Epic: Resolve Missing Converter Scripts and Preset Loader Failures for HF Models

[mrinalghoshh]

Overview

This Epic tracks the implementation of missing Hugging Face and timm converter scripts, preset loaders, and config resolutions required for Keras Hub 2026 checkpoint validation.

🐛 Missing Converters & Loaders

• Bloom: Add HF converter for 'bloom' model type #2984
• Clip: Add HF converter for 'clip' model type #2985
• D_Fine: Add HF converter for 'd_fine' model type #2986
• DeBerTa: Add HF converter for 'deberta-v2' model type #2987
• ELECTRA: Add HF converter for 'electra' model type #2989
• FNet: Add HF converter for 'fnet' model type #2988
• HGNetV2: Add HF converter for timm architecture 'hgnetv2_b6' #2990
• MedSigLIP: Add HF converter for 'siglip' model type #2991
• MiT: Add HF converter for 'mit' (Segformer) model type #2994
• moonshine: Fix converter dict assumption in preset_utils.py #2992
• OPT: Add HF converter for 'opt' model type #2995
• PaliGemma: Fix missing task preprocessor in convert_pali_gemma.py #2996
• Phi-3: Add HF converter for 'phi3' model type #2997
• SwinTransformer: Add 'swin' model type to preset_loader.py #2998
• T5: Add HF converter for 't5' model type (Backbone only) #2999
• VaultGemma: Add 'vaultgemma' model type to preset_loader.py #3000
• Whisper: Add HF converter for 'whisper' model type #3002
• Xception: Add HF converter for timm architecture 'xception41' #3001

🛠️ Weight Loading & Safetensors Errors

• Bart: Fix missing .safetensors and shared.weight keys #3003
• DeiT: Fix missing model.safetensors in preset directories #3004
• Llama2: Fix weight loading failure for int8 presets #3005
• Llama3: Fix weight loading failure for int8 presets #3006
• ViT: Fix SafetensorError (HF weight key name mismatch) #3007

🗺️ Preset Map & Config Issues

• BASNet: Add 'base' preset to preset mapping #3008
• Bert: Fix 'architectures' error for hf://google/bert_uncased #3009
• CSPNet: Add Kaggle presets to preset map #3010
• EfficientNet: Fix missing Classification Head & update preset map #3014
• Flux: Fix missing config.json for FLUX.1-schnell #3013
• Gemma 3: Fix 'head_dim' error & add 270m to preset map #3015
• PARseq: Fix missing config.json for parseq-tiny #3012
• StableDiffusion3: Fix missing config.json for stable-diffusion-3-medium #3011