tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 19:55:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add support for BERT embedding models (#5423)

Browse Source 
* BERT model graph construction (build_bert)
* WordPiece tokenizer (llm_tokenize_wpm)
* Add flag for non-causal attention models
* Allow for models that only output embeddings
* Support conversion of BERT models to GGUF
* Based on prior work by @xyzhang626 and @skeskinen

---------

Co-authored-by: Jared Van Bortel <jared@nomic.ai>
Co-authored-by: Jared Van Bortel <cebtenzzre@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Douglas Hanley
2024-02-11 10:21:38 -06:00
committed by GitHub
parent 97a336507e
commit 2891c8aa9a
8 changed files with 616 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
gguf-py/gguf/gguf_writer.py
View File

@ -357,6 +357,9 @@ class GGUFWriter:
def add_layer_norm_rms_eps(self, value: float) -> None:
self.add_float32(Keys.Attention.LAYERNORM_RMS_EPS.format(arch=self.arch), value)
def add_causal_attention(self, value: bool) -> None:
self.add_bool(Keys.Attention.CAUSAL.format(arch=self.arch), value)
def add_rope_dimension_count(self, count: int) -> None:
self.add_uint32(Keys.Rope.DIMENSION_COUNT.format(arch=self.arch), count)
@ -387,6 +390,9 @@ class GGUFWriter:
def add_token_types(self, types: Sequence[TokenType] | Sequence[int]) -> None:
self.add_array(Keys.Tokenizer.TOKEN_TYPE, types)
def add_token_type_count(self, value: int) -> None:
self.add_uint32(Keys.Tokenizer.TOKEN_TYPE_COUNT, value)
def add_token_scores(self, scores: Sequence[float]) -> None:
self.add_array(Keys.Tokenizer.SCORES, scores)