Spaces:
Running
Running
Commit
·
25fe98a
1
Parent(s):
a6407c2
Updated costs based on resource usage and actual api costs (created by HASHIRU AI)
Browse files
src/manager/manager.py
CHANGED
|
@@ -140,8 +140,13 @@ class GeminiManager:
|
|
| 140 |
role = "user"
|
| 141 |
if isinstance(message["content"], tuple):
|
| 142 |
path = message["content"][0]
|
| 143 |
-
|
| 144 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 145 |
continue
|
| 146 |
else:
|
| 147 |
parts = [types.Part.from_text(text=message.get("content", ""))]
|
|
@@ -197,19 +202,22 @@ class GeminiManager:
|
|
| 197 |
return results
|
| 198 |
|
| 199 |
def run(self, messages):
|
| 200 |
-
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
|
|
|
|
|
|
|
|
|
| 213 |
yield from self.invoke_manager(messages)
|
| 214 |
|
| 215 |
def invoke_manager(self, messages):
|
|
|
|
| 140 |
role = "user"
|
| 141 |
if isinstance(message["content"], tuple):
|
| 142 |
path = message["content"][0]
|
| 143 |
+
try:
|
| 144 |
+
file = self.client.files.upload(file=path)
|
| 145 |
+
formatted_history.append(file)
|
| 146 |
+
except Exception as e:
|
| 147 |
+
logger.error(f"Error uploading file: {e}")
|
| 148 |
+
formatted_history.append(
|
| 149 |
+
types.Part.from_text(text="Error uploading file: "+str(e)))
|
| 150 |
continue
|
| 151 |
else:
|
| 152 |
parts = [types.Part.from_text(text=message.get("content", ""))]
|
|
|
|
| 202 |
return results
|
| 203 |
|
| 204 |
def run(self, messages):
|
| 205 |
+
try:
|
| 206 |
+
if self.use_memory:
|
| 207 |
+
memories = self.get_k_memories(messages[-1]['content'], k=5, threshold=0.1)
|
| 208 |
+
if len(memories) > 0:
|
| 209 |
+
messages.append({
|
| 210 |
+
"role": "memories",
|
| 211 |
+
"content": f"{memories}",
|
| 212 |
+
})
|
| 213 |
+
messages.append({
|
| 214 |
+
"role": "assistant",
|
| 215 |
+
"content": f"Memories: {memories}",
|
| 216 |
+
"metadata": {"title": "Memories"}
|
| 217 |
+
})
|
| 218 |
+
yield messages
|
| 219 |
+
except Exception as e:
|
| 220 |
+
pass
|
| 221 |
yield from self.invoke_manager(messages)
|
| 222 |
|
| 223 |
def invoke_manager(self, messages):
|
src/tools/default_tools/agent_cost_manager.py
CHANGED
|
@@ -16,59 +16,59 @@ class AgentCostManager():
|
|
| 16 |
|
| 17 |
costs = {
|
| 18 |
"llama3.2": {
|
| 19 |
-
"description": "
|
| 20 |
-
"create_resource_cost":
|
| 21 |
-
"invoke_resource_cost":
|
| 22 |
},
|
| 23 |
"mistral": {
|
| 24 |
"description": "7 Billion parameter model",
|
| 25 |
-
"create_resource_cost":
|
| 26 |
-
"invoke_resource_cost":
|
| 27 |
},
|
| 28 |
"deepseek-r1": {
|
| 29 |
"description": "7 Billion reasoning model",
|
| 30 |
-
"create_resource_cost":
|
| 31 |
-
"invoke_resource_cost":
|
| 32 |
},
|
| 33 |
"gemini-2.5-flash-preview-04-17": {
|
| 34 |
"description": "Adaptive thinking, cost efficiency",
|
| 35 |
-
"create_expense_cost":
|
| 36 |
-
"invoke_expense_cost":
|
| 37 |
},
|
| 38 |
"gemini-2.5-pro-preview-03-25": {
|
| 39 |
