refactor: minor code cleanup

huggingface · Dec 30, 2024 · d3f00e0 · d3f00e0
1 parent d487b5d
commit d3f00e0
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 5 deletions.
diff --git a/src/transformers/integrations/ggml.py b/src/transformers/integrations/ggml.py
@@ -17,6 +17,7 @@
 Integration with GGML / The file is copied and adapted from https://github.com/99991/pygguf
 with extra methods beings exposed
 """
+
 from array import array
 
 import numpy as np
@@ -147,7 +148,7 @@
         ".output.": ".lm_head.",
         "output_norm": "ln_f",
     },
-        "t5": {
+    "t5": {
         "token_embd": "shared",
         "dec.blk.{bid}.attn_q": "decoder.block.{bid}.layer.0.SelfAttention.q",
         "dec.blk.{bid}.attn_k": "decoder.block.{bid}.layer.0.SelfAttention.k",
@@ -276,7 +277,7 @@
         "attn_k": "self_attn.k_proj",
         "attn_output": "self_attn.o_proj",
         "output_norm": "model.norm",
-    }
+    },
 }
 
 
@@ -374,7 +375,7 @@
         "vocab_size": "vocab_size",
         "attention.layer_norm_epsilon": "layer_norm_epsilon",
     },
-        "t5": {
+    "t5": {
         "context_length": "n_positions",
         "block_count": "num_layers",
         "feed_forward_length": "d_ff",
@@ -770,6 +771,7 @@ def converted(self) -> Tokenizer:
         tokenizer = super().converted(vocab, merges)
         return tokenizer
 
+
 class GGUFT5Converter(T5Converter):
     def __init__(self, tokenizer_dict):
         # set dummy data to avoid unnecessary merges calculation
@@ -832,6 +834,7 @@ def converted(self) -> Tokenizer:
 
         return tokenizer
 
+
 class GGUFGemmaConverter(GemmaConverter):
     def __init__(self, tokenizer_dict):
         # set dummy data to avoid unnecessary merges calculation
@@ -896,6 +899,7 @@ def converted(self) -> Tokenizer:
 
         return tokenizer
 
+
 GGUF_TO_FAST_CONVERTERS = {
     "llama": GGUFLlamaConverter,
     "qwen2": GGUFQwen2Converter,

diff --git a/src/transformers/modeling_gguf_pytorch_utils.py b/src/transformers/modeling_gguf_pytorch_utils.py
@@ -237,17 +237,19 @@ def process(self, weights, name, **kwargs):
             weights = np.log(-weights)
         return GGUFTensor(weights, name, {})
 
+
 class Gemma2TensorProcessor(TensorProcessor):
     def __init__(self, config=None):
         super().__init__(config=config)
 
-    #ref: https://github.com/ggerganov/llama.cpp/blob/d79d8f39b4da6deca4aea8bf130c6034c482b320/convert_hf_to_gguf.py#L3191
-    #ref: https://github.com/huggingface/transformers/blob/fc37f38915372c15992b540dfcbbe00a916d4fc6/src/transformers/models/gemma/modeling_gemma.py#L89
+    # ref: https://github.com/ggerganov/llama.cpp/blob/d79d8f39b4da6deca4aea8bf130c6034c482b320/convert_hf_to_gguf.py#L3191
+    # ref: https://github.com/huggingface/transformers/blob/fc37f38915372c15992b540dfcbbe00a916d4fc6/src/transformers/models/gemma/modeling_gemma.py#L89
     def process(self, weights, name, **kwargs):
         if "norm.weight" in name:
             weights = weights - 1
         return GGUFTensor(weights, name, {})
 
+
 TENSOR_PROCESSORS = {
     "llama": LlamaTensorProcessor,
     "qwen2moe": Qwen2MoeTensorProcessor,