RichardEllis/qwen3.5-4b-reduction-coldstart

Same weights, converted (fewer layers) vs Qwen/Qwen3.5-4B. RichardEllis/qwen3.5-4b-reduction-coldstart matches part of Qwen/Qwen3.5-4B (426 of its 738 tensors; the rest of Qwen/Qwen3.5-4B has no counterpart). On the shared part: Same weights stored in another dtype (BF16/F32 → BF16).

Tensors changed: Embeddings 0/1, MLP 0/96, Attention 0/224, Norms 0/105

12 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse