Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 09:14:35 +02:00
Code Issues Packages Projects Releases Wiki Activity

fix(anima): handle net. prefix on custom transformer files

Browse Source 
NVIDIA/Cosmos native exports prefix every key with net., including the
bundled llm_adapter half. Add it to KNOWN_PREFIXES so partition_adapter
runs on stripped keys.
This commit is contained in:
CalamitousFelicitousness
2026-04-25 11:47:45 +01:00
parent 4af3a57741
commit 0515da673d
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pipelines/anima/anima_transformer.py
+2 -1
View File
@@ -19,6 +19,7 @@ Supported input formats (safetensors only; GGUF and .pth are rejected early):
- Bare BFL keys: ``blocks.0.self_attn.q_proj.weight`` (e.g. ``rdbtAnima_v027``)
- ``model.diffusion_model.`` prefix (e.g. ``animaika_v35``)
- ``diffusion_model.`` prefix (ComfyUI-style export)
- ``net.`` prefix (NVIDIA/Cosmos native export, e.g. ``animayume_v04``)
Quantization: SDNQ (pre/post/auto) and ``layerwise_quantization`` are honored.
SDNQ pre-mode is applied post-load here because this path bypasses
@@ -35,7 +36,7 @@ from modules import shared, devices, sd_models, model_quant, errors
from modules.logger import log
KNOWN_PREFIXES = ("model.diffusion_model.", "diffusion_model.")
KNOWN_PREFIXES = ("model.diffusion_model.", "diffusion_model.", "net.")
ADAPTER_PREFIX = "llm_adapter."
COSMOS_1_MARKER = "net.blocks.block1.blocks.0.block.attn.to_q.0.weight"