vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

feat: Add prototype for bootstrapping registry

Browse source
This commit is contained in:
teleprint-me 2024-05-27 01:05:36 -04:00
parent 0732bd9051
commit 215394947e
No known key found for this signature in database
GPG key ID: B0D11345E65C4D48
1 changed files with 14 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

15
gguf-py/scripts/gguf-registry.py
View file

@ -2,6 +2,7 @@
from __future__ import annotations from __future__ import annotations
import argparse import argparse
import json
import logging import logging
import os import os
import sys import sys
@ -327,8 +328,8 @@ hub_tokenizer = HFHubTokenizer(
) )
metadata = []
for model in HF_MODEL_MAP: for model in HF_MODEL_MAP:
model_repo = model["model_repo"] model_repo = model["model_repo"]
model_arch = model["model_arch"] model_arch = model["model_arch"]
vocab_type = model["vocab_type"] vocab_type = model["vocab_type"]
@ -345,14 +346,26 @@ for model in HF_MODEL_MAP:
# log the downloaded results # log the downloaded results
hub_tokenizer.log_tokenizer_json_info(model_repo) hub_tokenizer.log_tokenizer_json_info(model_repo)
model['model_arch'] = MODEL_ARCH_NAMES[model_arch]
model['vocab_type'] = hub_tokenizer.get_vocab_name(vocab_type)
normalizer = hub_tokenizer.get_normalizer(model_repo) normalizer = hub_tokenizer.get_normalizer(model_repo)
# extract the normalizer metadata # extract the normalizer metadata
model['normalizer'] = normalizer
pre_tokenizer = hub_tokenizer.get_pre_tokenizer(model_repo) pre_tokenizer = hub_tokenizer.get_pre_tokenizer(model_repo)
# extract the pre-tokenizer metadata # extract the pre-tokenizer metadata
model['pre_tokenizer'] = pre_tokenizer
added_tokens = hub_tokenizer.get_added_tokens(model_repo) added_tokens = hub_tokenizer.get_added_tokens(model_repo)
# extract the added tokens metadata # extract the added tokens metadata
model['added_tokens'] = added_tokens
sha256sum = hub_tokenizer.get_tokenizer_json_hash(model_repo) sha256sum = hub_tokenizer.get_tokenizer_json_hash(model_repo)
# use the hash to validate the models vocabulary # use the hash to validate the models vocabulary
model['vocab_hash'] = sha256sum
metadata.append(model)
with open(f"{args.model_path}/registry.json", mode="w") as file:
json.dump(metadata, file, indent=2)