fix cache problem

using the huggingface plan and drop the cache

forward_check.py

@@ -0,0 +1,67 @@
+import torch
+import paddle
+import numpy as np
+from reprod_log import ReprodDiffHelper
+from reprod_log import ReprodLogger
+from paddle import nn
+from ppcls.arch.backbone.legendary_models import open_clip as clip_paddle
+import open_clip as clip_torch
+
+MODEL_param = {
+    "react": "pretrain_model_torch/react.pt",
+    "laclip": "pretrain_model_torch/laclip.pt",
+    "unicom": "pretrain_model_torch/unicom.pt",
+}
+
+
+def torch_to_numpy(name, x: torch.nn.Module):
+    try:
+        ckpt = torch.load(MODEL_param[name])["state_dict"]
+    except:
+        ckpt = torch.load(MODEL_param[name])
+    x.load_state_dict(ckpt)
+    state_dict_vision = x.visual.state_dict()
+    state_numpy = {}
+    for k in state_dict_vision.keys():
+        state_numpy[k] = state_dict_vision[k].numpy()
+    return x.visual, state_numpy
+
+
+def numpy_to_paddle(state_dict_numpy, x: paddle.nn.Layer):
+    state_paddle = {}
+    a = x.visual.state_dict()
+    for k in state_dict_numpy.keys():
+        state_paddle[k] = paddle.to_tensor(state_dict_numpy[k])
+
+    x.visual.set_state_dict(state_paddle)
+    return x.visual
+
+
+if __name__ == "__main__":
+    torch.manual_seed(0)
+    paddle.seed(0)
+    vit_b_32_paddle = clip_paddle.Laclip_vit_b_32()
+    vit_b_32_torch = clip_torch.create_model("ViT-B-32")
+    vision_torch, state_dict_paddle = torch_to_numpy("laclip", vit_b_32_torch)
+    vision_paddle = numpy_to_paddle(state_dict_paddle, vit_b_32_paddle)
+    input = torch.ones([1, 3, 224, 224]).numpy()
+    reprod_logger = ReprodLogger()
+    vision_torch.eval()
+    vision_paddle.eval()
+
+    paddle_out = vision_paddle(paddle.to_tensor(input))
+    reprod_logger.add("logits", paddle_out.numpy())
+    reprod_logger.save("./result/forward_paddle.npy")
+
+    torch_out = vision_torch(torch.Tensor(input))
+    reprod_logger.add("logits", torch_out.detach().numpy())
+    reprod_logger.save("./result/forward_torch.npy")
+
+    diff_helper = ReprodDiffHelper()
+    torch_info = diff_helper.load_info("./result/forward_torch.npy")
+    paddle_info = diff_helper.load_info("./result/forward_paddle.npy")
+
+    # compare result and produce log
+    diff_helper.compare_info(torch_info, paddle_info)
+    diff_helper.report(
+        path="./result/log/forward_diff.log", diff_threshold=1e-5)