SulphurAI/Sulphur-2-base
LoRA-style vs Lightricks/LTX-2.3. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of 0, to_k, to_q, to_v changed, in 0 of 3 layers; norms, biases and embeddings are untouched.
Tensors changed: other 0/552, Embeddings 0/161, Attention 247/1821, Norms 0/364
12 derivatives checked
- Abiray/Sulphur-2-base-GGUF: Same weights, converted
- PeppX/Sulphur-2-Prompt-Enhancer-Uncensored-GGUF: Not comparable yet
- coolthor/Sulphur-2-distilled-Q4_K_M-GGUF: Not comparable yet
- fairy322/Sulphur-2-base-GGUF: Same weights, converted
- coolthor/Sulphur-2-distilled-NVFP4: Not comparable yet
- aman5g/Sulphur-2-base-GGUF: Same weights, converted
- Aiwithus/Sulphur-2-base-GGUF: Same weights, converted
- rockerBOO/sulphur-2-nvfp4: Quantization
- MLXBits/sulphur-2-distill-mlx-q4: Different architecture
- MLXBits/sulphur-prompt-enhancer-uncensored-mlx-q8: Quantization
- chfm/Sulphur-2-base-GGUF: Same weights, converted
- MLXBits/sulphur-promptenhancer-mlx-q8: Quantization