ggerganov/llama.cpp

Pull Requests Commits

cuda : ggml_cuda_op_mul_mat_cublas support F32 precision

ggerganov committed 1 year ago

Verified a462159c

Update ggml-cuda.cu

ggerganov committed 1 year ago

Verified 30338c56

Update ggml-cuda.cu

ggerganov committed 1 year ago

Verified 3c8d6b16

ggml : add ggml_mul_mat_set_prec

ggerganov committed 1 year ago

Verified 18c67bdd

Merge branch 'master' into HEAD

ggerganov committed 1 year ago

Verified a8d2a6f3

decode : fix logits_valid for legacy API (#4516)

cebtenzzre committed 1 year ago

Verified 2994f0c5

readme : update hot topics

ggerganov committed 1 year ago

Verified b1306c43

llama.swiftui : add bench functionality (#4483)

ggerganov committed 1 year ago

Verified 800a489e

gguf-py : fail fast on nonsensical special token IDs (#4489)

cebtenzzre committed 1 year ago

Verified f7f468a9

build : Check the ROCm installation location (#4485)

Su3h7aM committed 1 year ago

Verified 919c4066

finetune : keep allocs alive until all allocations are done (#4486)

slaren committed 1 year ago

Verified 45668633

server : disable llm logs if SERVER_VERBOSE is off (#3792)

olexiyb committed 1 year ago

Verified 0ffc92d2

server : fix grammar being ignored (#4494)

AdithyanI committed 1 year ago

Verified 8edd2b40

server : fix possible ambiguity in content type charset (#4501)

z80maniac committed 1 year ago

Verified eb16dae7

server : allow requests larger than 8K (#4500)

mzcu committed 1 year ago

Verified 62bd52b7

Link to cublas dynamically on Windows even with LLAMA_STATIC (#4506)

bullno1 committed 1 year ago

Verified 5daa5f54

cuda : less diff in the rope_neox kernel

ggerganov committed 1 year ago

Verified 42e95258

ggml : fix NeoX rope to rotate just first n_dims

ggerganov committed 1 year ago

Verified f703ca8a

lora : add support for non-llama models (#3333)

slaren committed 1 year ago

Verified c6c4fc08

phi-2 : scale Q instead of KQ for better precision

ggerganov committed 1 year ago

Verified 0644c3be

convert : revert "added_tokens_decoder" change

ggerganov committed 1 year ago

Verified 0b6ffa58

convert : phi don't add BOS token

ggerganov committed 1 year ago

Verified a878be4c

llama : fix meta KV override bug

ggerganov committed 1 year ago

Verified 5469d82d

py : whitespaces

ggerganov committed 1 year ago

Verified 7500fa2f

phi-2 : use layer norm eps

ggerganov committed 1 year ago

Verified aa5c881a

phi-2 : various fixes

ggerganov committed 1 year ago

Verified a2a3d2c8

llama : sanity checks for access to logits (#4274)

cebtenzzre committed 1 year ago

Verified 8a5be3bd

fix breaking change

eabraham-1 committed 1 year ago

e2076553

phi2 implementation

eabraham-1 committed 1 year ago

12cc80cb

server : add optional API Key Authentication example (#4441)

ShadovvBeast committed 1 year ago

Verified 88ae8952

Older