ToriiGate 0.5 ComfyUI NVFP4
ComfyUI-native NVFP4 safetensors conversion of Minthy/ToriiGate-0.5.
Source du modèle
- Type
- Text model
- Taille du fichier
- 3,57 GB
- Modèle de base
- Minthy/ToriiGate-0.5
- Version
- vNVFP4
- Mis à jour
- 3 nov. 2026
- Téléchargements
- 0
- Tâches
- Légendes
- Source
- Hugging Face
Description de la source
ToriiGate-0.5 ComfyUI NVFP4
ComfyUI-native NVFP4 safetensors conversion of Minthy/ToriiGate-0.5.
Files
toriigate_0.5_qwen35_4b_nvfp4_comfyui.safetensorstoriigate_0.5_qwen35_4b_nvfp4_calibrated_comfyui.safetensors
Place the file in:
ComfyUI/models/text_encoders/
Then use ComfyUI CLIPLoader + TextGenerate.
Important Fix
model.visual.pos_embed.weight is intentionally kept full precision.
It is an Embedding table, not a Linear weight. ComfyUI expects it as [2304, 1024].
Packing it as NVFP4 would turn it into [2304, 512] and break loading.
Build
Quantized layers: 347 Copied tensors: 377 Base output size: 3.570 GiB Calibrated output size: 3.570 GiB Activation input scales added: 347 GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition Torch: 2.12.0+cu130 CUDA: 13.0
Sources
1 source
Sources
1 sourceVérifié 2 oct.
Artefacts du modèle
1 artefact
Artefacts du modèle
1 artefacttoriigate_0.5_qwen35_4b_nvfp4_comfyui.safetensors
safetensors · NVFP4 (unverified) · 3,57 GB · SHA-256 0f5180616afa…9048 · Hugging Face
Download fileExtraits de sources
2 extraits
Extraits de sources
2 extraits--- license: mit base_model: Minthy/ToriiGate-0.5 tags: - comfyui - qwen3.5 - qwen35 - toriigate - nvfp4 - image-captioning - vision-language --- # ToriiGate-0.5 ComfyUI NVFP4 ComfyUI-native NVFP4 safetensors conversion of `Minthy/ToriiGate-0.5`. ## Files - `toriigate_0.5_qwen35_4b_nvfp4_comfyui.safetensors` - `toriigate_0.5_qwen35_4b_nvfp4_calibrated_comfyui.safetensors` Place the file in: `ComfyUI/models/text_encoders/` Then use ComfyUI `CLIPLoader + TextGenerate`. ## Important Fix `model.visual.pos_embed.weight` is intentionally kept full precision. It is an Embedding table, not a Linear weight. ComfyUI expects it as `[2304, 1024]`. Packing it as NVFP4 would turn it into `[2304, 512]` and break loading. ## Build Quantized layers: 347 Copied tensors: 377 Base output size: 3.570 GiB Calibrated output size: 3.570 GiB Activation input scales added: 347 GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition Torch: 2.12.0+cu130 CUDA: 13.0
Source context: 0 downloads · 0 likes · Repo Moeblack/ToriiGate-0.5-ComfyUI-NVFP4
