GuminiResearch/Gumini-1.5B-Base
Full fine-tune (fewer layers) vs Qwen/Qwen2.5-3B. GuminiResearch/Gumini-1.5B-Base matches part of Qwen/Qwen2.5-3B (194 of its 434 tensors; the rest of Qwen/Qwen2.5-3B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 16 of 16 layers.
Tensors changed: Embeddings 1/1, Norms 31/33, MLP 48/48, Attention 112/112
1 derivatives checked
- GuminiResearch/Gumini-1.5B-Base-i1-GGUF: Same weights, converted