root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-11-14 06:49:54 +00:00
Code Issues Actions 14 Packages Projects Releases Wiki Activity

py: Add base_model_sources and dataset_sources to metadata heuristics

Browse Source 
This is to address "Model Card: Allow for dicts in datasets and base_model and also update spec" in https://github.com/huggingface/huggingface_hub/issues/2479 where we would like to add detailed metadata support for both base model and datashet but in a way that huggingface will eventually be able to support (They are currently using either a string or string list... we will be using a list of dict which would be extensible). They recommended creating a seperate metadata property for this.
This commit is contained in:
brian khuu 2024-10-07 22:54:51 +11:00
parent 640039106f
commit 9a465199a1
1 changed files with 8 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
gguf-py/gguf/metadata.py
View File

@ -348,12 +348,12 @@ class Metadata:
use_model_card_metadata("author", "model_creator")
use_model_card_metadata("basename", "model_type")
if "base_model" in model_card or "base_models" in model_card:
if "base_model" in model_card or "base_models" in model_card or "base_model_sources" in model_card:
# This represents the parent models that this is based on
# Example: stabilityai/stable-diffusion-xl-base-1.0. Can also be a list (for merges)
# Example of merges: https://huggingface.co/EmbeddedLLM/Mistral-7B-Merge-14-v0.1/blob/main/README.md
metadata_base_models = []
base_model_value = model_card.get("base_model", model_card.get("base_models", None))
base_model_value = model_card.get("base_model", model_card.get("base_models", model_card.get("base_model_sources", None)))
if base_model_value is not None:
if isinstance(base_model_value, str):
@ -402,14 +402,16 @@ class Metadata:
elif isinstance(model_id, dict):
base_model = model_id
else:
logger.error(f"base model entry '{str(model_id)}' not in a known format")
metadata.base_models.append(base_model)
if "datasets" in model_card or "dataset" in model_card:
if "datasets" in model_card or "dataset" in model_card or "dataset_sources" in model_card:
# This represents the datasets that this was trained from
metadata_datasets = []
dataset_value = model_card.get("datasets", model_card.get("dataset", None))
dataset_value = model_card.get("datasets", model_card.get("dataset", model_card.get("dataset_sources", None)))
if dataset_value is not None:
if isinstance(dataset_value, str):
@ -458,8 +460,10 @@ class Metadata:
elif isinstance(dataset_id, dict):
dataset = dataset_id
else:
logger.error(f"dataset entry '{str(dataset_id)}' not in a known format")
metadata.datasets.append(dataset)
use_model_card_metadata("license", "license")