Repository Issue Activity (beta)

#6883 Vulkan bf16_packed shader compile failure on shaderInt16-less GPUs

#6961 SIGFPE on linux where ncnn cannot read the CPU topology

#6873 python 版后处理例子基本上都跑不通了。。

#6932 ARM fp16 arithmetic produces all-NaN output for a ViT-tiny graph / model (ncnn 20260526)

#6765 Inference results are not deterministic due to uninitialized padding memory in Mat::create()

#6855 pnnx: bool attributes desynchronise the ncnn .bin — MemoryData declares element count but only 1 byte/element is written

#6753 feature plan

#6951 vulkan: vkCreateDescriptorPool passes VkDescriptorPoolSize with descriptorCount==0 (VUID-00302); aborts on Mesa v3dv

#6937 bug: heap-buffer-overflow (write) in Net::load_param_bin (src/net.cpp:1811)

#6939 bug: heap-buffer-overflow (write) in Spectrogram::load_param (src/layer/spectrogram.cpp:61)

#6935 Vulkan compute output corrupted on Imagination/PowerVR with opt.lightmode=1 blob recycling — allocator bookkeeping proven correct by internal tracing during a corrupted inference; barriers and submit boundaries ruled out

#6888 src/allocator.h's fastMalloc()/fastFree() pick between two mutually incompatible allocation schemes based on the preprocessor check _POSIX_C_SOURCE >= 200112L

#6791 QwenLM/Qwen3-TTS ncnn 移植与多平台部署

#6790 QwenLM/Qwen3-ASR ncnn 移植与多平台部署

#6789 TencentCloudADP/youtu-vl ncnn 移植与多平台部署

#6788 tencent-ailab/Penguin-VL ncnn 移植与多平台部署

#6787 Tencent-Hunyuan/HunyuanOCR ncnn 移植与多平台部署

#6926 [bug] MultiHeadAttention::load_param divides by zero when num_heads is 0, crashing Net::load_param on any param file that sets 1=0

#6915 bug: heap-buffer-overflow (read) in ParamDict::load_param string parsing (src/paramdict.cpp:387)

#6914 bug: heap-buffer-overflow (write) in Mat::create via ParamDict::load_param (src/mat.cpp:337)

#6913 bug: heap-buffer-overflow (read/write) in Net::load_param blob index out-of-bounds (src/net.cpp:1464)

#6911 bug: FPE (divide-by-zero) in ConvolutionDepthWise::load_param (src/layer/convolutiondepthwise.cpp:46)

#6512 ncnn KV cache mem opt ncnn 的 KVcache 模式优化

#6775 Android更新到20250503版本后偶现__kmp_abort_process崩溃

#6885 input layer changed in optimized param file