url/colibri
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

windows-dev: CUDA build script, expert budget + grouped quant research

Browse Source 
- build_cuda.bat: one-shot nvcc DLL compilation for sm_120 (RTX 5070 Ti)
- issue_budget.md: EXPERT_BUDGET research (cap distinct experts/layer, 2x tok/s)
- issue_grouped_quant.md: root cause of int4 incoherence (per-row vs group-128 scales)
- glm_fp8_emit.py: FP8 e4m3 test weight generator for converter validation
This commit is contained in:
woolcoxm
2026-07-15 01:04:32 -04:00
parent 2c6946c478
commit e71d4fbe29
3 changed files with 340 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
c/build_cuda.bat
+5
View File
@@ -0,0 +1,5 @@
@echo off
call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"
cd /d C:\Users\Mark\Desktop\Projects\colibri\c
"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\bin\nvcc" -O3 -std=c++17 -arch=sm_120 -Xcompiler=-W3 -shared -DCOLI_CUDA_BUILDING_DLL -L"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\lib/x64" -lcudart backend_cuda.cu -o coli_cuda.dll
echo EXITCODE=%ERRORLEVEL%