Code

split out rllm common from rllm-cuda#51

Code

Merged

mmoskal merged 56 commits intomainmicrosoft/aici:mainfrom traitsmicrosoft/aici:traitsCopy head branch name to clipboardFeb 12, 2024

+2,660-2,060

Commits

Commits on Feb 5, 2024

• start #ifdef (cfg) -> trait conversion

  mmoskal

  committed
  d17dfc2

• fix warnings

  mmoskal

  committed
  119af1c

• force GPU layers when --cuda on llama.cpp

  mmoskal

  committed
  4b74440

• parametric RllmConfig

  mmoskal

  committed
  3f39e75

• splitting out model loader args

  mmoskal

  committed
  0d388f5

Commits on Feb 6, 2024

• add sequence manager

  mmoskal

  committed
  19210e0

• cleanup

  mmoskal

  committed
  547b161

• use traits in tch version

  mmoskal

  committed
  83f55fd

• fix warnings

  mmoskal

  committed
  662fd23

• move paged into llm/

  mmoskal

  committed
  b0254f8

• limit usage of llm::

  mmoskal

  committed
  e8e32d1

• cleanup

  mmoskal

  committed
  594c8b0

• default unique --shm-prefix

  mmoskal

  committed
  9d336b3

Commits on Feb 7, 2024

• keep block mapping in BlockSpaceManager

  mmoskal

  committed
  535f5a0

• make cpp version build again

  mmoskal

  committed
  832abbd

• feature-containment

  mmoskal

  committed
  f2e22d7

• remove last #[cfg]

  mmoskal

  committed
  12ae462

• split cpp/cuda code into folders

  mmoskal

  committed
  469cf64

• make it compile again

  mmoskal

  committed
  6c50da4

• fix warnings

  mmoskal

  committed
  385e8c3

• fix start script

  mmoskal

  committed
  232d720

• disable ccache action (runs out of space)

  mmoskal

  committed
  e0c4d39

• rename rllm-server to rllm-cuda

  mmoskal

  committed
  6a233b3

• custom md link check

  mmoskal

  committed
  0ae83cc

• Merge remote-tracking branch 'origin/main' into traits

  mmoskal

  committed
  d390f1c

• fix links

  mmoskal

  committed
  07b5e12

• fix build script

  mmoskal

  committed
  ef7c432

• disable all build caching

  mmoskal

  committed
  90fcc46

• diagnose space problem

  mmoskal

  committed
  bd915b6

Commits on Feb 8, 2024

• update llama.cpp

  mmoskal

  committed
  7de3e0b

• try a hack for M1

  mmoskal

  committed
  963dffa

• disable failing cuda build

  mmoskal

  committed
  0cc32ed

• Revert "disable all build caching"

  Show description for b0bf6ba
  mmoskal

  committed
  b0bf6ba

Commits on Feb 9, 2024

• remove unused deps

  mmoskal

  committed
  8bf1ddb

• start moving files around; fixes #56

  mmoskal

  committed
  3434490

• update some file refs

  mmoskal

  committed
  1597110

• more folder renames

  mmoskal

  committed
  729a223

• updating links

  mmoskal

  committed
  16adf25

• more path updates

  mmoskal

  committed
  5ef6435

• cpp-server.sh -> server.sh

  mmoskal

  committed
  66de406

• Fix typo in REST.md

  mmoskal

  committed
  3e4fd97

• fix path

  mmoskal

  committed
  e46c772

• fix folders in release script

  mmoskal

  committed
  5ced06d

• fix #46: GPU acceleration on macOS

  mmoskal

  committed
  b30dd81

• switch to upstream rquickjs

  Show description for 73672c7
  mmoskal

  committed
  73672c7

• tone down CUDA instructions in the guide

  mmoskal

  committed
  4ab588a

• explain --cuda vs rllm-cuda

  mmoskal

  committed
  cc5c20a

• Add check for Cargo.toml existence in build_rust function

  mmoskal

  committed
  2fc7e40

• add tps stats

  mmoskal

  committed
  84a099e

• allow reordering server.sh args

  mmoskal

  committed
  3e5fce3

• upload .wasm files directly in release

  mmoskal

  committed
  7a056e3

Commits on Feb 10, 2024

• start on gh controllers; fixes #55

  mmoskal

  committed
  c8f214c

• implement module download

  mmoskal

  committed
  4326b45

• better error handling

  mmoskal

  committed
  ff1d3b5

• fix rllm-cuda phi2 crash

  mmoskal

  committed
  76ac24b

• have aici cli use gh: controllers

  mmoskal

  committed
  5e944b6