beehive-lab

Create or update the Spring AI jitllm integration (spring-ai-jitllm, spring-ai-autoconfigure-model-jitllm, spring-ai-starter-model-jitllm) for a jitllm release. Use when bumping io.github.beehive-lab:jitllm in spring-ai, adapting JitLlmChatModel to API or…

Port a new model architecture to jitllm. Use when a GGUF file is not recognized, when a family needs its own program description or lowering, or when deciding whether an existing family's implementation can be reused for a new one.

Update the Quarkus LangChain4j jitllm extension to a new jitllm release. Use when bumping io.github.beehive-lab:jitllm in quarkus-langchain4j, adapting the runtime/deployment modules to API or capability changes, validating JDK/TornadoVM combinations and…

Update the LangChain4j jitllm integration to a new jitllm release. Use when bumping io.github.beehive-lab:jitllm, adapting the adapter to API or capability changes, enabling inherited integration tests, validating JDK/TornadoVM combinations and demos, or…

Build jitllm with Maven and run it on an accelerator. Use when target/ is missing or stale, after pulling changes, or when switching JDK line or backend.

Build a TornadoVM SDK for jitllm. Use when TORNADOVM_HOME is unset, points at the wrong JDK line or backend, or the pinned version has moved.

Nsight Compute workflow for one jitllm CUDA kernel. Use after nsys, the TornadoVM profiler or the metrics output has already named the hot kernel.

Nsight Systems workflow for jitllm on the CUDA backend. Use to capture a timeline and decide where the time goes: generated kernels, launch overhead, synchronization, host-device transfers, or GPU idle.

Port a new model architecture to jllm. Use when a GGUF file is not recognized, when a family needs its own program description or lowering, or when deciding whether an existing family's implementation can be reused for a new one.

Build jllm with Maven and run it on an accelerator. Use when target/ is missing or stale, after pulling changes, or when switching JDK line or backend.

Update the Quarkus LangChain4j jllm extension to a new jllm release. Use when bumping io.github.beehive-lab:jllm in quarkus-langchain4j, adapting the runtime/deployment modules to API or capability changes, validating JDK/TornadoVM combinations and demos, or…

Build a TornadoVM SDK for jllm. Use when TORNADOVM_HOME is unset, points at the wrong JDK line or backend, or the pinned version has moved.

Nsight Compute workflow for one jllm CUDA kernel. Use after nsys, the TornadoVM profiler or the metrics output has already named the hot kernel.

Nsight Systems workflow for jllm on the CUDA backend. Use to capture a timeline and decide where the time goes: generated kernels, launch overhead, synchronization, host-device transfers, or GPU idle.

Update the LangChain4j jllm integration to a new jllm release. Use when bumping io.github.beehive-lab:jllm, adapting the adapter to API or capability changes, enabling inherited integration tests, validating JDK/TornadoVM combinations and demos, or updating…

Update the Quarkus LangChain4j GPULlama3 extension to a new GPULlama3.java release. Use when bumping io.github.beehive-lab:gpu-llama3 in quarkus-langchain4j, adapting the runtime/deployment modules to API or capability changes, validating JDK/TornadoVM…

Update the LangChain4j GPULlama3 integration to a new GPULlama3.java release. Use when bumping io.github.beehive-lab:gpu-llama3, adapting the adapter to API or capability changes, enabling inherited integration tests, validating JDK/TornadoVM combinations and…

Build GPULlama3.java from source with Maven and run. Use when the compiled jar/classes are missing or stale, or after pulling new changes.

Setup TornadoVM to use with GPULlama3.java. Use this when already on a system with GPUs visible.

Nsight Compute workflow for analyzing one specific GPULlama3/TornadoVM CUDA kernel. Use after nsys or TornadoVM profiler output identifies the hot kernel.

Nsight Systems workflow for GPULlama3/TornadoVM CUDA runs. Use to capture CUDA timelines, summarize kernel/API/memcpy time, identify hot kernels, launch overhead, syncs, and GPU idle behavior.

NVIDIA-specific TornadoVM workflows on the CUDA backend — profile with Nsight Systems (nsys) and read NVTX ranges, use the hybrid library-task API (cuBLAS/cuBLASLt/cuDNN/cuFFT/cuSPARSE/cuTENSOR/CUTLASS/NCCL native calls from Java), print the generated CUDA-C…

Build, test, and contribute to TornadoVM's OpenCL/CUDA/Metal backends — make BACKEND targets, tornado-test (quickPass/single-class), checkstyle-before-PR, one-line commits, the develop-branch PR flow, and the recurring graalJars/setvars build gotchas. Use…

Run a measured performance/correctness campaign on the TornadoVM runtime — profile first (JFR for host, Nsight Systems for driver), falsify the hypothesis with a standalone probe, A/B against a same-build baseline, then ship one PR per measured claim. Use…