nvidia/canary-qwen-2.5b
Adapter vs Qwen/Qwen3-1.7B. nvidia/canary-qwen-2.5b is a LoRA adapter, not full weights: rank 128, on q_proj, v_proj in 32 layers (56 adapted matrices). It modifies Qwen/Qwen3-1.7B at load time.
2 derivatives checked
- handy-computer/canary-qwen-2.5b-gguf: Adapter
- cstr/canary-qwen-2.5b-GGUF: Adapter
2 other repos hold byte-for-byte identical weight files.