Tonic commited on
Commit
552663e
·
verified ·
1 Parent(s): 18aa477

esp32: update export_job.py

Browse files
Files changed (1) hide show
  1. export_job.py +9 -0
export_job.py CHANGED
@@ -20,7 +20,9 @@ ckpt = hf_hub_download("Tonic/challenge5-acoustic-squeezenet1_1-s7", "best.pt",
20
  print("ckpt:", os.path.getsize(ckpt) // 1024, "KB", flush=True)
21
 
22
  here = os.path.dirname(os.path.abspath(__file__))
 
23
  ck = torch.load(ckpt, map_location="cpu", weights_only=False)
 
24
  sd = ck.get("state_dict", ck) if isinstance(ck, dict) else ck
25
  arch = ck.get("arch", "squeezenet1_1") if isinstance(ck, dict) else "squeezenet1_1"
26
 
@@ -28,7 +30,9 @@ import importlib.util
28
  jt_path = hf_hub_download("Tonic/challenge5-acoustic-code", "job_train.py", repo_type="model", token=tok)
29
  spec = importlib.util.spec_from_file_location("job_train", jt_path)
30
  jt = importlib.util.module_from_spec(spec)
 
31
  spec.loader.exec_module(jt)
 
32
  backbone, feat = jt.build_backbone(arch, pretrained=False)
33
  class WakeCNN(torch.nn.Module):
34
  def __init__(self, b, f):
@@ -41,7 +45,9 @@ class WakeCNN(torch.nn.Module):
41
  return self.head_wake(f).squeeze(-1), self.head_class(f)
42
  model = WakeCNN(backbone, feat)
43
  sd = {k: v for k, v in sd.items() if k.startswith("backbone.") or k.startswith("head_")}
 
44
  model.load_state_dict(sd, strict=True)
 
45
  model.eval()
46
  print("arch:", arch, "params:", sum(p.numel() for p in model.parameters()), flush=True)
47
 
@@ -52,13 +58,16 @@ torch.onnx.export(model, torch.zeros(1, 3, SIZE, SIZE), onnx_path, opset_version
52
  dynamic_axes={"spectrogram": {0: "batch"}})
53
  print("onnx exported", flush=True)
54
 
 
55
  import onnx2tf
56
  sm_dir = "/tmp/saved_model"
57
  onnx2tf.convert(input_onnx_file_path=onnx_path, output_folder_path=sm_dir,
58
  not_use_onnxsim=True, verbosity="error")
59
  print("onnx2tf done", flush=True)
60
 
 
61
  import tensorflow as tf
 
62
  conv = tf.lite.TFLiteConverter.from_saved_model(sm_dir)
63
  conv.optimizations = [tf.lite.Optimize.DEFAULT] # dynamic-range int8 weights
64
  tfl = conv.convert()
 
20
  print("ckpt:", os.path.getsize(ckpt) // 1024, "KB", flush=True)
21
 
22
  here = os.path.dirname(os.path.abspath(__file__))
23
+ print("step: torch.load", flush=True)
24
  ck = torch.load(ckpt, map_location="cpu", weights_only=False)
25
+ print("step: ckpt loaded", flush=True)
26
  sd = ck.get("state_dict", ck) if isinstance(ck, dict) else ck
27
  arch = ck.get("arch", "squeezenet1_1") if isinstance(ck, dict) else "squeezenet1_1"
28
 
 
30
  jt_path = hf_hub_download("Tonic/challenge5-acoustic-code", "job_train.py", repo_type="model", token=tok)
31
  spec = importlib.util.spec_from_file_location("job_train", jt_path)
32
  jt = importlib.util.module_from_spec(spec)
33
+ print("step: exec job_train", flush=True)
34
  spec.loader.exec_module(jt)
35
+ print("step: job_train imported", flush=True)
36
  backbone, feat = jt.build_backbone(arch, pretrained=False)
37
  class WakeCNN(torch.nn.Module):
38
  def __init__(self, b, f):
 
45
  return self.head_wake(f).squeeze(-1), self.head_class(f)
46
  model = WakeCNN(backbone, feat)
47
  sd = {k: v for k, v in sd.items() if k.startswith("backbone.") or k.startswith("head_")}
48
+ print("step: load_state_dict", flush=True)
49
  model.load_state_dict(sd, strict=True)
50
+ print("step: state loaded", flush=True)
51
  model.eval()
52
  print("arch:", arch, "params:", sum(p.numel() for p in model.parameters()), flush=True)
53
 
 
58
  dynamic_axes={"spectrogram": {0: "batch"}})
59
  print("onnx exported", flush=True)
60
 
61
+ print("step: import onnx2tf", flush=True)
62
  import onnx2tf
63
  sm_dir = "/tmp/saved_model"
64
  onnx2tf.convert(input_onnx_file_path=onnx_path, output_folder_path=sm_dir,
65
  not_use_onnxsim=True, verbosity="error")
66
  print("onnx2tf done", flush=True)
67
 
68
+ print("step: import tensorflow", flush=True)
69
  import tensorflow as tf
70
+ print("step: tf imported", flush=True)
71
  conv = tf.lite.TFLiteConverter.from_saved_model(sm_dir)
72
  conv.optimizations = [tf.lite.Optimize.DEFAULT] # dynamic-range int8 weights
73
  tfl = conv.convert()