Projects

Overview: The Loops Cookbook is a collection of example recipes and patterns designed to help Loops users get started with reinforcement learning (RL) and supervised fine-tuning (SFT) on Baseten Loops. Key features: - RL with a verifiable reward (RLVR) - RLVR with held-out evaluations - RLVR on images - RL with an LLM judge - An agent that uses tools in a sandbox - A coding agent on a real benchmark - A task with several verified steps - Asynchronous RL - One policy in two roles - An agent built on another framework - Your own episodes, step by step Tech stack: Python Why it matters: The Loops Cookbook provides a starting point for training post-training models on Baseten Loops, offering RL and SFT templates that can be easily adapted and used as a foundation for more complex projects.

Overview: This project is an Apple Developer Documentation MCP server that indexes 16,253+ documents covering all Apple platforms, WWDC session transcripts, and API references. It allows AI assistants to search through the complete documentation with precise answers and source links. Key features: - Comprehensive coverage of Apple's entire documentation, including all platforms, WWDC sessions, and API references. - Natural language search for asking questions in plain English. - Complete coverage of design and interface guidelines, including Human Interface Guidelines. - Working code examples extracted from the documents. Tech stack: JavaScript Why it matters: This tool provides a complete and accurate source of Apple's developer documentation, making it easier for AI assistants to find and understand the latest and most comprehensive information available. Repository: bbssppllvv/apple-docs-mcp-server Primary language: JavaScript GitHub stars: 9 GitHub description: MCP server for semantic search through Apple developer documentation, WWDC transcripts, and code examples.

Overview: MojoVec is a vector database implementation built entirely in Mojo, focusing on approximate nearest neighbor search using SIMD instructions. It aims to provide performance comparable to C++ libraries like FAISS and hnswlib, while maintaining the simplicity and ease of use of Mojo. Key features: - Flat and SQ8 HNSW collections with a Mojo-native Chroma-style API. - Support for squared L2, cosine, and inner-product distance metrics. - Record metadata with types like String, Int, Float64, and Bool. - Batched query and update operations. - Full-text search over collection documents. - Hybrid BM25 search with reciprocal rank fusion. - Atomic save/load, optional WAL recovery, and collection statistics. - IVF-PQ collections with explicit or automatic training and persistence. - IVF-Flat index implemented in Mojo. Tech stack: Mojo. Why it matters: MojoVec offers a native Mojo implementation that aims to match the performance of C++ libraries while keeping the simplicity of Mojo, making it suitable for applications that require both performance and ease of use. GitHub stars: 9 Repository: bewaffnete/MojoVec

Overview: Grok Imagine Image API (`grok-imagine-image`) is a powerful image generation and editing service that allows users to create stunning photorealistic images from text prompts or intelligently edit existing images using natural language. It supports over 300 models and is compatible with OpenAI's API. Key features: - Supports over 300 models including image, video, and language models - Compatible with OpenAI's API - Provides a single endpoint for both image and text generation - Supports image resizing and multiple image generation - Charges based on the number of images generated or edited - Supports image URLs and base64 data URLs for reference images - Assists with image editing tasks Tech stack: Python Why it matters: Grok Imagine Image API is a cost-effective and efficient solution for image generation and editing, making it ideal for developers and creatives who need to quickly create high-quality images without the need for complex setup or expensive software licenses. GitHub stars: 9 Repository: bhaiiya1p6r32/grok-image-api-pt

Overview: This repository provides a comprehensive guide to implementing Agentic AI solutions using the DAPR framework and cloud technologies such as OpenAI's Agents SDK, Memory Management, MCP, A2A, Knowledge Graphs, and various cloud-native tools including Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Kafka, Rancher Desktop, and Dapr. Key features: - Event-Driven Architecture - Scalability through Kubernetes - Intelligent Agents using OpenAI's SDK - Real-Time Data Processing with Kafka and RabbitMQ - Persistent Storage in PostgreSQL and Redis Tech stack: Dapr, OpenAI Agents SDK, Memory Management, MCP, A2A, Knowledge Graphs, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Kafka, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Kafka, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Rancher Desktop, Dapr, Docker, Kubernetes, PostgreSQL, Redis, RabbitMQ, Rancher Desktop, D

