LiquidAI/d1-3B
Partial fine-tune vs LiquidAI/LFM2.5-VL-3B. Partial change: 86% of the weights changed (attention, embeddings, mlp, norms, other), in 30 of 30 layers.
Tensors changed: Embeddings 2/5, other 66/70, MLP 90/198, Norms 76/186, Attention 32/248
7 derivatives checked
- LiquidAI/d1-3B-GGUF: Same weights, converted (fewer layers)
- AtomicChat/d1-3B-GGUF: Same weights, converted (fewer layers)
- webmp3/Sakura-d1-3B-HighQuality-GGUF: Quantization (fewer layers)
- prithivMLmods/d1-3B-GGUF: Same weights, converted (fewer layers)
- TechnoBaptist/d1-3B-GGUF: Partial fine-tune (fewer layers)
- DJLougen/d1-3B-MLX-8bit: Quantization
- LiquidAI/d1-3B-w8a8: Quantization
3 other repos hold byte-for-byte identical weight files.