Liyulingyue/PaddleOCR-VL-half-GGUF

Quantization (fewer layers) vs PaddlePaddle/PaddleOCR-VL. Liyulingyue/PaddleOCR-VL-half-GGUF matches part of PaddlePaddle/PaddleOCR-VL (165 of its 620 tensors; the rest of PaddlePaddle/PaddleOCR-VL has no counterpart). On the shared part: Liyulingyue/PaddleOCR-VL-half-GGUF is a GGUF quantization (Q8_0) 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