ggerganov/llama.cpp

Pull Requests Commits

cuda : fix multi-seq, quantized FA

ggerganov committed 1 year ago

55cf48de

tests : add non-cont K,V FA tests

ggerganov committed 1 year ago

Verified a856a566

model : fix build after merge conflict (#14754)

ggerganov committed 1 year ago

Verified eacdeb5b

model : add EXAONE 4.0 support (#14630)

lgai-exaone committed 1 year ago

Verified e0cb5c5c

CUDA: set_rows + cpy.cu refactor (#14712)

am17an committed 1 year ago

Verified f9a31eea

graph : refactor context to not pass gf explicitly (#14629)

ggerganov committed 1 year ago

Verified 8f974bc1

graph : Pass the graph placeholder message in debug mode (#14748)

Nexesenex committed 1 year ago

Verified 09651d09

use max work group size for device to replace the magic number (#14732)

NeoZhangJianyu committed 1 year ago

Verified 349ea79f

convert : fix Ernie4.5 MoE without shared experts (#14746)

pwilkin committed 1 year ago

Verified 670e1360

nix : use optionalAttrs for env mkDerivation attrset argument (#14726)

amozeo committed 1 year ago

Verified 760b4484

model: add Ernie 4.5 MoE support (#14658)

pwilkin committed 1 year ago

Verified cb887f1b

kv-cache : fix k-shift for multiple streams (#14742)

ggerganov committed 1 year ago

Verified d6fb3f6b

llama : reuse compute graphs (#14482)

ggerganov committed 1 year ago

Verified 01612b74

llama : fix parallel processing for lfm2 (#14705)

tdakhran committed 1 year ago

Verified 086cf81e

kv-cache : opt mask set input (#14600)

ggerganov committed 1 year ago

Verified d9b69108

batch : fix uninitialized has_cpl flag (#14733)

ggerganov committed 1 year ago

Verified ad57d3ed

ci : disable failing vulkan crossbuilds (#14723)

CISC committed 1 year ago

Verified 1ba45d49

convert : make hf token optional (#14717)

CISC committed 1 year ago

Verified 19e5943d

llama : fix parameter order for hybrid memory initialization (#14725)

dinerburger committed 1 year ago

Verified 496957e1

ggml: Add initial WebGPU backend (#14521)

reeselevine committed 1 year ago

Verified 21c02174

model : support output bias for qwen2 (#14711)

tempstudio committed 1 year ago

Verified b0f0ecc3

llama : add high-throughput mode (#14363)

ggerganov committed 1 year ago

Verified 225e7a14

Support diffusion models: Add Dream 7B (#14644)

am17an committed 1 year ago

Verified ab140198

ggml : add asserts (#14720)

ggerganov committed 1 year ago

Verified 64978340

server : pre-calculate EOG logit biases (#14721)

ggerganov committed 1 year ago

Verified 6ffd4e9c

llama : fix parallel processing for plamo2 (#14716)

mitmul committed 1 year ago

Verified e4841d24

server : fix handling of the ignore_eos flag (#14710)

ggerganov committed 1 year ago

Verified 538cc77f

scripts: synthetic prompt mode for server-bench.py (#14695)

JohannesGaessler committed 1 year ago

Verified 5cae7665

convert : only check for tokenizer folder if we need it (#14704)

CISC committed 1 year ago

Verified 4b91d6f7

convert : add pre-computed hashes first to prevent order mishaps (#14701)

CISC committed 1 year ago

Verified cf91f217

Older