Add dump tool.

chatglm/modeling_chatglm.py

@@ -502,13 +502,9 @@ class ChatGLMModel(nn.Module):
         # Rotary positional embeddings
         rotary_pos_emb = self.rotary_pos_emb(self.seq_length)
 
+        from tools import show
 
-        import plotly_express as px
-        img = px.imshow((rotary_pos_emb[:,:,0]*256).byte().cpu())
-        img.write_image("plot.png")
-
-
-
+        show.DumpTensorToImage(rotary_pos_emb[:, :, 0], "plot.png", scale=0.1)
 
         if position_ids is not None:
             rotary_pos_emb = rotary_pos_emb[position_ids]
@@ -710,7 +706,7 @@ class ChatGLMForConditionalGeneration(nn.Module):
             pad_token_id=generation_config.pad_token_id,
             eos_token_id=generation_config.eos_token_id,
             output_hidden_states=generation_config.output_hidden_states,
-            use_cache = generation_config.use_cache
+            use_cache=generation_config.use_cache,
         )
 
         outputs = outputs.tolist()[0][len(inputs["input_ids"][0]) : -1]
@@ -724,7 +720,7 @@ class ChatGLMForConditionalGeneration(nn.Module):
         pad_token_id: Optional[int] = None,
         eos_token_id: Optional[Union[int, List[int]]] = None,
         output_hidden_states: Optional[bool] = None,
-        use_cache: Optional[bool] = None
+        use_cache: Optional[bool] = None,
     ):
         if isinstance(eos_token_id, int):
             eos_token_id = [eos_token_id]

tools/show.py

@@ -0,0 +1,52 @@
+import plotly_express as px
+import torch
+import torch.nn.functional as F
+import torchvision.transforms.functional as Vision
+import cv2
+
+
+def DumpTensorToImage(tensor, name, autoPad=True, scale=1.0):
+    if len(tensor.shape) != 2:
+        raise ("Error input dims")
+    tensor = tensor.float()
+    maxv = torch.max(tensor)
+    minv = torch.min(tensor)
+    tensor = (((tensor - minv) / (maxv - minv)) * 256).byte().cpu()
+    img = tensor.numpy()
+    srp = img.shape
+    if autoPad and (max(srp) / min(srp) > 16):
+        img = cv2.resize(img,[max(srp),max(srp)])
+    srp = img.shape
+    if scale != 1.0:
+        img = cv2.resize(img, [int(srp[0] * scale), int(srp[1] * scale)])
+    srp = img.shape
+    cv2.imwrite(name, img)
+
+
+# def DumpTensorToImage(tensor, name, autoPad=True, scale=1.0):
+#     if len(tensor.shape) != 2:
+#         raise ("Error input dims")
+#     tensor = tensor.float()
+#     maxv = torch.max(tensor)
+#     minv = torch.min(tensor)
+#     tensor = (((tensor - minv) / (maxv - minv)) * 256).byte().cpu()
+#     srp = tensor.shape
+#     if autoPad and (max(srp) / min(srp) > 16):
+#         if srp[0] == min(srp):
+#             tensor = F.pad(tensor, [max(srp) - min(srp), 0], "replicate")
+#         else:
+#             tensor = F.pad(tensor, [0, max(srp) - min(srp)], "replicate")
+#     srp = tensor.shape
+
+#     tensor = tensor.unsqueeze(0)
+#     if scale != 1.0:
+#         tensor = Vision.resize(tensor, [int(srp[0] * scale), int(srp[1] * scale)])
+#     tensor = tensor.view([int(srp[0] * scale), int(srp[1] * scale)])
+#     srp = tensor.shape
+
+#     w = 1024 if max(srp) > 1024 else max(srp)
+#     scale = max(srp) / w
+#     # img = px.imshow(tensor)
+#     # img.write_image(name)
+#     cv2.imwrite(name, tensor.numpy())
+#     cv2.CreateMat(name, tensor.numpy())

tools/test.py

@@ -0,0 +1,6 @@
+import show
+import torch
+
+
+radata = torch.randn(8192, 128)
+show.DumpTensorToImage(radata, "test.png", autoPad=True,scale=0.2)