a simple and efficient way to integrate Seedream 5.0 Lite into projects without the need for a monthly subscription. Key features: - Access to Seedream 5.0 Lite and over 300 other image generation models via a single API endpoint. - Pricing: $0.028 per image, with the cost reflected in the response. - Asynchronous processing: Tasks are submitted, and results are fetched via the `GET /v1/tasks/{task_id}` endpoint. - No subscription required: Users can pay as they go, with charges based on actual usage. - Compatible with OpenAI: The API endpoint is designed to work seamlessly with the OpenAI API. Tech stack: The API is built using Python, utilizing the `requests` library for making HTTP requests and handling JSON responses. Why it matters: Seedream 5.0 Lite API simplifies the integration of Seedream 5.0 Lite into projects by providing a straightforward and cost-effective solution. It eliminates the need for a monthly subscription, making it ideal for developers and small businesses looking to experiment with image generation without the financial burden of a subscription.

**Overview:** This repository contains 8 tested recipes for GPT Image 1.5 prompts, including example renders, aspect ratios, and the cost per image on the APIMart API endpoint. The project aims to provide developers with practical image prompt recipes for their projects. **Key features:** - 8 GPT Image 1.5 prompts - Example renders, aspect ratios, and call costs - Model: `gpt-image-1.5-official` - Effective price: $4.00 per image - Estimated cost for the set: $32.00 **Tech stack:** Python **Why it matters:** This project simplifies the process of creating high-quality image prompts for developers, ensuring they can quickly and efficiently generate the desired output without the need for complex setup or manual adjustments.

### Overview: 本项目通过将纯文本模型（如 DeepSeek `deepseek-v4-flash`）与 StepFun `step-3.7-flash` 模型结合，为用户提供看图功能，从而在不支持图片输入的场景中，让主模型能够继续推理和写代码。这使得用户能够处理图片内容，而无需手动保存或上传图片文件。 ### Key features: 1. **零依赖 Node.js**：项目完全依赖于 Node.js，无需额外的依赖。 2. **`--latest`**：用户粘贴进 Codex 的图片（即使主模型从未收到），也能从 `~/.codex/sessions/*.jsonl` 里恢复并识别。 3. **本地文件 / 图片 URL / 多张图**：支持粘贴本地文件、图片 URL 和多张图片。 4. **带问题识图**：支持通过 `--prompt` 参数让视觉模型按需描述图片内容。 5. **推理模型适配**：`reasoning_effort`（默认低）+ `content` 空时回退 `reasoning`，确保结果必出。 6. **Provider 守卫**：只在主模型是 `deepseek-v4-flash / deepseek-v4-pro` 时启用。 7. **跨平台**：适用于 macOS、Windows 和 Linux 系统。 ### Tech stack: 项目使用 Node.js 作为后端，支持 `latest` 版本的 Node.js 和 `>= 1

### Overview: Seedance 1.5 Pro Prompts: A Comprehensive Guide to Image Prompt Recipes ### Key Features: - **Model ID:** `seedance-1.5-pro` - **Asynchronous Processing:** Tasks are submitted, retrieved, and can be retried in a queue. - **Cost Per Image:** $0.0104 per second, billed by resolution tier. ### Tech Stack: Python ### Why It Matters: Seedance 1.5 Pro offers a cost-effective way to generate high-quality images, making it ideal for developers and content creators looking to streamline their workflow without the need for a subscription. ### Repository: k2t3e2-a/doubao-seedance-1-5-pro-prompts-fr ### Primary Language: Python ### GitHub Stars: 9 ### GitHub Description: Seedance 1.5 Pro Prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`seedance-1.5-pro`) and the cost per image. ### Primary Language: Python ### GitHub Stars: 9