Spaces:
Sleeping
Sleeping
Dylan
commited on
Commit
·
598dcfa
1
Parent(s):
0160c44
added parallel map to call model multiple times
Browse files
agents.py
CHANGED
@@ -1,6 +1,9 @@
|
|
|
|
|
|
1 |
import torch
|
2 |
from langgraph.graph import END, StateGraph
|
3 |
-
from
|
|
|
4 |
|
5 |
from transformers import (
|
6 |
AutoProcessor,
|
@@ -23,23 +26,21 @@ class State(TypedDict):
|
|
23 |
image: Any
|
24 |
voice: str
|
25 |
caption: str
|
26 |
-
|
27 |
|
28 |
|
29 |
# Build the workflow graph
|
30 |
def build_graph():
|
31 |
workflow = StateGraph(State)
|
32 |
|
33 |
-
# Add nodes
|
34 |
-
# workflow.add_node("caption_image", caption_image_dummy)
|
35 |
-
# workflow.add_node("describe_with_voice", describe_with_voice_dummy)
|
36 |
-
|
37 |
workflow.add_node("caption_image", caption_image)
|
38 |
workflow.add_node("describe_with_voice", describe_with_voice)
|
39 |
|
40 |
# Add edges
|
41 |
workflow.set_entry_point("caption_image")
|
42 |
-
|
|
|
|
|
43 |
workflow.add_edge("describe_with_voice", END)
|
44 |
|
45 |
# Compile the graph
|
@@ -76,18 +77,19 @@ def describe_with_voice(state: State) -> State:
|
|
76 |
caption = state["caption"]
|
77 |
voice = state["voice"]
|
78 |
|
79 |
-
caption = "A golden retriever that seems to be smiling straight to the camera"
|
80 |
-
|
81 |
# Voice prompt templates
|
82 |
voice_prompts = {
|
83 |
"scurvy-ridden pirate": "You are a scurvy-ridden pirate, angry and drunk.",
|
84 |
"forgetful wizard": "You are a forgetful and easily distracted wizard.",
|
85 |
"sarcastic teenager": "You are a sarcastic and disinterested teenager.",
|
|
|
|
|
86 |
}
|
|
|
87 |
messages = [
|
88 |
{
|
89 |
"role": "system",
|
90 |
-
"content": [{"type": "text", "text":
|
91 |
},
|
92 |
{
|
93 |
"role": "user",
|
@@ -111,16 +113,22 @@ def describe_with_voice(state: State) -> State:
|
|
111 |
|
112 |
description = processor.decode(generation, skip_special_tokens=True)
|
113 |
|
114 |
-
|
|
|
115 |
print(description)
|
116 |
|
117 |
return state
|
118 |
|
119 |
|
|
|
|
|
|
|
|
|
|
|
120 |
def caption_image(state: State) -> State:
|
121 |
# image is PIL
|
122 |
image = state["image"]
|
123 |
-
image =
|
124 |
|
125 |
# Load models (in practice, do this once and cache)
|
126 |
messages = [
|
|
|
1 |
+
import operator
|
2 |
+
from helpers import image_to_base64
|
3 |
import torch
|
4 |
from langgraph.graph import END, StateGraph
|
5 |
+
from langgraph.types import Send
|
6 |
+
from typing import Annotated, TypedDict, Any
|
7 |
|
8 |
from transformers import (
|
9 |
AutoProcessor,
|
|
|
26 |
image: Any
|
27 |
voice: str
|
28 |
caption: str
|
29 |
+
descriptions: Annotated[list, operator.add]
|
30 |
|
31 |
|
32 |
# Build the workflow graph
|
33 |
def build_graph():
|
34 |
workflow = StateGraph(State)
|
35 |
|
|
|
|
|
|
|
|
|
36 |
workflow.add_node("caption_image", caption_image)
|
37 |
workflow.add_node("describe_with_voice", describe_with_voice)
|
38 |
|
39 |
# Add edges
|
40 |
workflow.set_entry_point("caption_image")
|
41 |
+
|
42 |
+
workflow.add_conditional_edges("caption_image", map_describe, ["describe_with_voice"])
|
43 |
+
# workflow.add_edge("caption_image", "describe_with_voice")
|
44 |
workflow.add_edge("describe_with_voice", END)
|
45 |
|
46 |
# Compile the graph
|
|
|
77 |
caption = state["caption"]
|
78 |
voice = state["voice"]
|
79 |
|
|
|
|
|
80 |
# Voice prompt templates
|
81 |
voice_prompts = {
|
82 |
"scurvy-ridden pirate": "You are a scurvy-ridden pirate, angry and drunk.",
|
83 |
"forgetful wizard": "You are a forgetful and easily distracted wizard.",
|
84 |
"sarcastic teenager": "You are a sarcastic and disinterested teenager.",
|
85 |
+
"private investigator": "You are a Victorian-age detective. Suave and intellectual.",
|
86 |
+
"shakespearian": "Talk like one of Shakespeare's characters. ",
|
87 |
}
|
88 |
+
system_prompt = voice_prompts.get(voice, "You are a pirate.") + " Output 5-10 sentences."
|
89 |
messages = [
|
90 |
{
|
91 |
"role": "system",
|
92 |
+
"content": [{"type": "text", "text": system_prompt}],
|
93 |
},
|
94 |
{
|
95 |
"role": "user",
|
|
|
113 |
|
114 |
description = processor.decode(generation, skip_special_tokens=True)
|
115 |
|
116 |
+
# note that the return value is a list
|
117 |
+
state["description"] = [description]
|
118 |
print(description)
|
119 |
|
120 |
return state
|
121 |
|
122 |
|
123 |
+
def map_describe(state: State) -> list:
|
124 |
+
# return list of `Send ` objects (3)
|
125 |
+
return [Send("describe_with_voice", {"caption" : state["caption"], "voice": state["voice"]})] * 3
|
126 |
+
|
127 |
+
|
128 |
def caption_image(state: State) -> State:
|
129 |
# image is PIL
|
130 |
image = state["image"]
|
131 |
+
image = image_to_base64(image)
|
132 |
|
133 |
# Load models (in practice, do this once and cache)
|
134 |
messages = [
|
app.py
CHANGED
@@ -15,8 +15,11 @@ def process_and_display(image, voice):
|
|
15 |
# Run the graph
|
16 |
result = graph.invoke(state)
|
17 |
|
|
|
|
|
|
|
18 |
# Return the caption and description
|
19 |
-
return result["caption"],
|
20 |
|
21 |
|
22 |
def create_interface():
|
@@ -38,6 +41,8 @@ def create_interface():
|
|
38 |
"scurvy-ridden pirate",
|
39 |
"forgetful wizard",
|
40 |
"sarcastic teenager",
|
|
|
|
|
41 |
],
|
42 |
label="Select a Voice",
|
43 |
value="scurvy-ridden pirate",
|
|
|
15 |
# Run the graph
|
16 |
result = graph.invoke(state)
|
17 |
|
18 |
+
descriptions:list[str] = result["descriptions"]
|
19 |
+
description = "\n---\n".join(descriptions)
|
20 |
+
|
21 |
# Return the caption and description
|
22 |
+
return result["caption"], description
|
23 |
|
24 |
|
25 |
def create_interface():
|
|
|
41 |
"scurvy-ridden pirate",
|
42 |
"forgetful wizard",
|
43 |
"sarcastic teenager",
|
44 |
+
"private investigator",
|
45 |
+
"shakespearian"
|
46 |
],
|
47 |
label="Select a Voice",
|
48 |
value="scurvy-ridden pirate",
|