 Michael Klimenko
|
52bb63c708
refactor : switch to emplace_back to avoid extra object (#5291)
|
há 1 ano atrás |
 kalomaze
|
191221178f
perplexity : fix KL divergence calculations on Windows (#5273)
|
há 1 ano atrás |
 Kawrakow
|
44879ee885
Additional KL-divergence statistics (#5081)
|
há 2 anos atrás |
 Georgi Gerganov
|
89758723c7
minor : clean-up some warnings and style (#5094)
|
há 2 anos atrás |
|
Kawrakow
|
6f9939d119
KL-divergence (#5076)
|
há 2 anos atrás |
|
Kawrakow
|
7dcbe39d36
Add ability to evauate multiple choice tasks (#5047)
|
há 2 anos atrás |
 Jared Van Bortel
|
97c1549808
perplexity : fix MSVC build after #5020 (#5043)
|
há 2 anos atrás |
|
Kawrakow
|
7051aacfac
winogrande: evaluate log-probs in parallel (#5036)
|
há 2 anos atrás |
|
Kawrakow
|
993fba8180
perplexity: avoid unnecessary alloocations and logit copies (#5035)
|
há 2 anos atrás |
|
Georgi Gerganov
|
8b20858e5e
perplexity : faster Winogrande via batching (#5024)
|
há 2 anos atrás |
|
Georgi Gerganov
|
d391ae9b49
perplexity : fix winogrande N tasks option
|
há 2 anos atrás |
|
Kawrakow
|
3e945cc1e9
HellaSwag: speed up by parallelizing log-prob evaluation (#5020)
|
há 2 anos atrás |
|
Georgi Gerganov
|
ad19812cda
perplexity : faster HellaSwag via batching (#5017)
|
há 2 anos atrás |
|
Kawrakow
|
682986a08e
Add Winogrande evaluation (#5015)
|
há 2 anos atrás |
|
Georgi Gerganov
|
959ef0c0df
perplexity : fix kv cache handling for hellaswag (#4981)
|
há 2 anos atrás |
 Kerfuffle
|
91f6499393
Respect tokenizer.ggml.add_bos_token value when tokenizing (#4040)
|
há 2 anos atrás |
 cebtenzzre
|
b12fa0d1c1
build : link against build info instead of compiling against it (#3879)
|
há 2 anos atrás |
|
Kerfuffle
|
6e08281e58
Extend llama_kv_cache_seq_rm to allow matching any sequence (#3843)
|
há 2 anos atrás |
 Marcus Dunn
|
5be6c803fa
llama : remove token functions with `context` args in favor of `model` (#3720)
|
há 2 anos atrás |
 slaren
|
16bc66d947
llama.cpp : split llama_context_params into model and context params (#3301)
|
há 2 anos atrás |
|
Georgi Gerganov
|
ec893798b7
llama : custom attention mask + parallel decoding + no context swaps (#3228)
|
há 2 anos atrás |
|
Cebtenzzre
|
8781013ef6
make : restore build-info.h dependency for several targets (#3205)
|
há 2 anos atrás |
|
Cebtenzzre
|
e6616cf0db
examples : add compiler version and target to build info (#2998)
|
há 2 anos atrás |
|
Cebtenzzre
|
3aefaab9e5
check C++ code with -Wmissing-declarations (#3184)
|
há 2 anos atrás |
|
Cebtenzzre
|
e64f5b5578
examples : make n_ctx warning work again (#3066)
|
há 2 anos atrás |
|
Cebtenzzre
|
00d62adb79
fix some warnings from gcc and clang-tidy (#3038)
|
há 2 anos atrás |
|
Georgi Gerganov
|
e36ecdccc8
build : on Mac OS enable Metal by default (#2901)
|
há 2 anos atrás |
|
Kawrakow
|
fa3582f509
Tell users attmepting to run perplexity with too few tokens to use more (#2882)
|
há 2 anos atrás |
 Johannes Gäßler
|
6b73ef1201
YAML result logging + preset script (#2657)
|
há 2 anos atrás |
|
Kawrakow
|
463173a6c0
llama : speedup tokenization (#2831)
|
há 2 anos atrás |
