Issues

• Memory preflight misses the MLX prefill for Llama Q4_0, and jitllm bench skips the preflight

  bug
  beehive-lab/jitllm#215 · Filter by author orionpapadakis opened on Oct 15, 2026

• MLX prefill: affine Q8_0/Q4_0 GEMMs lose precision at narrow chunks (MLX split-K keeps FP16 partial sums)

  prefill-decodequantization
  beehive-lab/jitllm#214 · Filter by author orionpapadakis opened on Oct 15, 2026

• MLX prefill: quantized projections are kept twice on the device and once more on the host

  enhancementquantization
  beehive-lab/jitllm#213 · Filter by author orionpapadakis opened on Oct 15, 2026

• MLX prefill: FP32/FP16 conversion kernels around every MLX call cost about twice the MLX kernels

  enhancementprefill-decode
  beehive-lab/jitllm#212 · Filter by author orionpapadakis opened on Oct 15, 2026

• MLX prefill: the K/V gather is launched over the whole context capacity, so prefill slows down as --ctx-size grows

  enhancementprefill-decode
  beehive-lab/jitllm#211 · Filter by author orionpapadakis opened on Oct 15, 2026

• Chunked batch prefill: chunks after the first are 5-14x slower (JIT paged attention re-streams KV per query)

  beehive-lab/jitllm#204 · Filter by author orionpapadakis opened on Oct 13, 2026

• Native-library prefill silently falls back to JIT attention when cuDNN is unavailable or declines the shape

  beehive-lab/jitllm#203 · Filter by author orionpapadakis opened on Oct 13, 2026

• CPU Q8_0/Q4_0/Q4_1 dot products throw on 512-bit (AVX-512) vector species

  beehive-lab/jitllm#201 · Filter by author kotselidis opened on Oct 8, 2026

• Framework integration compatibility for jitllm 1.0.0: LangChain4j, Spring AI, Quarkus

  1.0.0
  beehive-lab/jitllm#178 · Filter by author mikepapadim opened on Sep 25, 2026
  2

  

• Support the Qwen3.5 architecture

  beehive-lab/jitllm#143 · Filter by author PhillHenry opened on Aug 2, 2026
  1

  

• Reduce external implementation types exposed by the GPULlama3 integration API

  1.0.1good first issuerefactoring
  Task

  beehive-lab/jitllm#141 · Filter by author orionpapadakis opened on Jul 30, 2026

• Roadmap: vLLM-style serving features for TornadoVM batched decode

  beehive-lab/jitllm#130 · Filter by author mikepapadim opened on Jul 11, 2026
  1

  

• Support direct execution of K-quant GGUF weights on GPU (Q4_K, Q5_K, Q6_K)

  enhancementquantization
  Feature

  beehive-lab/jitllm#118 · Filter by author orionpapadakis opened on May 29, 2026

• [POC] Minimal Kernel Autotuning Support

  enhancement
  Feature

  beehive-lab/jitllm#96 · Filter by author mikepapadim opened on Dec 22, 2025
  

• TornadoTaskRuntimeException when using Phi-3-mini-4k-instruct-fp16.gguf during TornadoVM initialization

  bug
  Bug

  beehive-lab/jitllm#72 · Filter by author yrq0208 opened on Nov 25, 2025
  5

  

• support for llama3.2 vision

  question
  beehive-lab/jitllm#43 · Filter by author ahmedBouchaibi opened on Aug 5, 2025
  1

• OutOfMemory / Misleading Hint

  documentationenhancement
  Bug

  beehive-lab/jitllm#30 · Filter by author AdamBien opened on Jun 19, 2025
  1

  

• Build GUI Chatbox for GPULlama3.java Inference with embedded GPU resource monitoring

  good first issueTooling
  Feature

  beehive-lab/jitllm#24 · Filter by author mikepapadim opened on Jun 13, 2025
  2

• Set Up Linters and Code Style Checkers

  good first issue
  beehive-lab/jitllm#16 · Filter by author mikepapadim opened on Jun 9, 2025
  2