lllyasviel / lllyasviel/stable-diffusion-webui-forge
API - /sdapi/v1/options not being set
- Dominant language
- Python
- Stars
- 13k
- Forks
- 1.7k
- PR merge metrics
- No merged PRs in 30d
Description
I am unable to have the parameters of ```sdapi/v1/options``` persist in the webui session. I may just have a fundamental misunderstanding of `/options`.
I can't seem to get the objects to persists. txt2img API calls made after POSTing options result in images not using the values which were set for `/options`
For example, calling
`curl -X POST 192.168.3.47:7864/sdapi/v1/options \
-H "Content-Type: application/json" \
-d '{
"forge_unet_storage_dtype": "Automatic",
"forge_inference_memory": 1024,
"forge_async_loading": "Queue",
"forge_pin_shared_memory": "CPU",
"forge_preset": "flux",
"forge_additional_modules": ["ae.safetensors", "clip_l.safetensors", "t5xxl_fp16.safetensors"],
"sd_model_checkpoint": "flux/flux1-dev-fp8.safetensors",
"forge_unet_storage_dtype": "Automatic",
}'`
returns `INFO: 192.168.3.11:34048 - "GET /sdapi/v1/options HTTP/1.1" 200 OK`. Following calls to `sdapi/v1/txt2img`, and following calls to `txt2img` do not have the modules loaded, resulting in the output
`
Loading Model: {'checkpoint_info': {'filename': '/workspace/project/stable-diffusion-webui-forge/models/Stable-diffusion/flux/flux1-dev-fp8.safetensors', 'hash': '26acbda5'}, 'additional_modules': [], 'unet_storage_dtype': None}
[Unload] Trying to free all memory for cuda:0 with 0 models keep loaded ... Done.
StateDict Keys: {'transformer': 780, 'vae': 0, 'ignore': 0}
INFO: 192.168.3.11:60048 - "POST /sdapi/v1/txt2img HTTP/1.1" 500 Internal Server Error
ERROR: Exception in ASGI application
...
AssertionError: You do not have CLIP state dict!
`
edit: I am not sure whether this is an options issue or an issue loading the text encoders. I am still investigating. `override_settings` works in place of using `/options`, but this is not ideal.
Hopefully @altoiddealer or someone else can assist.
Contributor guide
No contributing guide indexed for this repository
Research direction
Start by reproducing the POST to /sdapi/v1/options followed by /sdapi/v1/txt2img, using valid JSON and checking the returned options. Trace the handlers for /sdapi/v1/options and /sdapi/v1/txt2img to determine whether settings persist or text encoders fail to load; done means the root cause is isolated and the expected behavior is verified without the CLIP assertion.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- api, backend
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 32/100