### Overview: Qwen-Image 3.0 prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`qwen-image-3.0`), and the cost per image. ### Key Features: - **Model ID:** `qwen-image-3.0` - **Aspect Ratios:** 1:1, 4:3, 16:9 - **Cost per Image:** $0.0206 ### Tech Stack: Python ### Why It Matters: - **Single API Key:** Access to Qwen-Image 3.0 and over 300+ models with a single `https://api.apimart.ai/v1` base URL and `Authorization` header. - **No Subscription Required:** No need for monthly plans or pre-defined quotas. - **Real-Time Pricing:** All charges are reflected in the response, with refunds for failed tasks. - **Asynchronous Design:** Tasks are completed by polling the `/v1/tasks/{id}` endpoint, not through batch or retry integration. ### Team Use Case: - **Catalog Management:** Easily catalog and manage a full catalog with a single `https://api.apimart.ai/v1` base URL and `Authorization` header. - **Cost Efficiency:** No need for pre-defined quotas or monthly subscriptions, allowing for flexible and cost-effective usage. - **Real-Time Pricing:** All charges are reflected in the response, with refunds for failed tasks. - **Asynchronous Design:** Tasks are completed by polling the `/v1/tasks/{id}` endpoint

Overview: Seedance 2.5 API (`seedance-2.5`) offers OpenAI-compatible pricing and per-second billing for video generation, including text-to-video and image-to-video models. It supports 300+ models and is priced at $0.0961 per second for 480P, with a minimum charge of $1 for the first $1000. Key features: - **OpenAI-Compatible Pricing**: $0.0961 per second for 480P, with $1 minimum for the first $1000. - **Per-Second Billing**: Charges based on the number of seconds generated. - **Model Id and Routes**: Supports 300+ models with specific tiered pricing. - **Model Page**: Accessible via the model id shown in the route table. - **Model Parameters**: Accepts fields like `model`, `prompt`, `duration`, `resolution`, `size`, and `image_urls`. Tech stack: Python Why it matters: Seedance 2.5 API provides a cost-effective solution for video generation, offering a single key for access to all models and a flexible pricing model that adjusts based on the generated content.

### Overview: 即梦视频 — Seedance 2.5 prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`seedance-2.5`) and the cost per image. ### Key features: - 4 tested recipes with example renders, durations, and costs. - Model id: `seedance-2.5`. - Effective price: $0.0961 per second (estimated cost for the entire set: $0.3844). ### Tech stack: Python ### Why it matters: - **Key, entire directory access**: Same `https://api.apimart.ai/v1` base URL and `Authorization` header for Seedance 2.5 (model `seedance-2.5`) and 300+ models. - **$1 minimum charge, pay per call**: No subscription or prepayment required; charges are based on actual usage. - **Real-time cost reporting**: Reports the actual cost per task (cost / credits_cost); failed tasks are automatically refunded. - **Native asynchronous design**: Simple queue-based processing for batch and retry functionality without custom integrations. Repository: c0j9o/doubao-seedance-2-5-prompts-cn Primary language: Python GitHub stars: 9

### Overview: Grok Imagine 2.0 Prompts: A Comprehensive Guide to Image Prompt Recipes ### Key Features: - **Model ID:** `grok-imagine-2.0-ext` - **Aspect Ratios:** 1:1, 4:3, 3:2, 16:9 - **Cost Per Image:** $0.015 (Region: $0.015) - **Example Renders:** From commercial product photography to food photography, showcasing various ratios and effects. ### Tech Stack: - Python ### Why It Matters: - **Efficient Pricing:** $0.015 per image, with no need for monthly billing. - **Model Compatibility:** OpenAI-compatible endpoint, allowing seamless integration with Grok Imagine 2.0. - **Minimal Setup:** One key for access to Grok Imagine 2.0 and 300+ additional models. - **Automatic Cost Tracking:** Summaries of completed tasks include the cost, ensuring transparency and accountability. ### Repository: - **Repository:** c3lmgfyau8zz/grok-imagine-2-0-prompts-ru - **Stars:** 9 - **Description:** Grok Imagine 2.0 prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`grok-imagine-2.0-ext`) and the cost per image. ### Primary Language: - **Python** ### GitHub Description: - **Description:** Grok Imagine 2.0 prompts: tested image prompt recipes with

### Overview: Seedream 4.0 prompts: tested image prompt recipes with example renders, aspect ratios, the model id (`seedream-4-0`), and the cost per image. ### Key features: - **8 Tested Recipes**: Eight different Seedream 4.0 prompts with example renders, aspect ratios, and the cost per image. - **Model ID**: `seedream-4-0` - **Effective Price**: $0.0195 per image, estimated cost of the set: $0.156 ### Tech stack: Python ### Why it matters: Seedream 4.0 (`seedream-4-0`) provides a versatile and cost-effective image generation service, allowing developers to easily integrate with OpenAI-compatible endpoints without the need for monthly subscriptions or additional fees. The prompts are tested and optimized for high-quality results, ensuring that developers can create professional-grade images without compromising on performance or cost.