"description": "Enhanced thinking and reasoning, multimodal understanding, advanced coding, and more",
|
| 40 |
-
"create_expense_cost":
|
| 41 |
-
"invoke_expense_cost":
|
| 42 |
},
|
| 43 |
"gemini-2.0-flash": {
|
| 44 |
"description": "Next generation features, speed, thinking, realtime streaming, and multimodal generation",
|
| 45 |
-
"create_expense_cost":
|
| 46 |
-
"invoke_expense_cost":
|
| 47 |
},
|
| 48 |
"gemini-2.0-flash-lite": {
|
| 49 |
"description": "Cost efficiency and low latency",
|
| 50 |
-
"create_expense_cost":
|
| 51 |
-
"invoke_expense_cost":
|
| 52 |
},
|
| 53 |
"gemini-1.5-flash": {
|
| 54 |
"description": "Fast and versatile performance across a diverse variety of tasks",
|
| 55 |
-
"create_expense_cost":
|
| 56 |
-
"invoke_expense_cost":
|
| 57 |
},
|
| 58 |
"gemini-1.5-flash-8b": {
|
| 59 |
"description": "High volume and lower intelligence tasks",
|
| 60 |
-
"create_expense_cost":
|
| 61 |
-
"invoke_expense_cost":
|
| 62 |
},
|
| 63 |
"gemini-1.5-pro": {
|
| 64 |
"description": "Complex reasoning tasks requiring more intelligence",
|
| 65 |
-
"create_expense_cost":
|
| 66 |
-
"invoke_expense_cost":
|
| 67 |
},
|
| 68 |
"gemini-2.0-flash-live-001": {
|
| 69 |
"description": "Low-latency bidirectional voice and video interactions",
|
| 70 |
-
"create_expense_cost":
|
| 71 |
-
"invoke_expense_cost":
|
| 72 |
}
|
| 73 |
}
|
| 74 |
|
|
|
|
| 16 |
|
| 17 |
costs = {
|
| 18 |
"llama3.2": {
|
| 19 |
+
"description": "1 Billion parameter model",
|
| 20 |
+
"create_resource_cost": 14,
|
| 21 |
+
"invoke_resource_cost": 15,
|
| 22 |
},
|
| 23 |
"mistral": {
|
| 24 |
"description": "7 Billion parameter model",
|
| 25 |
+
"create_resource_cost": 75,
|
| 26 |
+
"invoke_resource_cost": 40,
|
| 27 |
},
|
| 28 |
"deepseek-r1": {
|
| 29 |
"description": "7 Billion reasoning model",
|
| 30 |
+
"create_resource_cost": 28,
|
| 31 |
+
"invoke_resource_cost": 35,
|
| 32 |
},
|
| 33 |
"gemini-2.5-flash-preview-04-17": {
|
| 34 |
"description": "Adaptive thinking, cost efficiency",
|
| 35 |
+
"create_expense_cost": 0.005,
|
| 36 |
+
"invoke_expense_cost": 0.00017
|
| 37 |
},
|
| 38 |
"gemini-2.5-pro-preview-03-25": {
|
| 39 |
"description": "Enhanced thinking and reasoning, multimodal understanding, advanced coding, and more",
|
| 40 |
+
"create_expense_cost": 0.005,
|
| 41 |
+
"invoke_expense_cost": 0.0001275
|
| 42 |
},
|
| 43 |
"gemini-2.0-flash": {
|
| 44 |
"description": "Next generation features, speed, thinking, realtime streaming, and multimodal generation",
|
| 45 |
+
"create_expense_cost": 0.005,
|
| 46 |
+
"invoke_expense_cost": 0.00017
|
| 47 |
},
|
| 48 |
"gemini-2.0-flash-lite": {
|
| 49 |
"description": "Cost efficiency and low latency",
|
| 50 |
+
"create_expense_cost": 0.005,
|
| 51 |
+
"invoke_expense_cost": 0.00017
|
| 52 |
},
|
| 53 |
"gemini-1.5-flash": {
|
| 54 |
"description": "Fast and versatile performance across a diverse variety of tasks",
|
| 55 |
+
"create_expense_cost": 0.005,
|
| 56 |
+
"invoke_expense_cost": 0.00017
|
| 57 |
