minor : improve C++ and Python style (ggml-org#768)

* use some STL functions

* use self.field than setattr, use pathlib.Path

* recover some format

* const some iter

* Keep the original

* 2 space

Author: asukaminato0721

models/convert-h5-to-ggml.py

@@ -23,6 +23,7 @@
 import code
 import torch
 import numpy as np
+from pathlib import Path
 
 from transformers import WhisperForConditionalGeneration
 
@@ -75,16 +76,13 @@ def bytes_to_unicode():
     print("Usage: convert-h5-to-ggml.py dir_model path-to-whisper-repo dir-output [use-f32]\n")
     sys.exit(1)
 
-dir_model   = sys.argv[1]
-dir_whisper = sys.argv[2]
-dir_out     = sys.argv[3]
+dir_model   = Path(sys.argv[1])
+dir_whisper = Path(sys.argv[2])
+dir_out     = Path(sys.argv[3])
 
-with open(dir_model + "/vocab.json", "r", encoding="utf8") as f:
-    encoder = json.load(f)
-with open(dir_model + "/added_tokens.json", "r", encoding="utf8") as f:
-    encoder_added = json.load(f)
-with open(dir_model + "/config.json", "r", encoding="utf8") as f:
-    hparams = json.load(f)
+encoder = json.load((dir_model / "vocab.json").open("r", encoding="utf8"))
+encoder_added = json.load((dir_model / "added_tokens.json").open( "r", encoding="utf8"))
+hparams = json.load((dir_model / "config.json").open("r", encoding="utf8") )
 
 model = WhisperForConditionalGeneration.from_pretrained(dir_model)
 
@@ -96,16 +94,15 @@ def bytes_to_unicode():
 
 dir_tokenizer = dir_model
 
-fname_out = dir_out + "/ggml-model.bin"
+fname_out = dir_out / "ggml-model.bin"
 
-with open(dir_tokenizer + "/vocab.json", "r", encoding="utf8") as f:
-    tokens = json.load(f)
+tokens = json.load(open(dir_tokenizer / "vocab.json", "r", encoding="utf8"))
 
 # use 16-bit or 32-bit floats
 use_f16 = True
 if len(sys.argv) > 4:
     use_f16 = False
-    fname_out = dir_out + "/ggml-model-f32.bin"
+    fname_out = dir_out / "ggml-model-f32.bin"
 
 fout = open(fname_out, "wb")
 
@@ -171,18 +168,17 @@ def bytes_to_unicode():
     data = data.astype(np.float16)
 
     # reshape conv bias from [n] to [n, 1]
-    if name == "encoder.conv1.bias" or \
-       name == "encoder.conv2.bias":
+    if name in ["encoder.conv1.bias", "encoder.conv2.bias"]:
         data = data.reshape(data.shape[0], 1)
-        print("  Reshaped variable: " + name + " to shape: ", data.shape)
+        print("  Reshaped variable: " , name , " to shape: ", data.shape)
 
     n_dims = len(data.shape)
     print(name, n_dims, data.shape)
 
     # looks like the whisper models are in f16 by default
     # so we need to convert the small tensors to f32 until we fully support f16 in ggml
     # ftype == 0 -> float32, ftype == 1 -> float16
-    ftype = 1;
+    ftype = 1
     if use_f16:
         if n_dims < 2 or \
                 name == "encoder.conv1.bias"   or \
@@ -197,16 +193,16 @@ def bytes_to_unicode():
         ftype = 0
 
     # header
-    str = name.encode('utf-8')
-    fout.write(struct.pack("iii", n_dims, len(str), ftype))
+    str_ = name.encode('utf-8')
+    fout.write(struct.pack("iii", n_dims, len(str_), ftype))
     for i in range(n_dims):
         fout.write(struct.pack("i", data.shape[n_dims - 1 - i]))
-    fout.write(str);
+    fout.write(str_)
 
     # data
     data.tofile(fout)
 
 fout.close()
 
-print("Done. Output file: " + fname_out)
+print("Done. Output file: " , fname_out)
 print("")

models/convert-pt-to-ggml.py

@@ -40,7 +40,7 @@
 import torch
 import numpy as np
 import base64
-
+from pathlib import Path
 #from transformers import GPTJForCausalLM
 #from transformers import GPT2TokenizerFast
 
@@ -194,17 +194,17 @@ def bytes_to_unicode():
     print("Usage: convert-pt-to-ggml.py model.pt path-to-whisper-repo dir-output [use-f32]\n")
     sys.exit(1)
 
-fname_inp   = sys.argv[1]
-dir_whisper = sys.argv[2]
-dir_out     = sys.argv[3]
+fname_inp   = Path(sys.argv[1])
+dir_whisper = Path(sys.argv[2])
+dir_out     = Path(sys.argv[3])
 