### Overview: This repository contains code and models for exploring the relationships between fine-tuning on non-English languages and the effectiveness of pre-training on English models for language transfer learning. ### Key Features: - Reproduces pre-trained decoder-only transformers using hyperparameters from "Scaling Laws for Neural Languages" but trained on OpenWebtext2 instead of WebText. - Reproduces language transfer experiments for pre-trained English models to Chinese, Spanish, and German texts. - Includes English pre-trained models trained for 26 billion tokens with no repeats: - [x6small](https://huggingface.co/christina/decoder-only-transformer-x6small) 3.3M non-embedding parameters - [x5small](https://huggingface.co/christina/decoder-only-transformer-x5small) 16M non-embedding parameters - [x4small](https://huggingface.co/christina/decoder-only-transformer-x4small) 39M non-embedding parameters - [x3small](https://huggingface.co/christina/decoder-only-transformer-x3small) 51M non-embedding parameters - [x2small](https://huggingface.co/christina/decoder-only-transformer-x2small) 70M non-embedding parameters - [small](https://huggingface.co/christina/decoder-only-transformer-small) 124M non-embedding parameters ### Tech Stack: Python

Overview: This repository contains the codes for the paper "CAMERO: Consistency Regularized Ensemble of Perturbed Language Models with Weight Sharing" (ACL 2022). The project proposes a parameter-efficient ensemble approach for large-scale language models using consistency-regularized perturbed models with weight sharing. Key features: - Consistency Regularized Ensemble of Perturbed Language Models with Weight Sharing - Fine-tuning scripts for various language models (e.g., BERT-base, RoBERTa) - Preprocessing data and model training scripts - Hyperparameters for fine-tuning (e.g., number of models, perturbation type, consistency loss weight) Tech stack: Python Why it matters: This project is significant for researchers and developers interested in improving the performance and efficiency of large language models, particularly in the context of ensemble learning and perturbed models.

Overview: Grok Imagine Image Quality API is a powerful tool for generating stunning photorealistic images and intelligently editing existing images using text prompts. It offers a flexible pricing model and supports various models and resolutions. Key features: - Supports multiple models and resolutions - Flexible pricing model: $0.04 per image, per resolution tier - Real-time pricing and cost reporting - Easy integration with an API key - Supports image, video, and language models Tech stack: Python Why it matters: This API simplifies the process of generating high-quality images using natural language prompts, making it accessible and cost-effective for developers and creatives alike. Repository: cm2dnh/grok-imagine-image-quality-api-cn Primary language: Python GitHub stars: 9

Overview: This project is designed to create LangChain-powered applications, including a Research Assistant that generates reports using web scraping and GPT-4o-mini, and a Chat With Video app that transcribes videos and enables content-based Q&A via embeddings. Key features: - Research Assistant: Generates comprehensive research reports using web scraping and GPT-4o-mini. - Chat With Video: Transcribes videos using OpenAI's Whisper, stores embeddings in Pinecone Vector DB, and enables content-based Q&A. Tech stack: Python Why it matters: These applications leverage LangChain and OpenAI's models to automate and enhance research and content-based interaction, making them valuable tools for professionals and students.

Overview: This project introduces a new method for interacting with Large Language Models (LLMs) by utilizing screen sharing instead of typing, offering a more intuitive and natural interface. Key features: - Screen sharing for LLM interaction - Minimal GUI for background operation - API integration with various LLMs - Keyboard listener for capturing screen images Tech stack: Python Why it matters: This approach simplifies the interaction with LLMs, providing a more natural and user-friendly experience without requiring additional effort for integration. Repository: csabakecskemeti/ai-interact Primary language: Python GitHub stars: 9 GitHub description: A new way to communicate with LLMs by sharing a portion of your screen instead of typing. README excerpt: The aihub project aims to change the way we interact with LLMs by offering a more natural interface. It captures screen images for LLMs to process and extract text, enhancing the interaction experience. The project supports various platforms and requires Python 3.11.7, Tesseract, and OpenAI API-compatible LLMs.

Overview: Wan 2.7 Image API provides a single endpoint compatible with OpenAI, allowing users to generate images at a cost of $0.0216 per image. Key features: - A single endpoint for generating images - Compatible with OpenAI's API - Supports multiple image, video, and language models - Pricing model based on the number of images generated Tech stack: Python Why it matters: This API simplifies the process of generating images for developers, offering a cost-effective solution without the need for a monthly subscription.

### Overview: Grok Imagine Image API is a powerful tool for generating photorealistic images and text-to-image transformations using a single endpoint. It offers a seamless experience with a unified API key and pricing model, making it ideal for developers and creatives looking to integrate AI-powered image generation into their projects. ### Key Features: - **Unified API Key**: Use the same API key for both image generation and text-to-image transformations. - **Pricing Model**: $0.016 per image, with a minimum of $1 required for each 1K or 2K image. - **Model Compatibility**: Compatible with OpenAI's API, allowing for seamless integration with existing workflows. - **Asynchronous Processing**: Tasks are processed asynchronously, with results returned in the response. - **Model Tiers**: Available for 1K and 2K image resolutions, with tiered pricing based on resolution. ### Tech Stack: - Python ### Why It Matters: - **Cost Efficiency**: Ideal for developers who want to pay as they go, with no upfront costs or monthly subscriptions. - **Model Diversity**: Access to over 300 different models, including text-to-image and image editing capabilities. - **Integration**: Seamless integration into existing projects, with the ability to switch models without changing the client. - **Real-Time Pricing**: Prices are updated in real-time, ensuring accurate billing and transparency. ### Repository: - **Repository**: cvyxx31f3pw/grok-imagine-image-api-de - **GitHub Stars**: 9 - **GitHub Description**: Gro

Overview: The CyborgDB Go SDK is a Go client for CyborgDB, an encrypted vector database that performs search over ciphertext. It provides encryption-in-use, encrypted ANN, filters on encrypted metadata, BYOK/HYOK, and idiomatic Go API support. Key features: - Encryption-in-use: Search runs over ciphertext, only candidates are decrypted. - Encrypted ANN: Disk-backed encrypted DiskIVF index. - Filters on encrypted metadata: Combine vector similarity with equality and range predicates. - BYOK / HYOK: Wrap keys with AWS KMS or Secrets Manager. - Per-tenant key isolation: Per-index, per-user keys with cryptographic RBAC. - Idiomatic Go API: Strong typing with `context` support for cancellation and timeouts. Tech stack: Go. Why it matters: This SDK simplifies the integration of CyborgDB into Go applications, providing a secure and efficient way to perform vector search on encrypted data. Repository: cyborginc/cyborgdb-go GitHub stars: 9 GitHub description: Go Client for CyborgDB: The Encrypted Vector Database.

### Overview: SemStash is a powerful semantic storage solution designed for humans and AI agents, offering unlimited storage capacity and the ability to find content using natural language. Unlike traditional file systems, SemStash stores and retrieves data based on semantic meaning rather than file names or folder structures. ### Key Features: - **Semantic Storage**: Store any type of content—text, images, audio, video, or documents. - **Natural Language Search**: Find content by describing what you're looking for, such as "the presentation about Q3 revenue" or "photos from the beach trip." - **Command-Line Interface (CLI)**: Access SemStash through the command line for terminal workflows. - **Python Library**: Integrate SemStash into applications using the Python library. - **Web Interface**: Browse content via the web interface. - **REST API**: Use HTTP-based integrations. - **MCP Server**: Provide persistent semantic memory for AI assistants. ### Tech Stack: - **AWS S3**: Stores virtually unlimited, durable storage for files. - **Amazon S3 Vectors**: Stores vector embeddings capturing semantic meaning. - **Amazon Nova**: Generates embeddings, understanding text, images, audio, and video in a unified way. ### Why It Matters: SemStash simplifies content management by eliminating the need to remember exact file names or organize folders. It allows users to describe what they're looking for, making it easier to find relevant information across different types of content.

Overview: axonmesh is an inference decision runtime designed to optimize bandwidth and GPU usage by deciding whether a camera frame needs the cloud at all. It leverages a small model on edge devices to answer 53% of frames, sending only detections, while 47% escalate to a 11.2 KB JPEG frame using a YOLO11M model. This approach saves bandwidth and GPU resources, optimizing performance and reducing costs. Key features: - Decides whether a camera frame needs the cloud. - Uses a small model on edge devices to answer 53% of frames. - Escalates to a large model for 47% of frames. - Sends only detections for 53% of frames. - Uses a 11.2 KB JPEG frame for 47% of frames. - Saves bandwidth and GPU resources. - Optimizes performance and reduces costs. Tech stack: Python Why it matters: axonmesh is a cost-effective solution for edge-to-cloud inference, optimizing bandwidth and GPU usage by deciding whether a camera frame needs the cloud at all. This approach saves resources and reduces costs, making it ideal for scenarios where the input is produced where you control the software, and sending everything actually costs you.