vidore/colqwen2.5-base
Same weights, repacked vs Qwen/Qwen2.5-VL-3B-Instruct. Same weights in different files (re-saved, re-sharded or new metadata), plus 2 new tensors.
Tensors changed: Embeddings 0/2, Norms 0/138, MLP 0/304, Attention 0/380
3 derivatives checked
- vidore/colqwen2.5-v0.2: Adapter
- ApsaraStackMaaS/EvoQwen2.5-VL-Retriever-3B-v1: Partial fine-tune
- vidore/colqwen2.5-v0.1: Adapter
1 other repos hold byte-for-byte identical weight files.