Spaces:
Running
Running
Commit History
Today
-
fix root cause: IPAdapterFaceID(..., torch_dtype=float32) instead of deep_float (lib defaults to fp16 internally) Running
982ac0c -
deep_float: recursively cast all IP-Adapter modules to fp32 (fixes Half vs Float); surface traceback in UI
b050965 -
fix: cast IP-Adapter attn processors + proj layers to fp32 (h94 bin is fp16, pipe is fp32 on CPU)
963550e -
fix: FaceAnalysis root=/app (insightface expects <root>/models/antelopev2)
44bdce0 -
add g++ for insightface source build; official IPAdapterFaceID impl
4b0a5d8 -
fix: install git in slim image for git+ IP-Adapter pip install
e71de8f -
switch to official tencent-ailab IP-Adapter IPAdapterFaceID (community-proven), drop diffusers ip_adapter path
e0c41aa -
use plain ip-adapter-faceid_sd15.bin (plusv2 needs CLIP encoder -> NoneType crash)
53dacf1 -
fix: stack zero+face embeds to (2,1,512) when CFG on (chunk(2) requirement)
cad34c9 -
fix: reshape face embedding to (1,1,512) for ip_adapter_image_embeds 3D check
12a1504 -
pin gradio_client 1.3.0 / pydantic 2.10.6 / fastapi 0.115.7 to fix api_info bool schema crash
22957ca -
pre-download antelopev2 at build time; fix cv2 system libs; extra-index-url
df15d4b -
fix: load_ip_adapter subfolder='' required in diffusers 0.30
7c40a75 -
fix: install libgl1/libglib2.0-0 for cv2; extra-index-url for cpu torch
3a3dd5f -
fix: extra-index-url for cpu torch so deps resolve from PyPI
303ff90 -
CharGen CPU: SD1.5 + LCM-LoRA + FaceID-PlusV2 fixed-character generation on cpu-basic
330f5e4
-
Update run.sh
Verified signatureeb1d7cd -
Update Dockerfile
Verified signature4fe1a13 -
Update run.sh
Verified signaturec74994c -
Update Dockerfile
Verified signaturecc1b5a0
-
Add pkg-config: ggml-blas/CMakeLists.txt find_package(PkgConfig) fails configure without it
openclawd14f219 -
Fix hostname typo in README (jacky3105 -> jacky3102)
openclaw19a5116 -
Replace Video Captioner with ACE-Step music generation (acestep.cpp, 0.6B CPU set)
openclaw- Dockerfile: cmake build from source (buildcpu.sh path, GGML_BLAS=ON), bake 4 weights ~4.2GB - run.sh: boot diagnostics + ace-server on 7860 + --keep-loaded, hold-alive after crash - README: async /lm -> /job -> /synth API usage + CPU limits - old Video Captioner (MiniCPM-V) fully removed; backup kept locally5f21217