Update patch.diff

patch.diff

@@ -1,23 +1,21 @@
 diff --git a/src/transformers/models/llama/convert_llama_weights_to_hf.py b/src/transformers/models/llama/convert_llama_weights_to_hf.py
-index a0fbe4680a..a6de9fe3ef 100644
+index a0fbe4680a..73944d2c2c 100644
 --- a/src/transformers/models/llama/convert_llama_weights_to_hf.py
 +++ b/src/transformers/models/llama/convert_llama_weights_to_hf.py
-@@ -17,10 +17,12 @@ import json
+@@ -17,10 +17,10 @@ import json
  import os
  import shutil
  import warnings
 -
-+from turtledemo.sorting_animate import instructions1
 +from typing import List
  import torch
  
 -from transformers import LlamaConfig, LlamaForCausalLM, LlamaTokenizer, PreTrainedTokenizerFast
-+from setup import install_requires
 +from transformers import LlamaConfig, LlamaForCausalLM, LlamaTokenizer, PreTrainedTokenizerFast, GenerationConfig
  from transformers.convert_slow_tokenizer import TikTokenConverter
  
  
-@@ -85,8 +87,12 @@ NUM_SHARDS = {
+@@ -85,8 +85,12 @@ NUM_SHARDS = {
      "65B": 8,
      "70B": 8,
      "70Bf": 8,
@@ -30,7 +28,7 @@ index a0fbe4680a..a6de9fe3ef 100644
  
  def compute_intermediate_size(n, ffn_dim_multiplier=1, multiple_of=256):
      return multiple_of * ((int(ffn_dim_multiplier * int(8 * n / 3)) + multiple_of - 1) // multiple_of)
-@@ -107,9 +113,10 @@ def write_model(
+@@ -107,9 +111,10 @@ def write_model(
      input_base_path,
      model_size=None,
      safe_serialization=True,
@@ -42,7 +40,7 @@ index a0fbe4680a..a6de9fe3ef 100644
  ):
      os.makedirs(model_path, exist_ok=True)
      tmp_model_path = os.path.join(model_path, "tmp")
-@@ -125,18 +132,11 @@ def write_model(
+@@ -125,18 +130,11 @@ def write_model(
      dims_per_head = dim // n_heads
      base = params.get("rope_theta", 10000.0)
      inv_freq = 1.0 / (base ** (torch.arange(0, dims_per_head, 2).float() / dims_per_head))
@@ -64,7 +62,7 @@ index a0fbe4680a..a6de9fe3ef 100644
      if params.get("n_kv_heads", None) is not None:
          num_key_value_heads = params["n_kv_heads"]  # for GQA / MQA
          num_key_value_heads_per_shard = num_key_value_heads // num_shards
-@@ -144,8 +144,7 @@ def write_model(
+@@ -144,8 +142,7 @@ def write_model(
      else:  # compatibility with other checkpoints
          num_key_value_heads = n_heads
          num_key_value_heads_per_shard = n_heads_per_shard
@@ -74,7 +72,7 @@ index a0fbe4680a..a6de9fe3ef 100644
  
      # permute for sliced rotary
      def permute(w, n_heads, dim1=dim, dim2=dim):
-@@ -159,11 +158,9 @@ def write_model(
+@@ -159,11 +156,9 @@ def write_model(
          loaded = torch.load(os.path.join(input_base_path, "consolidated.00.pth"), map_location="cpu")
      else:
          # Sharded
@@ -89,7 +87,7 @@ index a0fbe4680a..a6de9fe3ef 100644
      param_count = 0
      index_dict = {"weight_map": {}}
      for layer_i in range(n_layers):
-@@ -263,7 +260,7 @@ def write_model(
+@@ -263,7 +258,7 @@ def write_model(
              "lm_head.weight": loaded["output.weight"],
          }
      else:
@@ -98,7 +96,7 @@ index a0fbe4680a..a6de9fe3ef 100644
          state_dict = {
              "model.norm.weight": loaded[0]["norm.weight"],
              "model.embed_tokens.weight": torch.cat(
-@@ -282,6 +279,18 @@ def write_model(
+@@ -282,6 +277,18 @@ def write_model(
      write_json(index_dict, os.path.join(tmp_model_path, "pytorch_model.bin.index.json"))
      ffn_dim_multiplier = params["ffn_dim_multiplier"] if "ffn_dim_multiplier" in params else 1
      multiple_of = params["multiple_of"] if "multiple_of" in params else 256
@@ -117,7 +115,7 @@ index a0fbe4680a..a6de9fe3ef 100644
      config = LlamaConfig(
          hidden_size=dim,
          intermediate_size=compute_intermediate_size(dim, ffn_dim_multiplier, multiple_of),
-@@ -292,11 +301,21 @@ def write_model(
+@@ -292,11 +299,21 @@ def write_model(
          vocab_size=vocab_size,
          rope_theta=base,
          max_position_embeddings=max_position_embeddings,
@@ -141,7 +139,7 @@ index a0fbe4680a..a6de9fe3ef 100644
      # Make space so we can load the model properly now.
      del state_dict
      del loaded
-@@ -313,7 +332,7 @@ def write_model(
+@@ -313,7 +330,7 @@ def write_model(
  
  
  class Llama3Converter(TikTokenConverter):
@@ -150,7 +148,7 @@ index a0fbe4680a..a6de9fe3ef 100644
          super().__init__(vocab_file, **kwargs)
          tokenizer = self.converted()
          chat_template = (
-@@ -327,34 +346,27 @@ class Llama3Converter(TikTokenConverter):
+@@ -327,34 +344,27 @@ class Llama3Converter(TikTokenConverter):
              "{% endfor %}"
              "{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}"
          )
@@ -196,7 +194,7 @@ index a0fbe4680a..a6de9fe3ef 100644
      else:
          tokenizer = tokenizer_class(input_tokenizer_path)
      print(f"Saving a {tokenizer_class.__name__} to {tokenizer_path}.")
-@@ -362,6 +374,37 @@ def write_tokenizer(tokenizer_path, input_tokenizer_path, llama_version=2):
+@@ -362,6 +372,37 @@ def write_tokenizer(tokenizer_path, input_tokenizer_path, llama_version=2):
      return tokenizer
  
  
@@ -234,7 +232,7 @@ index a0fbe4680a..a6de9fe3ef 100644
  def main():
      parser = argparse.ArgumentParser()
      parser.add_argument(
-@@ -383,9 +426,9 @@ def main():
+@@ -383,9 +424,9 @@ def main():
      # Different Llama versions used different default values for max_position_embeddings, hence the need to be able to specify which version is being used.
      parser.add_argument(
          "--llama_version",
@@ -247,7 +245,7 @@ index a0fbe4680a..a6de9fe3ef 100644
          help="Version of the Llama model to convert. Currently supports Llama1 and Llama2. Controls the context size",
      )
      parser.add_argument(
-@@ -394,11 +437,34 @@ def main():
+@@ -394,11 +435,34 @@ def main():
          type=int,
          help="The number of individual shards used for the model. Does not have to be the same as the number of consolidated_xx.pth",
      )
@@ -283,7 +281,7 @@ index a0fbe4680a..a6de9fe3ef 100644
      if args.model_size != "tokenizer_only":
          write_model(
              model_path=args.output_dir,
-@@ -408,6 +474,7 @@ def main():
+@@ -408,6 +472,7 @@ def main():
              llama_version=args.llama_version,
              vocab_size=vocab_size,
              num_shards=args.num_shards,