},
|
| 58 |
"gemini-1.5-flash-8b": {
|
| 59 |
"description": "High volume and lower intelligence tasks",
|
| 60 |
+
"create_expense_cost": 0.005,
|
| 61 |
+
"invoke_expense_cost": 0.00017
|
| 62 |
},
|
| 63 |
"gemini-1.5-pro": {
|
| 64 |
"description": "Complex reasoning tasks requiring more intelligence",
|
| 65 |
+
"create_expense_cost": 0.005,
|
| 66 |
+
"invoke_expense_cost": 0.0001275
|
| 67 |
},
|
| 68 |
"gemini-2.0-flash-live-001": {
|
| 69 |
"description": "Low-latency bidirectional voice and video interactions",
|
| 70 |
+
"create_expense_cost": 0.005,
|
| 71 |
+
"invoke_expense_cost": 0.000635
|
| 72 |
}
|
| 73 |
}
|
| 74 |
|
src/tools/user_tools/get_website_tool.py
CHANGED
|
@@ -1,4 +1,7 @@
|
|
| 1 |
import importlib
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
__all__ = ['GetWebsiteTool']
|
| 4 |
|
|
@@ -8,7 +11,7 @@ class GetWebsiteTool():
|
|
| 8 |
|
| 9 |
inputSchema = {
|
| 10 |
"name": "GetWebsiteTool",
|
| 11 |
-
"description": "Returns the content of a website based on a query string.",
|
| 12 |
"parameters": {
|
| 13 |
"type": "object",
|
| 14 |
"properties": {
|
|
@@ -21,6 +24,57 @@ class GetWebsiteTool():
|
|
| 21 |
}
|
| 22 |
}
|
| 23 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
def run(self, **kwargs):
|
| 25 |
headers = {
|
| 26 |
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:137.0) Gecko/20100101 Firefox/137.0',
|
|
@@ -46,7 +100,7 @@ class GetWebsiteTool():
|
|
| 46 |
"message": "Missing required parameters: 'url'",
|
| 47 |
"output": None
|
| 48 |
}
|
| 49 |
-
|
| 50 |
output = None
|
| 51 |
requests = importlib.import_module("requests")
|
| 52 |
bs4 = importlib.import_module("bs4")
|
|
@@ -57,17 +111,16 @@ class GetWebsiteTool():
|
|
| 57 |
# Parse the content using BeautifulSoup
|
| 58 |
soup = BeautifulSoup(response.content, 'html.parser')
|
| 59 |
# Extract text from the parsed HTML
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
| 61 |
else:
|
| 62 |
return {
|
| 63 |
"status": "error",
|
| 64 |
"message": f"Failed to fetch content from {url}. Status code: {response.status_code}",
|
| 65 |
"output": None
|
| 66 |
}
|
| 67 |
-
|
| 68 |
-
# truncate the results to avoid excessive output
|
| 69 |
-
if len(output) > 1000:
|
| 70 |
-
output = output[:1000] + "... (truncated)"
|
| 71 |
|
| 72 |
return {
|
| 73 |
"status": "success",
|
|
|
|
| 1 |
import importlib
|
| 2 |
+
from collections import defaultdict
|
| 3 |
+
import re
|
| 4 |
+
import time
|
| 5 |
|
| 6 |
__all__ = ['GetWebsiteTool']
|
| 7 |
|
|
|
|
| 11 |
|
| 12 |
inputSchema = {
|
| 13 |
"name": "GetWebsiteTool",
|
| 14 |
+
"description": "Returns a summary of the content of a website based on a query string.",
|
| 15 |
"parameters": {
|
| 16 |
"type": "object",
|
| 17 |
"properties": {
|
|
|
|
| 24 |
}
|
| 25 |
}
|
| 26 |
|
| 27 |
+
def summarize_text(self, text):
|
| 28 |
+
# Clean the text more thoroughly
|
| 29 |
+
text = re.sub(r'\[[0-9]*\]', ' ', text)
|
| 30 |
+
text = re.sub(r'\s+', ' ', text)
