Use the correct way of initializing using latest transformers latest branch v5.0.0
Amazing model. Thank you.
(sglangenv) GLM-4.7$ python3 -m sglang.launch_server --model-path /models/zai-org/GLM-4.7-Flash --tp-size 4 --tool-call-parser glm47 --reasoning-parser glm45 --speculative-algorithm EAGLE --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 --mem-fraction-static 0.8 --served-model-name glm-4.7-flash --host 0.0.0.0 --port 8000
Traceback (most recent call last):
File "", line 198, in _run_module_as_main
File "", line 88, in _run_code
File "/home/rmehta/sglangenv/lib/python3.12/site-packages/sglang/launch_server.py", line 7, in
from sglang.srt.server_args import prepare_server_args
File "/home/rmehta/sglangenv/lib/python3.12/site-packages/sglang/srt/server_args.py", line 67, in
from sglang.srt.utils.hf_transformers_utils import check_gguf_file
File "/home/rmehta/sglangenv/lib/python3.12/site-packages/sglang/srt/utils/hf_transformers_utils.py", line 46, in
from sglang.srt.configs import (
File "/home/rmehta/sglangenv/lib/python3.12/site-packages/sglang/srt/configs/init.py", line 9, in
from sglang.srt.configs.janus_pro import MultiModalityConfig
File "/home/rmehta/sglangenv/lib/python3.12/site-packages/sglang/srt/configs/janus_pro.py", line 634, in
register_image_processor(MultiModalityConfig, VLMImageProcessor)
File "/home/rmehta/sglangenv/lib/python3.12/site-packages/sglang/srt/configs/utils.py", line 18, in register_image_processor
AutoImageProcessor.register(config, None, image_processor, None, exist_ok=True)
TypeError: AutoImageProcessor.register() got multiple values for argument 'exist_ok'
I am using latest sglang 0.5.8 & and transformers using stable latest release 5.0.0.
This should be supported and not rely on transformers specific commit. It should be merged in the stable release in order for deployment to work.
This requires you to modify sglang and comment out the line of code register_image_processor(MultiModalityConfig, VLMImageProcessor). This is an upgrade compatibility issue and is not related to this model. We have already synchronized with the sglang committer, and it may be fixed soon.