Update convert.py to check for dependencies on startup

Author: thomasantony

llamacpp/convert.py

@@ -17,13 +17,21 @@
 # and vocabulary.
 #
 
+# Check if torch is installed and show and error and exit if not
 import sys
 import json
 import struct
-import numpy as np
-import torch
 
-from sentencepiece import SentencePieceProcessor
+try:
+    import torch
+    import numpy as np
+    from sentencepiece import SentencePieceProcessor
+except ImportError:
+    print("Error: torch, sentencepiece and numpy are required to run this script.")
+    print("Please install using the following command:")
+    print("  pip install torch sentencepiece numpy")
+    sys.exit(1)
+
 
 def main():
     if len(sys.argv) < 3:
@@ -35,7 +43,7 @@ def main():
     # output in the same directory as the model
     dir_model = sys.argv[1]
 
-    fname_hparams   = sys.argv[1] + "/params.json"
+    fname_hparams = sys.argv[1] + "/params.json"
     fname_tokenizer = sys.argv[1] + "/../tokenizer.model"
 
     def get_n_parts(dim):
@@ -76,35 +84,35 @@ def get_n_parts(dim):
     n_parts = get_n_parts(hparams["dim"])
 
     print(hparams)
-    print('n_parts = ', n_parts)
+    print("n_parts = ", n_parts)
 
     for p in range(n_parts):
-        print('Processing part ', p)
+        print("Processing part ", p)
 
-        #fname_model = sys.argv[1] + "/consolidated.00.pth"
+        # fname_model = sys.argv[1] + "/consolidated.00.pth"
         fname_model = sys.argv[1] + "/consolidated.0" + str(p) + ".pth"
         fname_out = sys.argv[1] + "/ggml-model-" + ftype_str[ftype] + ".bin"
-        if (p > 0):
+        if p > 0:
             fname_out = sys.argv[1] + "/ggml-model-" + ftype_str[ftype] + ".bin" + "." + str(p)
 
         # weights_only requires torch 1.13.1, remove this param or update if you get an "invalid keyword argument" error
         model = torch.load(fname_model, map_location="cpu", weights_only=True)
 
         fout = open(fname_out, "wb")
 
-        fout.write(struct.pack("i", 0x67676d6c)) # magic: ggml in hex
+        fout.write(struct.pack("i", 0x67676D6C))  # magic: ggml in hex
         fout.write(struct.pack("i", hparams["vocab_size"]))
         fout.write(struct.pack("i", hparams["dim"]))
         fout.write(struct.pack("i", hparams["multiple_of"]))
         fout.write(struct.pack("i", hparams["n_heads"]))
         fout.write(struct.pack("i", hparams["n_layers"]))
-        fout.write(struct.pack("i", hparams["dim"] // hparams["n_heads"])) # rot (obsolete)
+        fout.write(struct.pack("i", hparams["dim"] // hparams["n_heads"]))  # rot (obsolete)
         fout.write(struct.pack("i", ftype))
 
         # Is this correct??
         for i in range(32000):
             # TODO: this is probably wrong - not sure how this tokenizer works
-            text = tokenizer.decode([29889, i]).encode('utf-8')
+            text = tokenizer.decode([29889, i]).encode("utf-8")
             # remove the first byte (it's always '.')
             text = text[1:]
             fout.write(struct.pack("i", len(text)))
@@ -120,16 +128,16 @@ def get_n_parts(dim):
 
             print("Processing variable: " + name + " with shape: ", shape, " and type: ", v.dtype)
 
-            #data = tf.train.load_variable(dir_model, name).squeeze()
+            # data = tf.train.load_variable(dir_model, name).squeeze()
             data = v.numpy().squeeze()
-            n_dims = len(data.shape);
+            n_dims = len(data.shape)
 
             # for efficiency - transpose some matrices
             # "model/h.*/attn/c_attn/w"
             # "model/h.*/attn/c_proj/w"
             # "model/h.*/mlp/c_fc/w"
             # "model/h.*/mlp/c_proj/w"
-            #if name[-14:] == "/attn/c_attn/w" or \
+            # if name[-14:] == "/attn/c_attn/w" or \
             #   name[-14:] == "/attn/c_proj/w" or \
             #   name[-11:] == "/mlp/c_fc/w" or \
             #   name[-13:] == "/mlp/c_proj/w":
@@ -146,11 +154,11 @@ def get_n_parts(dim):
                 ftype_cur = 0
 
             # header
-            sname = name.encode('utf-8')
+            sname = name.encode("utf-8")
             fout.write(struct.pack("iii", n_dims, len(sname), ftype_cur))
             for i in range(n_dims):
                 fout.write(struct.pack("i", dshape[n_dims - 1 - i]))
-            fout.write(sname);
+            fout.write(sname)
 
             # data
             data.tofile(fout)
@@ -163,5 +171,6 @@ def get_n_parts(dim):
         print("Done. Output file: " + fname_out + ", (part ", p, ")")
         print("")
 
-if __name__ == '__main__':
+
+if __name__ == "__main__":
     main()