Formed in 2009, the Archive Team (not to be confused with the archive.org Archive-It Team) is a rogue archivist collective dedicated to saving copies of rapidly dying or deleted websites for the sake of history and digital heritage. The group is 100% composed of volunteers and interested parties, and has expanded into a large amount of related projects for saving online and digital history.
ggerganov
Follow
Pinned
2,079 contributions in the last year
Contribution activity
July 2023
Created 67 commits in 4 repositories
Created 1 repository
Created a pull request in ggerganov/llama.cpp that received 1 comment
Opened 10 other pull requests in 4 repositories
ggerganov/llama.cpp
5
merged
1
closed
ggerganov/ggml
2
merged
ggerganov/whisper.cpp
1
open
evanmiller/llama.cpp
1
merged
Reviewed 66 pull requests in 4 repositories
ggerganov/llama.cpp
25 pull requests
- Fix #2221, use pkg-config
- Allocate all temporary ggml_tensor_extra_gpu from a fixed-size buffer
- Fix static_assert with older compilers
- Add functions that work directly on model
- build.zig: install config header
- examples: fixed path typos in embd-input
- Support broadcast add & mul on CUDA (fixed)
- Enable LLAMA_METAL and LLAMA_MPI in Makefile
- fixed runtime bugs and compile errors related to GGML_PERF and GGML_DEBUG
- Metal: faster Q4_0 and Q4_1 matrix x vector kernels
- metal: new q4_0 mat-vec mul kernel
- Add missing quotes to bash script
- Fixed __dp4a compute capability: 6.0 -> 6.1
- ggml : remove src0 and src1 from ggml_tensor and rename opt to src
- Implement classifier-free guidance
- Distributed inference via MPI
- Remove "first token must be BOS" restriction
- Escape prompt prefix/suffix along with prompt
- Possible solution to allow K-quants on models with n_vocab!=32000
- Update README.md
- Fix buidling with Intel MKL but ask for "cblas.h" issue (#2104)
- ggml : change ggml_graph_compute() API to not require context
- update for baichuan
-
Update model file name in
examples/alpaca.sh - Expose generation timings from server & update completions.js
- Some pull request reviews not shown.
ggerganov/ggml
25 pull requests
- Support GLM-style RoPE on CUDA
- Added vector scaling using Accelerate
- ggml : fix mul_mat src1 indexing when src1 is not contiguous
- Batch inference
- ggml : basic implementation of 1d and 2d pools
- callback to abort ggml_graph_compute()
- Build on Windows fixed (dynamic linking of stdc++)
- Use a single kernel for cuda mul op
- tests : ifdef for #pragma GCC
- Support broadcast addition for fp32
- Support norm op on CUDA
- Fix for starcoder ggml-model.bin being saved in wrong directory. Moderrnise by using argparse.
- Fix for #329
- Correct Alibi implementation
- Add <|end_of_turn|> token handling to starcoder in order to support openchat/opencoderplus
- ggml : restore GGML_RESTRICT
- build: Add pkg-config file
- cmake: Install the header file to ggml/ggml.h
- examples : use GGML_FILE_MAGIC where possible
- zig : add tests codes using zig
- Add interactive prompt and port mode for dollyv2
- MPT-30B: Convert model weights part by part to save memory
-
ggml: add ops in
encodec.cpp - ggml : reduce code duplication with helper macros for local variables
- Starcoder repeat penalty
ggerganov/whisper.cpp
4 pull requests
ggerganov/ggwave
1 pull request
Created an issue in ggerganov/llama.cpp that received 18 comments
mpi : attempt inference of 65B LLaMA on a cluster of Raspberry Pis
Now that distributed inference is supported thanks to the work of @evanmiller in #2099 it would be fun to try to utilize it for something cool. One…
18
comments