 # try to load PyTorch binary data
 try:
     model_bytes = open(fname_inp, "rb").read()
     with io.BytesIO(model_bytes) as fp:
         checkpoint = torch.load(fp, map_location="cpu")
-except:
-    print("Error: failed to load PyTorch model file: %s" % fname_inp)
+except Exception:
+    print("Error: failed to load PyTorch model file:" , fname_inp)
     sys.exit(1)
 
 hparams = checkpoint["dims"]
@@ -218,17 +218,17 @@ def bytes_to_unicode():
 
 # load mel filters
 n_mels = hparams["n_mels"]
-with np.load(os.path.join(dir_whisper, "whisper/assets", "mel_filters.npz")) as f:
+with np.load(dir_whisper / "whisper" / "assets" / "mel_filters.npz") as f:
     filters = torch.from_numpy(f[f"mel_{n_mels}"])
     #print (filters)
 
 #code.interact(local=locals())
 
 multilingual = hparams["n_vocab"] == 51865
-tokenizer = os.path.join(dir_whisper, "whisper/assets", multilingual and "multilingual.tiktoken" or "gpt2.tiktoken")
+tokenizer = dir_whisper / "whisper" / "assets" / (multilingual and "multilingual.tiktoken" or "gpt2.tiktoken")
 
 # output in the same directory as the model
-fname_out = dir_out + "/ggml-model.bin"
+fname_out = dir_out / "ggml-model.bin"
 
 with open(tokenizer, "rb") as f:
     contents = f.read()
@@ -238,9 +238,9 @@ def bytes_to_unicode():
 use_f16 = True
 if len(sys.argv) > 4:
     use_f16 = False
-    fname_out = dir_out + "/ggml-model-f32.bin"
+    fname_out = dir_out / "ggml-model-f32.bin"
 
-fout = open(fname_out, "wb")
+fout = fname_out.open("wb")
 
 fout.write(struct.pack("i", 0x67676d6c)) # magic: ggml in hex
 fout.write(struct.pack("i", hparams["n_vocab"]))
@@ -273,20 +273,19 @@ def bytes_to_unicode():
 
 for name in list_vars.keys():
     data = list_vars[name].squeeze().numpy()
-    print("Processing variable: " + name + " with shape: ", data.shape)
+    print("Processing variable: " , name ,  " with shape: ", data.shape)
 
     # reshape conv bias from [n] to [n, 1]
-    if name == "encoder.conv1.bias" or \
-       name == "encoder.conv2.bias":
+    if name in ["encoder.conv1.bias", "encoder.conv2.bias"]:
         data = data.reshape(data.shape[0], 1)
-        print("  Reshaped variable: " + name + " to shape: ", data.shape)
+        print(f"  Reshaped variable: {name} to shape: ", data.shape)
 
-    n_dims = len(data.shape);
+    n_dims = len(data.shape)
 
     # looks like the whisper models are in f16 by default
     # so we need to convert the small tensors to f32 until we fully support f16 in ggml
     # ftype == 0 -> float32, ftype == 1 -> float16
-    ftype = 1;
+    ftype = 1
     if use_f16:
         if n_dims < 2 or \
                 name == "encoder.conv1.bias"   or \
@@ -307,16 +306,16 @@ def bytes_to_unicode():
     #        data = data.transpose()
 
     # header
-    str = name.encode('utf-8')
-    fout.write(struct.pack("iii", n_dims, len(str), ftype))
+    str_ = name.encode('utf-8')
+    fout.write(struct.pack("iii", n_dims, len(str_), ftype))
     for i in range(n_dims):
         fout.write(struct.pack("i", data.shape[n_dims - 1 - i]))
-    fout.write(str);
+    fout.write(str_)
 
     # data
     data.tofile(fout)
 
 fout.close()
 
-print("Done. Output file: " + fname_out)
+print("Done. Output file: " , fname_out)
 print("")

models/convert-whisper-to-coreml.py

@@ -20,7 +20,7 @@ def linear_to_conv2d_map(state_dict, prefix, local_metadata, strict,
     """
     for k in state_dict:
         is_attention = all(substr in k for substr in ['attn', '.weight'])
-        is_mlp = any([k.endswith(s) for s in ['mlp.0.weight', 'mlp.2.weight']])
+        is_mlp = any(k.endswith(s) for s in ['mlp.0.weight', 'mlp.2.weight'])
 
