ggerganov/llama.cpp

Pull Requests Commits

hellaswag: speed up even more by parallelizing log-prob evaluation

Iwan Kawrakow committed 1 year ago

ccc78a20

perplexity : faster HellaSwag via batching (#5017)

ggerganov committed 1 year ago

Verified ad19812c

Add Winogrande evaluation (#5015)

ikawrakow committed 1 year ago

Verified 682986a0

scritps : add helper script to get hellaswag data in txt format

ggerganov committed 1 year ago

Verified dcad445d

metal : fix memory leak, dangling pointer and unused autorel (#5007)

ptsochantaris committed 1 year ago

Verified 1e605f41

ggerganov committed 1 year ago

Verified 6b6916b2

ggml : add IQ2 to test-backend-ops + refactoring (#4990)

ggerganov committed 1 year ago

Verified 38566680

imatrix : offload to GPU support (#4957)

ggerganov committed 1 year ago

Verified ba69bbc8

backend : add eval callback (#4935)

ggerganov committed 1 year ago

Verified 44a1a4a4

metal : create autorelease pool during library build (#4970)

ggerganov committed 1 year ago

Verified c918fe8d

py : fix whitespace

ggerganov committed 1 year ago

Verified 0f83e727

py : fix missing added_tokens_dict for SPM and BPE vocabs (#4971)

ggerganov committed 1 year ago

Verified 4f4bf35f

llama : use Q4_K for attn_v for Q2_K_S when n_gqa >= 4 (#4996)

ikawrakow committed 1 year ago

Verified 2b3a665d

metal : remove unnecessary nil check (#4986)

ptsochantaris committed 1 year ago

Verified 75632936

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

dwrensha committed 1 year ago

Verified f46c0c1b

py : remove unnecessary hasattr (#4903)

ggerganov committed 1 year ago

Verified 5c999609

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

philiptaron committed 1 year ago

Verified bee938da

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

danbev committed 1 year ago

Verified cec8a484

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

ikawrakow committed 1 year ago

Verified 334a835a

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

Maximilian-Winter committed 1 year ago

Verified 4feb4b33

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

ggerganov committed 1 year ago

Verified 959ef0c0

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

ggerganov committed 1 year ago

Verified c37b3474

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

ptsochantaris committed 1 year ago

Verified 158f8c9e

android : introduce starter project example (#4926)

luciferous committed 1 year ago

Verified 862f5e41

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

azarovalex committed 1 year ago

Verified 3a48d558

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

azarovalex committed 1 year ago

Verified 7c8d3abd

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

Maximilian-Winter committed 1 year ago

Verified 122ed484

ggml : introduce GGML_CALL function annotation (#4850)

jart committed 1 year ago

Verified a0b3ac8c

finetune : use LLAMA_FILE_MAGIC_GGLA (#4961)

danbev committed 1 year ago

Verified d75c232e

speculative : threading options (#4959)

stduhpf committed 1 year ago

Verified e0324285

Older