Code

MPT support in llama.cpp#3417

Code

Merged

ggerganov merged 17 commits intoggerganov:masterggerganov/llama.cpp:masterfrom jploski:mptjploski/llama.cpp:mptCopy head branch name to clipboardOct 10, 2023

+685-9

Commits

Commits on Sep 30, 2023

• CUDA: added support for ggml_clamp (see also: ggerganov/ggml#545)

  jploski

  committed
  b49792b

• mpt : added an implementation based (mostly) on falcon integration, modified with deltas from ggml/examples/mpt

  jploski

  committed
  15236e8

• mpt : protect against "clip_qkv": null in mpt-7b

  jploski

  committed
  84e30e8

• mpt : quick fix to avoid "Strange model" warning when quantizing MPT models

  jploski

  committed
  00e8c5c

• mpt : addendum to changeset:84e30e8 - leave parameter clamp_kqv out from metadata rather than use 0.0 to indicate "no clamping" (more compliant with the current GGUF spec?)

  jploski

  committed
  1be89c4

• mpt : standardized all tensor names to follow GGUF spec

  jploski

  committed
  26c253e

• mpt : addendum to changeset:1be89c40 - use "req" parameter of GGUF_GET_KEY macro instead of duplicate code

  jploski

  committed
  df072d2

Commits on Oct 2, 2023

• mpt : fixed comment s/gptneox/mpt/

  jploski

  committed
  90e7d6d

• mpt : remove tabs, trailing whitespace

  jploski

  committed
  4708012

Commits on Oct 3, 2023

• mpt : removed ne01 + n_past == ne00 assertion from alibi (cuda/f32) and rope_shift from build_mpt

  jploski

  committed
  1364bcd

Commits on Oct 9, 2023

• mpt : updated convert-mpt-hf-to-gguf.py to reflect changes made to convert-gptneox-hf-to-gguf.py in pr:3252

  

  jploski

  authored and

  cebtenzzre

  committed
  7d6a24a

• Merge branch 'master' of https://github.com/ggerganov/llama.cpp into pull-3417

  cebtenzzre

  committed
  292363e

• comment out n_past instead of marking it unused

  cebtenzzre

  committed
  ad3c2f3

• mpt : removed hardcoded +178 from convert script in favor of utilizing hparams["vocab_size"]

  jploski

  committed
  1a454eb

• mpt : remove unused tokenizer_json in convert script

  cebtenzzre

  committed
  32172f1

Commits on Oct 10, 2023

• ggml : remove obsolete n_past assert in ggml_alibi

  ggerganov

  committed
  96cf3f5

• llama : print clam_kqv and max_alibi_bias hparams

  ggerganov

  committed
  9b66378