         if (is_attention or is_mlp) and len(state_dict[k].shape) == 2:
             state_dict[k] = state_dict[k][:, :, None, None]
@@ -42,11 +42,10 @@ def __init__(self, *args, **kwargs):
 class MultiHeadAttentionANE(MultiHeadAttention):
     def __init__(self, n_state: int, n_head: int):
         super().__init__(n_state, n_head)
-
-        setattr(self, 'query', nn.Conv2d(n_state, n_state, kernel_size=1))
-        setattr(self, 'key', nn.Conv2d(n_state, n_state, kernel_size=1, bias=False))
-        setattr(self, 'value', nn.Conv2d(n_state, n_state, kernel_size=1))
-        setattr(self, 'out', nn.Conv2d(n_state, n_state, kernel_size=1))
+        self.query =  nn.Conv2d(n_state, n_state, kernel_size=1)
+        self.key = nn.Conv2d(n_state, n_state, kernel_size=1, bias=False)
+        self.value = nn.Conv2d(n_state, n_state, kernel_size=1)
+        self.out = nn.Conv2d(n_state, n_state, kernel_size=1)
 
     def forward(self,
                 x: Tensor,
@@ -104,30 +103,28 @@ def qkv_attention_ane(self, q: Tensor, k: Tensor, v: Tensor, mask: Optional[Tens
 class ResidualAttentionBlockANE(ResidualAttentionBlock):
     def __init__(self, n_state: int, n_head: int, cross_attention: bool = False):
         super().__init__(n_state, n_head, cross_attention)
-
-        setattr(self, 'attn', MultiHeadAttentionANE(n_state, n_head))
-        setattr(self, 'attn_ln', LayerNormANE(n_state))
-
-        setattr(self, 'cross_attn', MultiHeadAttentionANE(n_state, n_head) if cross_attention else None)
-        setattr(self, 'cross_attn_ln', LayerNormANE(n_state) if cross_attention else None)
+        self.attn =  MultiHeadAttentionANE(n_state, n_head)
+        self.attn_ln = LayerNormANE(n_state)
+        self.cross_attn =  MultiHeadAttentionANE(n_state, n_head) if cross_attention else None
+        self.cross_attn_ln =  LayerNormANE(n_state) if cross_attention else None
 
         n_mlp = n_state * 4
-        setattr(self, 'mlp', nn.Sequential(
+        self.mlp =  nn.Sequential(
             nn.Conv2d(n_state, n_mlp, kernel_size=1),
             nn.GELU(),
             nn.Conv2d(n_mlp, n_state, kernel_size=1)
-        ))
-        setattr(self, 'mlp_ln', LayerNormANE(n_state))
+        )
+        self.mlp_ln = LayerNormANE(n_state)
 
 
 class AudioEncoderANE(AudioEncoder):
     def __init__(self, n_mels: int, n_ctx: int, n_state: int, n_head: int, n_layer: int):
         super().__init__(n_mels, n_ctx, n_state, n_head, n_layer)
 
-        setattr(self, 'blocks', nn.ModuleList(
+        self.blocks = nn.ModuleList(
             [ResidualAttentionBlockANE(n_state, n_head) for _ in range(n_layer)]
-        ))
-        setattr(self, 'ln_post', LayerNormANE(n_state))
+        )
+        self.ln_post = LayerNormANE(n_state)
 
     def forward(self, x: Tensor):
         """
@@ -168,10 +165,10 @@ class TextDecoderANE(TextDecoder):
     def __init__(self, n_vocab: int, n_ctx: int, n_state: int, n_head: int, n_layer: int):
         super().__init__(n_vocab, n_ctx, n_state, n_head, n_layer)
 
-        setattr(self, 'blocks', nn.ModuleList(
+        self.blocks= nn.ModuleList(
             [ResidualAttentionBlockANE(n_state, n_head, cross_attention=True) for _ in range(n_layer)]
-        ))
-        setattr(self, 'ln', LayerNormANE(n_state))
+        )
+        self.ln= LayerNormANE(n_state)
 
     def forward(self, x: Tensor, xa: Tensor, kv_cache: Optional[dict] = None):
         """
@@ -213,20 +210,20 @@ class WhisperANE(Whisper):
     def __init__(self, dims: ModelDimensions):
         super().__init__(dims)
 
-        setattr(self, 'encoder', AudioEncoderANE(
+        self.encoder = AudioEncoderANE(
             self.dims.n_mels,
             self.dims.n_audio_ctx,
             self.dims.n_audio_state,
             self.dims.n_audio_head,
             self.dims.n_audio_layer,
-        ))
-        setattr(self, 'decoder', TextDecoderANE(
+        )
+        self.decoder = TextDecoderANE(
             self.dims.n_vocab,
             self.dims.n_text_ctx,
             self.dims.n_text_state,
             self.dims.n_text_head,
             self.dims.n_text_layer,
-        ))
+        )
 
         self._register_load_state_dict_pre_hook(linear_to_conv2d_map)