OpenMOSS-Team/MOSS-VL-Instruct-0708

Full fine-tune vs OpenMOSS-Team/MOSS-VL-Base-0708. Fine-tune: 100% of the weights changed, across 48 of 48 layers.

Tensors changed: lm_head 1/1, Embeddings 4/4, Norms 251/309, MLP 252/256, Attention 324/324, other 1/1

2 derivatives checked

• OpenMOSS-Team/MOSS-VL-Instruct-0708-FP8: Quantization
• OpenMOSS-Team/MOSS-VL-Instruct-0708-NF4: Quantization

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

On the Hub · Downloads on Model Pulse