You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi @saudet@HGuillemet javacpp team,
I’d like to share my recent work and progress on a series of new JavaCPP-based toolchains.
These are all fully functional or nearly complete, and I plan to donate and migrate them to the Bytedeco organization and publish them to Maven Central later.
Here’s what has been built:
javacpp-open3d
Fully compiled and working (Linux x86-64). Built almost entirely with large models, and Claude Code helped solve many critical issues. This will enable robotics, GIS, and point cloud processing in Java.
2.javacpp-tensorRT-llm
Full Java glue code completed. Platform generation still has small issues I’m fixing with Claude Code. This will become the ultimate LLM deployment platform for JavaCPP.
3.javacpp-pytorch-geometric (Java)
Full GNN framework implemented with many benchmarks. Production-ready.
4.javacpp-pytorch-geometric (Scala 3)
Full GNN framework in Scala 3, also with benchmarks. Production-ready.
5.javacpp-pytorch-distribution-rl (Java & Scala 3)
Implemented 39 probability distributions required for reinforcement learning.
6.javacpp-pytorch-rl (Java & Scala 3)
Deep reinforcement learning framework modeled after HuggingFace TRL. Basically usable.
7.javacpp-pytorch-distribute-train (DDP / FSDP)
Full distributed training framework supporting NCCL and Gloo.
Only blocked by Module.forward() overloads — otherwise already perfect.
8.javacpp-cuda-triton-dsl
A major breakthrough. Built in 2 weeks, comparable to OpenAI Triton for writing high-performance CUDA kernels.
Verified FlashAttention, PageAttention — industrial level quality.
9.javacpp-pytorch-cuda-vllm-nano
LLM inference server inspired by vllm-nano. Supports KV cache. Not fully finished yet, but will be excellent thanks to javacpp-cuda-triton-dsl.
10.javacpp-safetensors-module
Directly maps Hugging Face safetensors models to JavaCPP PyTorch Module.
Also supports:
numpy / pickle / gguf read/write
pandas Java version
javacpp-transformers
Implementation similar to Hugging Face Transformers.
Model loading, downloading, and architecture tracking work — still being polished but fully functional.
All these projects are being cleaned up and organized.
I believe they will greatly enrich the JavaCPP ecosystem and make it stronger.
Also, I have a GitHub account:
mullerhe
(created for GitHub Copilot)
which contains many projects I want to migrate into Bytedeco.
Please add this account to the Bytedeco organization when you have time.
Thank you!
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi @saudet @HGuillemet javacpp team,
I’d like to share my recent work and progress on a series of new JavaCPP-based toolchains.
These are all fully functional or nearly complete, and I plan to donate and migrate them to the Bytedeco organization and publish them to Maven Central later.
Here’s what has been built:
Fully compiled and working (Linux x86-64). Built almost entirely with large models, and Claude Code helped solve many critical issues. This will enable robotics, GIS, and point cloud processing in Java.
2.javacpp-tensorRT-llm
Full Java glue code completed. Platform generation still has small issues I’m fixing with Claude Code. This will become the ultimate LLM deployment platform for JavaCPP.
3.javacpp-pytorch-geometric (Java)
Full GNN framework implemented with many benchmarks. Production-ready.
4.javacpp-pytorch-geometric (Scala 3)
Full GNN framework in Scala 3, also with benchmarks. Production-ready.
5.javacpp-pytorch-distribution-rl (Java & Scala 3)
Implemented 39 probability distributions required for reinforcement learning.
6.javacpp-pytorch-rl (Java & Scala 3)
Deep reinforcement learning framework modeled after HuggingFace TRL. Basically usable.
7.javacpp-pytorch-distribute-train (DDP / FSDP)
Full distributed training framework supporting NCCL and Gloo.
Only blocked by Module.forward() overloads — otherwise already perfect.
8.javacpp-cuda-triton-dsl
A major breakthrough. Built in 2 weeks, comparable to OpenAI Triton for writing high-performance CUDA kernels.
Verified FlashAttention, PageAttention — industrial level quality.
9.javacpp-pytorch-cuda-vllm-nano
LLM inference server inspired by vllm-nano. Supports KV cache. Not fully finished yet, but will be excellent thanks to javacpp-cuda-triton-dsl.
10.javacpp-safetensors-module
Directly maps Hugging Face safetensors models to JavaCPP PyTorch Module.
Also supports:
numpy / pickle / gguf read/write
pandas Java version
javacpp-transformers
Implementation similar to Hugging Face Transformers.
Model loading, downloading, and architecture tracking work — still being polished but fully functional.
All these projects are being cleaned up and organized.
I believe they will greatly enrich the JavaCPP ecosystem and make it stronger.
Also, I have a GitHub account:
mullerhe
(created for GitHub Copilot)
which contains many projects I want to migrate into Bytedeco.
Please add this account to the Bytedeco organization when you have time.
Thank you!
Beta Was this translation helpful? Give feedback.
All reactions