ggerganov
Follow
Pinned
1,808 contributions in the last year
Contribution activity
June 2023
Created 25 commits in 3 repositories
Created a pull request in ggerganov/llama.cpp that received 2 comments
k-quants : allow to optionally disable at compile time
The new quantization types Q2_K, Q3_K, Q4_K, Q5_K and Q6_K can now optionally be disabled at compile time:
make:
LLAMA_NO_K_QUANTS=1 make
CMake:
cm…
+251
−229
•
2
comments
Opened 2 other pull requests in 2 repositories
ggerganov/llama.cpp
1
merged
ggerganov/ggml
1
open
Reviewed 35 pull requests in 3 repositories
ggerganov/llama.cpp
25 pull requests
- Train Text from scratch
- Fix issue with ggml-metal.metal path. Closes #1769
- Allow "quantizing" to f16 and f32
- doc: fix the wrong address of BLIS.md
- Updated CMakeLists.txt slightly to allow building this project in Android Studio with the latest Android NDK
- metal : add Q4_1 implementation
- docker: add support for CUDA in docker
- Support requantizing models instead of only allowing quantization from 16/32bit
- fix(avx): workaround for missing _mm256_set_m128i in GCC < 8
- Add detection code for avx/avx2/etc
- Numa
- Update Makefile to add SSSE3 compilation use cases
- Windows nvcc workaround
- Add gelu implementation to metal.
- metal : faster q40
- metal : add Q2_K implementation
- Q6_K implementation for Metal
- Fix warning on fprintf
- Q4_K implementation for Metal
- Add missing compile definition to CMakeLists for k_quants
- update flake to support metal on m1/m2
- Multi GPU support, CUDA refactor, CUDA scratch buffer
- Add checks for buffer size with Metal
- docs(performance): Add performance troubleshoot + example benchmark documentation
- fix small typo in README.md
- Some pull request reviews not shown.
ggerganov/ggml
4 pull requests
ggerganov/whisper.spm
1 pull request
Started 1 discussion in 1 repository
ggerganov/llama.cpp
ggerganov/llama.cpp
-
Roadmap June 2023
This contribution was made on Jun 7