LunarOilRig/PaddleOCR-VL-1.6-GGUF-Q4

Quantization (fewer layers) vs PaddlePaddle/PaddleOCR-VL-1.6. LunarOilRig/PaddleOCR-VL-1.6-GGUF-Q4 matches part of PaddlePaddle/PaddleOCR-VL-1.6 (165 of its 620 tensors; the rest of PaddlePaddle/PaddleOCR-VL-1.6 has no counterpart). On the shared part: LunarOilRig/PaddleOCR-VL-1.6-GGUF-Q4 is a GGUF quantization (Q4_K_M) of the same weights: 37 tensors match exactly and 128 within quantization error.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/37, MLP 0/54, Attention 0/72

On the Hub · Downloads on Model Pulse