Update tooling.py
Browse files- tooling.py +128 -125
tooling.py
CHANGED
@@ -1,125 +1,128 @@
|
|
1 |
-
from smolagents import Tool
|
2 |
-
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
|
3 |
-
import torch
|
4 |
-
from wikipedia_utils import *
|
5 |
-
from youtube_utils import *
|
6 |
-
|
7 |
-
|
8 |
-
class MathModelQuerer(Tool):
|
9 |
-
name = "math_model"
|
10 |
-
description = "Answers advanced math questions using a pretrained math model."
|
11 |
-
|
12 |
-
inputs = {
|
13 |
-
"problem": {
|
14 |
-
"type": "string",
|
15 |
-
"description": "Math problem to solve.",
|
16 |
-
}
|
17 |
-
}
|
18 |
-
|
19 |
-
output_type = "string"
|
20 |
-
|
21 |
-
def __init__(self, model_name="deepseek-ai/deepseek-math-7b-base"):
|
22 |
-
print(f"Loading math model: {model_name}")
|
23 |
-
|
24 |
-
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
|
25 |
-
print("loaded tokenizer")
|
26 |
-
self.model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16)
|
27 |
-
print("loaded auto model")
|
28 |
-
|
29 |
-
self.model.generation_config = GenerationConfig.from_pretrained(model_name)
|
30 |
-
print("loaded coonfig")
|
31 |
-
|
32 |
-
self.model.generation_config.pad_token_id = self.model.generation_config.eos_token_id
|
33 |
-
print("loaded pad token")
|
34 |
-
|
35 |
-
def forward(self, problem: str) -> str:
|
36 |
-
try:
|
37 |
-
print(f"[MathModelTool] Question: {problem}")
|
38 |
-
|
39 |
-
inputs = self.tokenizer(problem, return_tensors="pt")
|
40 |
-
outputs = self.model.generate(**inputs, max_new_tokens=100)
|
41 |
-
|
42 |
-
result = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
43 |
-
|
44 |
-
return result
|
45 |
-
except:
|
46 |
-
return f"Failed using the tool {self.name}"
|
47 |
-
|
48 |
-
|
49 |
-
class CodeModelQuerer(Tool):
|
50 |
-
name = "code_querer"
|
51 |
-
description = "Given a problem description, generates a piece of code used specialized LLM model. Returns output of the model."
|
52 |
-
|
53 |
-
inputs = {
|
54 |
-
"problem": {
|
55 |
-
"type": "string",
|
56 |
-
"description": "Description of a code sample to be generated",
|
57 |
-
}
|
58 |
-
}
|
59 |
-
|
60 |
-
output_type = "string"
|
61 |
-
|
62 |
-
def __init__(self, model_name="Qwen/Qwen2.5-Coder-32B-Instruct"):
|
63 |
-
from smolagents import HfApiModel
|
64 |
-
print(f"Loading llm for Code tool: {model_name}")
|
65 |
-
self.model = HfApiModel()
|
66 |
-
|
67 |
-
def forward(self, problem: str) -> str:
|
68 |
-
try:
|
69 |
-
return self.model.generate(problem, max_new_tokens=512)
|
70 |
-
except:
|
71 |
-
return f"Failed using the tool {self.name}"
|
72 |
-
|
73 |
-
|
74 |
-
class WikipediaPageFetcher(Tool):
|
75 |
-
name = "wiki_page_fetcher"
|
76 |
-
description = "Searches Wikipedia and provides summary about the queried topic as a string
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
return
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
"
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
|
|
|
|
|
|
|
1 |
+
from smolagents import Tool
|
2 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
|
3 |
+
import torch
|
4 |
+
from wikipedia_utils import *
|
5 |
+
from youtube_utils import *
|
6 |
+
|
7 |
+
|
8 |
+
class MathModelQuerer(Tool):
|
9 |
+
name = "math_model"
|
10 |
+
description = "Answers advanced math questions using a pretrained math model."
