Spaces:
Sleeping
Sleeping
srijaydeshpande
commited on
Commit
•
aa98be2
1
Parent(s):
b26fc16
Update app.py
Browse files
app.py
CHANGED
@@ -82,9 +82,9 @@ def txt_to_html(text):
|
|
82 |
|
83 |
@spaces.GPU(duration=120)
|
84 |
def deidentify_doc(pdftext, maxtokens, temperature, top_probability):
|
85 |
-
prompt = "In the following text replace any person name and any address with term [redacted], replace any Date of Birth and NHS number with term [redacted]"
|
86 |
-
|
87 |
-
|
88 |
# model_id = "models/Meta-Llama-3-70B-Instruct-Q3_K_M.gguf"
|
89 |
# # model = Llama(model_path=model_id, n_ctx=2048, n_threads=8, n_gpu_layers=-1, n_batch=128)
|
90 |
# model = Llama(
|
@@ -96,20 +96,16 @@ def deidentify_doc(pdftext, maxtokens, temperature, top_probability):
|
|
96 |
# )
|
97 |
|
98 |
chat_template = MessagesFormatterType.LLAMA_3
|
99 |
-
|
100 |
-
print('Control 1-----------------------------------')
|
101 |
|
102 |
llm = Llama(
|
103 |
model_path="models/Meta-Llama-3-70B-Instruct-Q3_K_M.gguf",
|
104 |
flash_attn=True,
|
105 |
-
n_gpu_layers
|
106 |
n_batch=1024,
|
107 |
n_ctx=8192,
|
108 |
)
|
109 |
-
print('Control 2-----------------------------------')
|
110 |
|
111 |
provider = LlamaCppPythonProvider(llm)
|
112 |
-
print('Control 3-----------------------------------')
|
113 |
|
114 |
agent = LlamaCppAgent(
|
115 |
provider,
|
@@ -117,7 +113,6 @@ def deidentify_doc(pdftext, maxtokens, temperature, top_probability):
|
|
117 |
predefined_messages_formatter_type=chat_template,
|
118 |
debug_output=True
|
119 |
)
|
120 |
-
print('Control 4-----------------------------------')
|
121 |
|
122 |
settings = provider.get_provider_default_settings()
|
123 |
settings.temperature = 0.7
|
@@ -130,14 +125,13 @@ def deidentify_doc(pdftext, maxtokens, temperature, top_probability):
|
|
130 |
messages = BasicChatHistory()
|
131 |
|
132 |
stream = agent.get_chat_response(
|
133 |
-
prompt,
|
134 |
llm_sampling_settings=settings,
|
135 |
chat_history=messages,
|
136 |
returns_streaming_generator=True,
|
137 |
print_output=False
|
138 |
)
|
139 |
-
|
140 |
-
print(stream)
|
141 |
outputs = ""
|
142 |
for output in stream:
|
143 |
outputs += output
|
|
|
82 |
|
83 |
@spaces.GPU(duration=120)
|
84 |
def deidentify_doc(pdftext, maxtokens, temperature, top_probability):
|
85 |
+
# prompt = "In the following text replace any person name and any address with term [redacted], replace any Date of Birth and NHS number with term [redacted]"
|
86 |
+
prompt = "Anonymize the following report"
|
87 |
+
|
88 |
# model_id = "models/Meta-Llama-3-70B-Instruct-Q3_K_M.gguf"
|
89 |
# # model = Llama(model_path=model_id, n_ctx=2048, n_threads=8, n_gpu_layers=-1, n_batch=128)
|
90 |
# model = Llama(
|
|
|
96 |
# )
|
97 |
|
98 |
chat_template = MessagesFormatterType.LLAMA_3
|
|
|
|
|
99 |
|
100 |
llm = Llama(
|
101 |
model_path="models/Meta-Llama-3-70B-Instruct-Q3_K_M.gguf",
|
102 |
flash_attn=True,
|
103 |
+
n_gpu_layers=-1,
|
104 |
n_batch=1024,
|
105 |
n_ctx=8192,
|
106 |
)
|
|
|
107 |
|
108 |
provider = LlamaCppPythonProvider(llm)
|
|
|
109 |
|
110 |
agent = LlamaCppAgent(
|
111 |
provider,
|
|
|
113 |
predefined_messages_formatter_type=chat_template,
|
114 |
debug_output=True
|
115 |
)
|
|
|
116 |
|
117 |
settings = provider.get_provider_default_settings()
|
118 |
settings.temperature = 0.7
|
|
|
125 |
messages = BasicChatHistory()
|
126 |
|
127 |
stream = agent.get_chat_response(
|
128 |
+
prompt + ' : ' + pdftext,
|
129 |
llm_sampling_settings=settings,
|
130 |
chat_history=messages,
|
131 |
returns_streaming_generator=True,
|
132 |
print_output=False
|
133 |
)
|
134 |
+
|
|
|
135 |
outputs = ""
|
136 |
for output in stream:
|
137 |
outputs += output
|