Qwen or not?
#1
by
clover-supply - opened
The model page says it's based on Qwen but when I try to gguf the model it says the architecture is not supported? Is it too much differrent now?
Yes, it's based on Qwen2.5. However, as described in the technical report, we apply a PMA layer on top of the model, so you will need to load with trust_remote_code=True.
How many input tokens and how many embeddings dimensions please?
C2LLM-0.5B has an embedding dimension of 896, and C2LLM-7B has an embedding dimension of 3584. Both models support 8192 input tokens.