Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -13,7 +13,7 @@ import time
|
|
13 |
|
14 |
|
15 |
|
16 |
-
|
17 |
gpt2 = AutoModelForCausalLM.from_pretrained("gpt2", return_dict_in_generate=True)
|
18 |
tokenizer = AutoTokenizer.from_pretrained("gpt2")
|
19 |
|
@@ -61,22 +61,25 @@ def inference(audio, state=""):
|
|
61 |
|
62 |
# length penalty for gpt2.generate???
|
63 |
#Prompt
|
64 |
-
generated_outputs = gpt2.generate(input_ids, do_sample=True, num_return_sequences=3, output_scores=True, max_length=4)
|
65 |
-
|
|
|
|
|
66 |
# print("outputs generated ", generated_outputs[0])
|
67 |
# only use id's that were generated
|
68 |
# gen_sequences has shape [3, 15]
|
69 |
-
|
70 |
-
|
|
|
71 |
|
72 |
# let's stack the logits generated at each step to a tensor and transform
|
73 |
# logits to probs
|
74 |
-
probs = torch.stack(generated_outputs.scores, dim=1).softmax(-1) # -> shape [3, 15, vocab_size]
|
75 |
|
76 |
# now we need to collect the probability of the generated token
|
77 |
# we need to add a dummy dim in the end to make gather work
|
78 |
-
gen_probs = torch.gather(probs, 2, gen_sequences[:, :, None]).squeeze(-1)
|
79 |
-
print("gen probs result: ", gen_probs)
|
80 |
# now we can do all kinds of things with the probs
|
81 |
|
82 |
# 1) the probs that exactly those sequences are generated again
|
@@ -99,11 +102,11 @@ def inference(audio, state=""):
|
|
99 |
print(state)
|
100 |
gt = [gt['generated_text'] for gt in state]
|
101 |
print(type(gt))
|
102 |
-
|
103 |
|
104 |
# result.text
|
105 |
#return getText, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
|
106 |
-
return result.text, state,
|
107 |
|
108 |
|
109 |
|
|
|
13 |
|
14 |
|
15 |
|
16 |
+
|
17 |
gpt2 = AutoModelForCausalLM.from_pretrained("gpt2", return_dict_in_generate=True)
|
18 |
tokenizer = AutoTokenizer.from_pretrained("gpt2")
|
19 |
|
|
|
61 |
|
62 |
# length penalty for gpt2.generate???
|
63 |
#Prompt
|
64 |
+
#generated_outputs = gpt2.generate(input_ids, do_sample=True, num_return_sequences=3, output_scores=True, max_length=4)
|
65 |
+
output = model.generate(input_ids, max_length=5, do_sample=True, top_k=50, top_p=0.95, num_return_sequences=5)
|
66 |
+
print("output ", output)
|
67 |
+
#outputs = [output[-4:] for output in output.tolist()]
|
68 |
# print("outputs generated ", generated_outputs[0])
|
69 |
# only use id's that were generated
|
70 |
# gen_sequences has shape [3, 15]
|
71 |
+
|
72 |
+
#gen_sequences = outputs.sequences[:, input_ids.shape[-1]:]
|
73 |
+
#print("gen sequences: ", gen_sequences)
|
74 |
|
75 |
# let's stack the logits generated at each step to a tensor and transform
|
76 |
# logits to probs
|
77 |
+
#probs = torch.stack(generated_outputs.scores, dim=1).softmax(-1) # -> shape [3, 15, vocab_size]
|
78 |
|
79 |
# now we need to collect the probability of the generated token
|
80 |
# we need to add a dummy dim in the end to make gather work
|
81 |
+
#gen_probs = torch.gather(probs, 2, gen_sequences[:, :, None]).squeeze(-1)
|
82 |
+
#print("gen probs result: ", gen_probs)
|
83 |
# now we can do all kinds of things with the probs
|
84 |
|
85 |
# 1) the probs that exactly those sequences are generated again
|
|
|
102 |
print(state)
|
103 |
gt = [gt['generated_text'] for gt in state]
|
104 |
print(type(gt))
|
105 |
+
gtTrim = [gt.lstrip(result) for val in gt]
|
106 |
|
107 |
# result.text
|
108 |
#return getText, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
|
109 |
+
return result.text, state, gtTrim
|
110 |
|
111 |
|
112 |
|