LiquidAI/LFM2.5-VL-450M-Extract

Full fine-tune vs LiquidAI/LFM2.5-VL-450M. Fine-tune: 100% of the weights changed, across 16 of 16 layers.

Tensors changed: Embeddings 4/5, other 34/34, MLP 77/96, Norms 30/94, Attention 96/120

2 derivatives checked

• LiquidAI/LFM2.5-VL-450M-Extract-GGUF: Same weights, converted (fewer layers)
• lucas-vitrus/liquid-crow: Quantization (fewer layers)

On the Hub · Downloads on Model Pulse