Circuit-Forge/circuitforge-core
3
0
Fork 0
Code Issues 6 Pull requests Projects Releases 5 Packages Wiki Activity Actions

fix: work around transformers 5.2.0 SigLIP processor crash
Some checks are pending
CI / test (push) Waiting to run
Details
Mirror / mirror (push) Waiting to run
Details

Browse source 
SiglipProcessor.from_pretrained() hits a NoneType AttributeError in
tokenization_auto.py:652 on transformers 5.2.0 (TOKENIZER_MAPPING_NAMES
returns None for 'siglip' model type). Load SiglipTokenizer and
SiglipImageProcessor separately then compose them manually to bypass the
broken auto-detection path. Falls back to from_pretrained() on older
builds. Requires sentencepiece in the runtime env.
This commit is contained in:
pyr0ball 2026-04-09 20:01:54 -07:00
parent 8c1daf3b6c
commit 7623c3edaf
1 changed files with 16 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

19
circuitforge_core/vision/backends/siglip.py
View file

@ -49,7 +49,7 @@ class SigLIPBackend:
) -> None: ) -> None:
try: try:
import torch import torch
from transformers import AutoProcessor, AutoModel from transformers import SiglipProcessor, SiglipModel
except ImportError as exc: except ImportError as exc:
raise ImportError( raise ImportError(
"SigLIPBackend requires torch and transformers. " "SigLIPBackend requires torch and transformers. "
@ -68,8 +68,21 @@ class SigLIPBackend:
self._model_path = model_path self._model_path = model_path
self._vram_mb = _estimate_vram(model_path) self._vram_mb = _estimate_vram(model_path)
self._processor = AutoProcessor.from_pretrained(model_path) # transformers 5.2.0 broke SiglipProcessor.from_pretrained() via the
self._model = AutoModel.from_pretrained( # auto-detection path (TOKENIZER_MAPPING_NAMES.get() returns None for
# 'siglip', causing AttributeError on .replace()). Load components
# directly and compose manually to bypass that code path.
try:
from transformers import SiglipTokenizer, SiglipImageProcessor
_tokenizer = SiglipTokenizer.from_pretrained(model_path)
_image_proc = SiglipImageProcessor.from_pretrained(model_path)
self._processor = SiglipProcessor(
image_processor=_image_proc, tokenizer=_tokenizer
)
except Exception:
# Fallback: try the standard path (may work on older transformers builds)
self._processor = SiglipProcessor.from_pretrained(model_path)
self._model = SiglipModel.from_pretrained(
model_path, model_path,
torch_dtype=self._torch_dtype, torch_dtype=self._torch_dtype,
).to(device) ).to(device)