|
| 31 |
+
text = re.sub(r'[^a-zA-Z0-9.\s]', '', text) # Remove special characters except periods
|
| 32 |
+
|
| 33 |
+
# Tokenize into sentences
|
| 34 |
+
sentences = re.split(r'(?<=[.!?])\s+', text)
|
| 35 |
+
sentences = [s.strip() for s in sentences if s]
|
| 36 |
+
|
| 37 |
+
# Calculate word frequencies
|
| 38 |
+
word_frequencies = defaultdict(int)
|
| 39 |
+
for sentence in sentences:
|
| 40 |
+
words = sentence.lower().split()
|
| 41 |
+
for word in words:
|
| 42 |
+
word_frequencies[word] += 1
|
| 43 |
+
|
| 44 |
+
# Normalize word frequencies
|
| 45 |
+
total_words = sum(word_frequencies.values())
|
| 46 |
+
if total_words > 0:
|
| 47 |
+
for word in word_frequencies:
|
| 48 |
+
word_frequencies[word] /= total_words
|
| 49 |
+
|
| 50 |
+
# Calculate sentence scores based on word frequencies, sentence length, and coherence
|
| 51 |
+
sentence_scores = {}
|
| 52 |
+
for i, sentence in enumerate(sentences):
|
| 53 |
+
score = 0
|
| 54 |
+
words = sentence.lower().split()
|
| 55 |
+
for word in words:
|
| 56 |
+
score += word_frequencies[word]
|
| 57 |
+
|
| 58 |
+
# Consider sentence length
|
| 59 |
+
sentence_length_factor = 1 - abs(len(words) - 15) / 15 # Prefer sentences around 15 words
|
| 60 |
+
score += sentence_length_factor * 0.1
|
| 61 |
+
|
| 62 |
+
# Add a coherence score
|
| 63 |
+
if i > 0 and sentences[i-1] in sentence_scores:
|
| 64 |
+
previous_sentence_words = sentences[i-1].lower().split()
|
| 65 |
+
common_words = set(words) & set(previous_sentence_words)
|
| 66 |
+
coherence_score = len(common_words) / len(words)
|
| 67 |
+
score += coherence_score * 0.1
|
| 68 |
+
|
| 69 |
+
sentence_scores[sentence] = score
|
| 70 |
+
|
| 71 |
+
# Get the top 3 sentences with the highest scores
|
| 72 |
+
ranked_sentences = sorted(sentence_scores, key=sentence_scores.get, reverse=True)[:3]
|
| 73 |
+
|
| 74 |
+
# Generate the summary
|
| 75 |
+
summary = ". ".join(ranked_sentences) + "."
|
| 76 |
+
return summary
|
| 77 |
+
|
| 78 |
def run(self, **kwargs):
|
| 79 |
headers = {
|
| 80 |
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:137.0) Gecko/20100101 Firefox/137.0',
|
|
|
|
| 100 |
"message": "Missing required parameters: 'url'",
|
| 101 |
"output": None
|
| 102 |
}
|
| 103 |
+
|
| 104 |
output = None
|
| 105 |
requests = importlib.import_module("requests")
|
| 106 |
bs4 = importlib.import_module("bs4")
|
|
|
|
| 111 |
# Parse the content using BeautifulSoup
|
| 112 |
soup = BeautifulSoup(response.content, 'html.parser')
|
| 113 |
# Extract text from the parsed HTML
|
| 114 |
+
text = soup.get_text()
|
| 115 |
+
|
| 116 |
+
# Summarize the text
|
| 117 |
+
output = self.summarize_text(text)
|
| 118 |
else:
|
| 119 |
return {
|
| 120 |
"status": "error",
|
| 121 |
"message": f"Failed to fetch content from {url}. Status code: {response.status_code}",
|
| 122 |
"output": None
|
| 123 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 124 |
|
| 125 |
return {
|
| 126 |
"status": "success",
|