PR #3417 MPT support in llama.cpp

CUDA: added support for ggml_clamp (see also: https://github.com/ggerganov/ggml/issues/545)

jploski committed 2 years ago

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

jploski committed 2 years ago

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

jploski committed 2 years ago

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

jploski committed 2 years ago

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 2 years ago

mpt : standardized all tensor names to follow GGUF spec

jploski committed 2 years ago

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

jploski committed 2 years ago

mpt : fixed comment s/gptneox/mpt/

jploski committed 2 years ago

mpt : remove tabs, trailing whitespace

jploski committed 2 years ago

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

jploski committed 2 years ago

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

cebtenzzre committed 2 years ago

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

cebtenzzre committed 2 years ago

comment out n_past instead of marking it unused

cebtenzzre committed 2 years ago

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

jploski committed 2 years ago

mpt : remove unused tokenizer_json in convert script

cebtenzzre committed 2 years ago

ggml : remove obsolete n_past assert in ggml_alibi

ggerganov committed 2 years ago

llama : print clam_kqv and max_alibi_bias hparams

ggerganov committed 2 years ago

llama.cpp MPT support in llama.cpp #3417 Merged

llama.cpp
MPT support in llama.cpp
#3417

Merged