|
11 |
+
|
12 |
+
inputs = {
|
13 |
+
"problem": {
|
14 |
+
"type": "string",
|
15 |
+
"description": "Math problem to solve.",
|
16 |
+
}
|
17 |
+
}
|
18 |
+
|
19 |
+
output_type = "string"
|
20 |
+
|
21 |
+
def __init__(self, model_name="deepseek-ai/deepseek-math-7b-base"):
|
22 |
+
print(f"Loading math model: {model_name}")
|
23 |
+
|
24 |
+
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
|
25 |
+
print("loaded tokenizer")
|
26 |
+
self.model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16)
|
27 |
+
print("loaded auto model")
|
28 |
+
|
29 |
+
self.model.generation_config = GenerationConfig.from_pretrained(model_name)
|
30 |
+
print("loaded coonfig")
|
31 |
+
|
32 |
+
self.model.generation_config.pad_token_id = self.model.generation_config.eos_token_id
|
33 |
+
print("loaded pad token")
|
34 |
+
|
35 |
+
def forward(self, problem: str) -> str:
|
36 |
+
try:
|
37 |
+
print(f"[MathModelTool] Question: {problem}")
|
38 |
+
|
39 |
+
inputs = self.tokenizer(problem, return_tensors="pt")
|
40 |
+
outputs = self.model.generate(**inputs, max_new_tokens=100)
|
41 |
+
|
42 |
+
result = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
43 |
+
|
44 |
+
return result
|
45 |
+
except:
|
46 |
+
return f"Failed using the tool {self.name}"
|
47 |
+
|
48 |
+
|
49 |
+
class CodeModelQuerer(Tool):
|
50 |
+
name = "code_querer"
|
51 |
+
description = "Given a problem description, generates a piece of code used specialized LLM model. Returns output of the model."
|
52 |
+
|
53 |
+
inputs = {
|
54 |
+
"problem": {
|
55 |
+
"type": "string",
|
56 |
+
"description": "Description of a code sample to be generated",
|
57 |
+
}
|
58 |
+
}
|
59 |
+
|
60 |
+
output_type = "string"
|
61 |
+
|
62 |
+
def __init__(self, model_name="Qwen/Qwen2.5-Coder-32B-Instruct"):
|
63 |
+
from smolagents import HfApiModel
|
64 |
+
print(f"Loading llm for Code tool: {model_name}")
|
65 |
+
self.model = HfApiModel()
|
66 |
+
|
67 |
+
def forward(self, problem: str) -> str:
|
68 |
+
try:
|
69 |
+
return self.model.generate(problem, max_new_tokens=512)
|
70 |
+
except:
|
71 |
+
return f"Failed using the tool {self.name}"
|
72 |
+
|
73 |
+
|
74 |
+
class WikipediaPageFetcher(Tool):
|
75 |
+
name = "wiki_page_fetcher"
|
76 |
+
description = "Searches Wikipedia and provides summary about the queried topic as a string.\
|
77 |
+
Use for all wikipedia queries regardless of the language and version.\
|
78 |
+
Only provide query as an input parameter."
|
79 |
+
|
80 |
+
inputs = {
|
81 |
+
"query": {
|
82 |
+
"type": "string",
|
83 |
+
"description": "Topic of wikipedia search",
|
84 |
+
}
|
85 |
+
}
|
86 |
+
|
87 |
+
output_type = "string"
|
88 |
+
|
89 |
+
def forward(self, query: str) -> str:
|
90 |
+
try:
|
91 |
+
wiki_query = query(query)
|
92 |
+
wiki_page = fetch_wikipedia_page(wiki_query)
|
93 |
+
return wiki_page
|
94 |
+
except:
|
95 |
+
return f"Failed using the tool {self.name}"
|
96 |
+
|
97 |
+
|
98 |
+
class YoutubeTranscriptFetcher(Tool):
|
99 |
+
name = "youtube_transcript_fetcher"
|
100 |
+
description = "Attempts to fetch a youtube transcript in english, if provided with a query \\" \
|
101 |
+
" that contains a youtube link with video id. Returns a transcript content as a string. Alternatively, if tool is provided with a\\"" \
|
102 |
+
youtube video id, it can fetch the transcript directly. Video id consist of last 11 strings of the url. Only provide this parameter, if the video id doesn't have\
|
103 |
+
to be parsed from a url."
|
104 |
+
|
105 |
+
inputs = {
|
106 |
+
"query": {
|
107 |
+
"type": "string",
|
108 |
+
"description": "A query that includes youtube id."
|
109 |
+
},
|
110 |
+
"video_id" : {
|
111 |
+
"type" : "string",
|
112 |
+
"description" : "Optional string with video id from youtube.",
|
113 |
+
"nullable" : True
|
114 |
+
}
|
115 |
+
}
|
116 |
+
|
117 |
+
output_type = "string"
|
118 |
+
|
119 |
+
def forward(self, query: str, video_id=None) -> str:
|
120 |
+
try:
|
121 |
+
if video_id is None:
|
122 |
+
video_id = get_youtube_video_id(query)
|
123 |
+
|
124 |
+
fetched_transcript = fetch_transcript_english(video_id)
|
125 |
+
|
126 |
+
return post_process_transcript(fetched_transcript)
|
127 |
+
except:
|
128 |
+
return f"Failed using the tool {self.name}"
|