ggml-org/llama.cpp

Pull Requests Commits

Use Q4_K for attn_v for Q2_K_S when n_gqa >= 4

Iwan Kawrakow committed 2 years ago

9fd1e83f

metal : remove unnecessary nil check (#4986)

ptsochantaris committed 2 years ago

Verified 75632936

llama : fix copy/paste error in llama_sampling_params comment (#4994)

dwrensha committed 2 years ago

Verified f46c0c1b

py : remove unnecessary hasattr (#4903)

ggerganov committed 2 years ago

Verified 5c999609

nix: remove nixConfig from flake.nix (#4984)

philiptaron committed 2 years ago

Verified bee938da

finetune : add training data file to log message (#4979)

danbev committed 2 years ago

Verified cec8a484

ggml : importance matrix support for legacy quants (#4969)

ikawrakow committed 2 years ago

Verified 334a835a

examples : add complete parallel function calling example (#4974)

Maximilian-Winter committed 2 years ago

Verified 4feb4b33

perplexity : fix kv cache handling for hellaswag (#4981)

ggerganov committed 2 years ago

Verified 959ef0c0

flake.lock: update flake-parts, flake-parts/nixpkgs-lib, and nixpkgs (#4920)

ggerganov committed 2 years ago

Verified c37b3474

metal : localized logic in `ggml_metal_graph_compute` (#4924)

ptsochantaris committed 2 years ago

Verified 158f8c9e

android : introduce starter project example (#4926)

luciferous committed 2 years ago

Verified 862f5e41

metal : replace loop of dispatch_async with dispatch_apply (#4934)

azarovalex committed 2 years ago

Verified 3a48d558

metal : log `recommendedMaxWorkingSetSize` on iOS 16+ (#4936)

azarovalex committed 2 years ago

Verified 7c8d3abd

examples : fix and improv docs for the grammar generator (#4909)

Maximilian-Winter committed 2 years ago

Verified 122ed484

ggml : introduce GGML_CALL function annotation (#4850)

jart committed 2 years ago

Verified a0b3ac8c

finetune : use LLAMA_FILE_MAGIC_GGLA (#4961)

danbev committed 2 years ago

Verified d75c232e

speculative : threading options (#4959)

stduhpf committed 2 years ago

Verified e0324285

pass cpu-architecture arguments only to host code (C;C++) (#4943)

ngc92 committed 2 years ago

Verified 3e5ca793

llama : apply classifier-free guidance to logits directly (#4951)

dfriehs committed 2 years ago

Verified 44833967

awq-py : fix typo in awq-py/README.md (#4947)

ziliangpeng committed 2 years ago

Verified d9aa4ffa

cuda : fix dequantize kernel names (#4938)

ggerganov committed 2 years ago

Verified ddb008d8

llama : check for 256 divisibility for IQ2_XS, IQ2_XXS (#4950)

ikawrakow committed 2 years ago

Verified 2faaef39

CUDA: faster dequantize kernels for Q4_0 and Q4_1 (#4938)

ikawrakow committed 2 years ago

Verified 4a3156de

llama : fix missing quotes (#4937)

David Pflug committed 2 years ago

Verified a836c8f5

Add ability to use importance matrix for all k-quants (#4930)

ikawrakow committed 2 years ago

Verified 467a882f

llama : check LLAMA_TRACE env for extra logging (#4929)

ggerganov committed 2 years ago

Verified bb0c1392

scripts : sync-ggml-am.sh option to skip commits

ggerganov committed 2 years ago

Verified 9408cfda

llama : use LLAMA_LOG_ macros for logging

ggerganov committed 2 years ago

Verified 03c52674

Fix ffn_down quantization mix for MoE models (#4927)

ikawrakow committed 2 years ago

Verified a128c38d

Older