ggerganov/llama.cpp

Pull Requests Commits
wip
ggerganov committed 1 year ago
Verified 6ccbd177
wip
ggerganov committed 1 year ago
Verified af3eda9c
wip
ggerganov committed 1 year ago
Verified 5cbdba69
wip
ggerganov committed 1 year ago
Verified 035c4f01
wip
ggerganov committed 1 year ago
Verified 06c2d0d1
metal : parallel reduce across heads
ggerganov committed 1 year ago
Verified 17720fad
metal : parallelize across KV size
ggerganov committed 1 year ago
Verified 77d08f32
wip : template for rows per warp
ggerganov committed 1 year ago
Verified a4b6341c
wip : 4 rows per simd group
ggerganov committed 1 year ago
Verified f31955f5
wip : 8 rows per simd group
ggerganov committed 1 year ago
Verified 8cde449b
metal : specialize for head size
ggerganov committed 1 year ago
Verified b9732580
metal : reduce branches
ggerganov committed 1 year ago
Verified 52ae0857
metal : f16 precision
ggerganov committed 1 year ago
Verified 528da751
metal : initial implementation
ggerganov committed 1 year ago
Verified 1173f49c
ggml : online attention (CPU)
ggerganov committed 1 year ago
Verified a9681feb
Merge branch 'master' into gg/flash-attn
ggerganov committed 1 year ago
Verified c3cdfffa
cuda : fix compile error in jetson platform (#4975)
KyL0N committed 1 year ago
Verified cca894f1
finetune : fix ggml_allocr lifetimes (tmp workaround) (#5033)
bzuzo committed 1 year ago
Verified 381ee195
ggml : fix GQA support in ggml_flash_attn_ext
ggerganov committed 1 year ago
fa7ebcca
imatrix : add README.md
ggerganov committed 1 year ago
Verified a5cacb22
llama : support upcoming Qwen2 (#5037)
simonJJJ committed 1 year ago
Verified 9b75cb2b
py : fix flake8 lint
ggerganov committed 1 year ago
de9a147d
winogrande: evaluate log-probs in parallel (#5036)
ikawrakow committed 1 year ago
Verified 7051aacf
llama : add CodeShell support (#5016)
chiranko committed 1 year ago
Verified 2b3b999c
perplexity: avoid unnecessary alloocations and logit copies (#5035)
ikawrakow committed 1 year ago
Verified 993fba81
perplexity : faster Winogrande via batching (#5024)
ggerganov committed 1 year ago
Verified 8b20858e
llama : fix falcon arch for tied output embeddings (#4978)
cmp-nct committed 1 year ago
Verified 57e2a7a5
cmake : add ggml public headers (#5011)
ggerganov committed 1 year ago
Verified 9b6ea426
server : defer tasks when "slot unavailable" (#5018)
ngxson committed 1 year ago
Verified 821f0a27
llama : fix mlock with no-mmap with Metal (#5025)
slaren committed 1 year ago
Verified 96d7f56d
Terms Of Use Privacy Policy Refund Policy Impressum