Spaces:
Paused
Paused
Update utils.py
Browse files
utils.py
CHANGED
|
@@ -150,6 +150,64 @@ def greedy_search(input_ids: torch.Tensor,
|
|
| 150 |
gc.collect()
|
| 151 |
return
|
| 152 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 153 |
def convert_to_markdown(text):
|
| 154 |
text = text.replace("$","$")
|
| 155 |
def replace_leading_tabs_and_spaces(line):
|
|
|
|
| 150 |
gc.collect()
|
| 151 |
return
|
| 152 |
|
| 153 |
+
########################################
|
| 154 |
+
#Predict
|
| 155 |
+
def predict(text,
|
| 156 |
+
history,
|
| 157 |
+
top_p,
|
| 158 |
+
temperature,
|
| 159 |
+
max_length_tokens,
|
| 160 |
+
max_context_length_tokens,):
|
| 161 |
+
if text=="":
|
| 162 |
+
yield history,"Empty context."
|
| 163 |
+
return
|
| 164 |
+
try:
|
| 165 |
+
model
|
| 166 |
+
except:
|
| 167 |
+
yield [[text,"No Model Found"]],[],"No Model Found"
|
| 168 |
+
return
|
| 169 |
+
|
| 170 |
+
inputs = generate_prompt_with_history(text,history,tokenizer,max_length=max_context_length_tokens)
|
| 171 |
+
if inputs is None:
|
| 172 |
+
yield history,"Input too long."
|
| 173 |
+
return
|
| 174 |
+
else:
|
| 175 |
+
prompt,inputs=inputs
|
| 176 |
+
begin_length = len(prompt)
|
| 177 |
+
|
| 178 |
+
input_ids = inputs["input_ids"][:,-max_context_length_tokens:].to(device)
|
| 179 |
+
torch.cuda.empty_cache()
|
| 180 |
+
|
| 181 |
+
#torch.no_grad() bedeutet, dass für die betreffenden tensoren keine Ableitungen berechnet werden bei der backpropagation
|
| 182 |
+
#hier soll das NN ja auch nicht geändert werden 8backprop ist nicht nötig), da es um interference-prompts geht!
|
| 183 |
+
with torch.no_grad():
|
| 184 |
+
#die vergangenen prompts werden alle als Tupel in history abgelegt sortiert nach 'Human' und 'AI'- dass sind daher auch die stop-words, die den jeweils nächsten Eintrag kennzeichnen
|
| 185 |
+
for x in greedy_search(input_ids,model,tokenizer,stop_words=["[|Human|]", "[|AI|]"],max_length=max_length_tokens,temperature=temperature,top_p=top_p):
|
| 186 |
+
if is_stop_word_or_prefix(x,["[|Human|]", "[|AI|]"]) is False:
|
| 187 |
+
if "[|Human|]" in x:
|
| 188 |
+
x = x[:x.index("[|Human|]")].strip()
|
| 189 |
+
if "[|AI|]" in x:
|
| 190 |
+
x = x[:x.index("[|AI|]")].strip()
|
| 191 |
+
x = x.strip()
|
| 192 |
+
a, b= [[y[0],convert_to_markdown(y[1])] for y in history]+[[text, convert_to_markdown(x)]],history + [[text,x]]
|
| 193 |
+
yield a, b, "Generating..."
|
| 194 |
+
if shared_state.interrupted:
|
| 195 |
+
shared_state.recover()
|
| 196 |
+
try:
|
| 197 |
+
yield a, b, "Stop: Success"
|
| 198 |
+
return
|
| 199 |
+
except:
|
| 200 |
+
pass
|
| 201 |
+
del input_ids
|
| 202 |
+
gc.collect()
|
| 203 |
+
torch.cuda.empty_cache()
|
| 204 |
+
|
| 205 |
+
try:
|
| 206 |
+
yield a,b,"Generate: Success"
|
| 207 |
+
except:
|
| 208 |
+
pass
|
| 209 |
+
|
| 210 |
+
|
| 211 |
def convert_to_markdown(text):
|
| 212 |
text = text.replace("$","$")
|
| 213 |
def replace_leading_tabs_and_spaces(line):
|