Taylor Yeonbok Lee

Results 12 issues of Taylor Yeonbok Lee

### Details: - *item1* - *...* ### Tickets: - *ticket-id*

category: GPU

### Details: - Previously, memory conflict check was done using std::string (primitive_id), and it was time consuming - Fixed to use unique_id as mem_dep, instead of std::string ### Tickets: -...

category: GPU

### Details: - Optimize memory pool by reducing conflict check time ### Tickets: - 140135

category: GPU

### Details: - Not to assign a memory slot for a memory request < 50% of the available memory ### Tickets: - 137490

category: GPU
do not merge

### Details: - Relaxed memory reclaim policy - Fixed memory pool not to assign too big memory to much smaller request ### Tickets: - 137490 - 132376

category: GPU
do not merge
under_perf_check

Hello, I followed the instruction (install opencl-icd and set library path to it) (llm) D:\dev\taylor\pti-gpu\tools\onetrace\build>cmake -G "NMake Makefiles" -DCMAKE_BUILD_TYPE=Release -DCMAKE_LIBRARY_PATH=D:\dev\taylor\OpenCL-ICD-Loader\install .. However the build was crashed as below: ![image](https://github.com/intel/pti-gpu/assets/25142117/dc8d0746-67e9-4850-a9dd-6488d284498f) Do...

### Details: - Fixed crash & accuracy issue in beam search scenario when initial input is not batched ### Tickets: - 140755

category: GPU

### Details: - Reorder which is only converting type can be fused to the prior node ### Tickets: - 144957

category: GPU

### Details: - Fuse QKV FCs to one FC ### Tickets: - 142815

category: Core
category: GPU

### Details: - Added detailed description about the kv cache prealloc policy ### Tickets: - *ticket-id*

category: GPU