Spaces:
Sleeping
Sleeping
s-a-malik
commited on
Commit
·
3dc5f5e
1
Parent(s):
f838d5b
todos
Browse files
app.py
CHANGED
@@ -10,6 +10,12 @@ import gradio as gr
|
|
10 |
import torch
|
11 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
MAX_MAX_NEW_TOKENS = 2048
|
15 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
@@ -40,7 +46,6 @@ if torch.cuda.is_available():
|
|
40 |
tokenizer.use_default_system_prompt = False
|
41 |
|
42 |
# load the probe data
|
43 |
-
# TODO compare accuracy and SE probe in different tabs/sections
|
44 |
with open("./model/20240625-131035_demo.pkl", "rb") as f:
|
45 |
probe_data = pkl.load(f)
|
46 |
# take the NQ open one
|
@@ -52,7 +57,6 @@ if torch.cuda.is_available():
|
|
52 |
else:
|
53 |
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
54 |
|
55 |
-
|
56 |
@spaces.GPU
|
57 |
def generate(
|
58 |
message: str,
|
@@ -62,7 +66,7 @@ def generate(
|
|
62 |
top_p: float = 0.9,
|
63 |
top_k: int = 50,
|
64 |
repetition_penalty: float = 1.2,
|
65 |
-
) ->
|
66 |
conversation = []
|
67 |
if system_prompt:
|
68 |
conversation.append({"role": "system", "content": system_prompt})
|
@@ -74,55 +78,6 @@ def generate(
|
|
74 |
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
75 |
input_ids = input_ids.to(model.device)
|
76 |
|
77 |
-
|
78 |
-
# streamer = CustomStreamer(skip_prompt=True, timeout=10.0)
|
79 |
-
|
80 |
-
# def generate_with_states():
|
81 |
-
# with torch.no_grad():
|
82 |
-
# model.generate(
|
83 |
-
# input_ids=input_ids,
|
84 |
-
# max_new_tokens=max_new_tokens,
|
85 |
-
# do_sample=True,
|
86 |
-
# top_p=top_p,
|
87 |
-
# top_k=top_k,
|
88 |
-
# temperature=temperature,
|
89 |
-
# repetition_penalty=repetition_penalty,
|
90 |
-
# output_hidden_states=True,
|
91 |
-
# return_dict_in_generate=True,
|
92 |
-
# streamer=streamer
|
93 |
-
# )
|
94 |
-
|
95 |
-
# thread = Thread(target=generate_with_states)
|
96 |
-
# thread.start()
|
97 |
-
|
98 |
-
# se_highlighted_text = ""
|
99 |
-
# acc_highlighted_text = ""
|
100 |
-
# for token_id in streamer:
|
101 |
-
# print
|
102 |
-
# hidden_states = streamer.hidden_states_queue.get()
|
103 |
-
# if hidden_states is streamer.stop_signal:
|
104 |
-
# break
|
105 |
-
|
106 |
-
# # Semantic Uncertainty Probe
|
107 |
-
# token_embeddings = torch.stack([generated_token[0, 0, :].cpu() for generated_token in hidden_states]).numpy() # (num_layers, hidden_size)
|
108 |
-
# se_concat_layers = token_embeddings[se_layer_range[0]:se_layer_range[1]].reshape(-1)
|
109 |
-
# se_probe_pred = se_probe.predict_proba(se_concat_layers.reshape(1, -1))[0][1] * 2 - 1
|
110 |
-
|
111 |
-
# # Accuracy Probe
|
112 |
-
# acc_concat_layers = token_embeddings[acc_layer_range[0]:acc_layer_range[1]].reshape(-1)
|
113 |
-
# acc_probe_pred = (1 - acc_probe.predict_proba(acc_concat_layers.reshape(1, -1))[0][1]) * 2 - 1
|
114 |
-
|
115 |
-
# # decode latest token
|
116 |
-
# new_text = tokenizer.decode(token_id)
|
117 |
-
# print(new_text, se_probe_pred, acc_probe_pred)
|
118 |
-
|
119 |
-
# se_new_highlighted_text = highlight_text(new_text, se_probe_pred)
|
120 |
-
# acc_new_highlighted_text = highlight_text(new_text, acc_probe_pred)
|
121 |
-
# se_highlighted_text += f" {se_new_highlighted_text}"
|
122 |
-
# acc_highlighted_text += f" {acc_new_highlighted_text}"
|
123 |
-
|
124 |
-
# yield se_highlighted_text, acc_highlighted_text
|
125 |
-
|
126 |
#### Generate without threading
|
127 |
generation_kwargs = dict(
|
128 |
input_ids=input_ids,
|
|
|
10 |
import torch
|
11 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
12 |
|
13 |
+
# TODO Sentence level highlighting instead (prediction after every word is not what it was trained on). Also solves token-level highlighting issues.
|
14 |
+
# TODO log prob output scaling highlighting instead?
|
15 |
+
# TODO make it look nicer
|
16 |
+
# TODO streaming output (need custom generation function because of probes)
|
17 |
+
# TODO add options to switch between models, SLT/TBG, layers?
|
18 |
+
# TODO full semantic entropy calculation
|
19 |
|
20 |
MAX_MAX_NEW_TOKENS = 2048
|
21 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
|
|
46 |
tokenizer.use_default_system_prompt = False
|
47 |
|
48 |
# load the probe data
|
|
|
49 |
with open("./model/20240625-131035_demo.pkl", "rb") as f:
|
50 |
probe_data = pkl.load(f)
|
51 |
# take the NQ open one
|
|
|
57 |
else:
|
58 |
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
59 |
|
|
|
60 |
@spaces.GPU
|
61 |
def generate(
|
62 |
message: str,
|
|
|
66 |
top_p: float = 0.9,
|
67 |
top_k: int = 50,
|
68 |
repetition_penalty: float = 1.2,
|
69 |
+
) -> Tuple[str, str]:
|
70 |
conversation = []
|
71 |
if system_prompt:
|
72 |
conversation.append({"role": "system", "content": system_prompt})
|
|
|
78 |
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
79 |
input_ids = input_ids.to(model.device)
|
80 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
81 |
#### Generate without threading
|
82 |
generation_kwargs = dict(
|
83 |
input_ids=input_ids,
|