Yoxas commited on
Commit
bce5d54
·
verified ·
1 Parent(s): cfee418

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +6 -22
app.py CHANGED
@@ -19,7 +19,10 @@ data = dataset["train"]
19
  # Convert the string embeddings to numerical arrays
20
  def convert_and_ensure_2d_embeddings(example):
21
  # Convert the string to a numpy array
22
- embeddings = np.fromstring(example['embedding'].strip("[]"), sep=' ', dtype=np.float32)
 
 
 
23
  # Ensure the embeddings are 2-dimensional
24
  if embeddings.ndim == 1:
25
  embeddings = embeddings.reshape(1, -1)
@@ -34,7 +37,7 @@ model_id = "meta-llama/Meta-Llama-3-8B-Instruct"
34
 
35
  # use quantization to lower GPU usage
36
  bnb_config = BitsAndBytesConfig(
37
- load_in_4bit=True, bnb_4bit_use_double_quant=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.bfloat16
38
  )
39
 
40
  tokenizer = AutoTokenizer.from_pretrained(model_id, token=token)
@@ -119,23 +122,4 @@ A rag pipeline with a chatbot feature
119
  Resources used to build this project :
120
  * embedding model : https://huggingface.co/mixedbread-ai/mxbai-embed-large-v1
121
  * dataset : https://huggingface.co/datasets/not-lain/wikipedia
122
- * faiss docs : https://huggingface.co/docs/datasets/v2.18.0/en/package_reference/main_classes#datasets.Dataset.add_faiss_index
123
- * chatbot : https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct
124
- """
125
-
126
- demo = gr.ChatInterface(
127
- fn=talk,
128
- chatbot=gr.Chatbot(
129
- show_label=True,
130
- show_share_button=True,
131
- show_copy_button=True,
132
- likeable=True,
133
- layout="bubble",
134
- bubble_full_width=False,
135
- ),
136
- theme="Soft",
137
- examples=[["what's anarchy ? "]],
138
- title=TITLE,
139
- description=DESCRIPTION,
140
- )
141
- demo.launch(debug=True)
 
19
  # Convert the string embeddings to numerical arrays
20
  def convert_and_ensure_2d_embeddings(example):
21
  # Convert the string to a numpy array
22
+ embedding_str = example['embedding']
23
+ embedding_str = embedding_str.replace('\n', ' ')
24
+ embedding_list = list(map(float, embedding_str.strip("[]").split()))
25
+ embeddings = np.array(embedding_list, dtype=np.float32)
26
  # Ensure the embeddings are 2-dimensional
27
  if embeddings.ndim == 1:
28
  embeddings = embeddings.reshape(1, -1)
 
37
 
38
  # use quantization to lower GPU usage
39
  bnb_config = BitsAndBytesConfig(
40
+ load_in 4bit=True, bnb_4bit_use_double_quant=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.bfloat16
41
  )
42
 
43
  tokenizer = AutoTokenizer.from_pretrained(model_id, token=token)
 
122
  Resources used to build this project :
123
  * embedding model : https://huggingface.co/mixedbread-ai/mxbai-embed-large-v1
124
  * dataset : https://huggingface.co/datasets/not-lain/wikipedia
125
+ * faiss docs : https://huggingface.co/docs/datasets/v2.18.0/en/package_reference/main_classes#datasets