openvino

ca5a5030 - [GPU] Extending the pattern for the IncreasePositionIdsPrecisionForQwen3VL transformation (#35160)

[GPU] Extending the pattern for the IncreasePositionIdsPrecisionForQwen3VL transformation (#35160) ### Details: - *The `IncreasePositionIdsPrecisionForQwen3VL` pattern was extended due to a match with Gemma4, which caused data type mismatches* - *The current Gemma4 pipeline already decompresses RoPE subgraph in FP32 (only `MatMul` is not included), apparently due to the mark decomposed RoPE subgraph as precision sensitive and propagation of the subgraph above. So perhaps this is enough for the accuracy of the model without additional processing* ### AI Assistance: - *AI assistance used: no*