Spaces:
Running
Running
Update gen_api_answer.py
Browse files- gen_api_answer.py +71 -14
gen_api_answer.py
CHANGED
@@ -10,6 +10,8 @@ from prompts import (
|
|
10 |
JUDGE_SYSTEM_PROMPT,
|
11 |
PROMETHEUS_PROMPT,
|
12 |
PROMETHEUS_PROMPT_WITH_REFERENCE,
|
|
|
|
|
13 |
)
|
14 |
|
15 |
# Initialize clients
|
@@ -18,10 +20,8 @@ openai_client = OpenAI()
|
|
18 |
together_client = Together()
|
19 |
hf_api_key = os.getenv("HF_API_KEY")
|
20 |
cohere_client = cohere.ClientV2(os.getenv("CO_API_KEY"))
|
21 |
-
|
22 |
-
|
23 |
-
api_key=hf_api_key
|
24 |
-
)
|
25 |
|
26 |
def get_openai_response(model_name, prompt, system_prompt=JUDGE_SYSTEM_PROMPT, max_tokens=500, temperature=0):
|
27 |
"""Get response from OpenAI API"""
|
@@ -70,7 +70,7 @@ def get_together_response(model_name, prompt, system_prompt=JUDGE_SYSTEM_PROMPT,
|
|
70 |
except Exception as e:
|
71 |
return f"Error with Together model {model_name}: {str(e)}"
|
72 |
|
73 |
-
def
|
74 |
"""Get response from Hugging Face model"""
|
75 |
try:
|
76 |
headers = {
|
@@ -83,7 +83,8 @@ def get_hf_response(model_name, prompt, max_tokens=500):
|
|
83 |
"inputs": prompt,
|
84 |
"parameters": {
|
85 |
"max_new_tokens": max_tokens,
|
86 |
-
"return_full_text": False
|
|
|
87 |
}
|
88 |
}
|
89 |
|
@@ -96,6 +97,34 @@ def get_hf_response(model_name, prompt, max_tokens=500):
|
|
96 |
except Exception as e:
|
97 |
return f"Error with Hugging Face model {model_name}: {str(e)}"
|
98 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
def get_cohere_response(model_name, prompt, system_prompt=JUDGE_SYSTEM_PROMPT, max_tokens=500, temperature=0):
|
100 |
"""Get response from Cohere API"""
|
101 |
try:
|
@@ -132,20 +161,23 @@ def get_model_response(
|
|
132 |
api_model = model_info["api_model"]
|
133 |
organization = model_info["organization"]
|
134 |
|
135 |
-
# Determine if model is Prometheus
|
136 |
is_prometheus = (organization == "Prometheus")
|
|
|
137 |
|
138 |
-
# For non-Prometheus models, use the Judge system prompt
|
139 |
-
system_prompt = None if is_prometheus else JUDGE_SYSTEM_PROMPT
|
140 |
|
141 |
# Select the appropriate base prompt
|
142 |
-
if
|
|
|
|
|
143 |
base_prompt = PROMETHEUS_PROMPT_WITH_REFERENCE
|
144 |
else:
|
145 |
base_prompt = PROMETHEUS_PROMPT
|
146 |
|
147 |
-
# For non-Prometheus models, replace the specific instruction
|
148 |
-
if not is_prometheus:
|
149 |
base_prompt = base_prompt.replace(
|
150 |
'3. The output format should look as follows: "Feedback: (write a feedback for criteria) [RESULT] (an integer number between 1 and 5)"',
|
151 |
'3. Your output format should strictly adhere to JSON as follows: {{"feedback": "<write feedback>", "result": <numerical score>}}. Ensure the output is valid JSON, without additional formatting or explanations.'
|
@@ -177,8 +209,12 @@ def get_model_response(
|
|
177 |
api_model, final_prompt, system_prompt, max_tokens, temperature
|
178 |
)
|
179 |
elif organization == "Prometheus":
|
180 |
-
return
|
181 |
-
api_model, final_prompt, max_tokens
|
|
|
|
|
|
|
|
|
182 |
)
|
183 |
elif organization == "Cohere":
|
184 |
return get_cohere_response(
|
@@ -269,4 +305,25 @@ def prometheus_parse_model_response(output):
|
|
269 |
|
270 |
except Exception as e:
|
271 |
print(f"Failed to parse response: {str(e)}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
272 |
return "Error", f"Exception during parsing: {str(e)}"
|
|
|
10 |
JUDGE_SYSTEM_PROMPT,
|
11 |
PROMETHEUS_PROMPT,
|
12 |
PROMETHEUS_PROMPT_WITH_REFERENCE,
|
13 |
+
ATLA_PROMPT,
|
14 |
+
ATLA_PROMPT_WITH_REFERENCE,
|
15 |
)
|
16 |
|
17 |
# Initialize clients
|
|
|
20 |
together_client = Together()
|
21 |
hf_api_key = os.getenv("HF_API_KEY")
|
22 |
cohere_client = cohere.ClientV2(os.getenv("CO_API_KEY"))
|
23 |
+
|
24 |
+
|
|
|
|
|
25 |
|
26 |
def get_openai_response(model_name, prompt, system_prompt=JUDGE_SYSTEM_PROMPT, max_tokens=500, temperature=0):
|
27 |
"""Get response from OpenAI API"""
|
|
|
70 |
except Exception as e:
|
71 |
return f"Error with Together model {model_name}: {str(e)}"
|
72 |
|
73 |
+
def get_prometheus_response(model_name, prompt, max_tokens=500, temperature=0.01): # temperature needs to be > 0 for hf to work
|
74 |
"""Get response from Hugging Face model"""
|
75 |
try:
|
76 |
headers = {
|
|
|
83 |
"inputs": prompt,
|
84 |
"parameters": {
|
85 |
"max_new_tokens": max_tokens,
|
86 |
+
"return_full_text": False,
|
87 |
+
"temperature": temperature
|
88 |
}
|
89 |
}
|
90 |
|
|
|
97 |
except Exception as e:
|
98 |
return f"Error with Hugging Face model {model_name}: {str(e)}"
|
99 |
|
100 |
+
def get_atla_response(model_name, prompt, max_tokens=500, temperature=0.01):
|
101 |
+
"""Get response from HF endpoint for Atla model"""
|
102 |
+
try:
|
103 |
+
headers = {
|
104 |
+
"Accept": "application/json",
|
105 |
+
"Authorization": f"Bearer {hf_api_key}",
|
106 |
+
"Content-Type": "application/json"
|
107 |
+
}
|
108 |
+
|
109 |
+
payload = {
|
110 |
+
"inputs": prompt,
|
111 |
+
"parameters": {
|
112 |
+
"max_new_tokens": max_tokens,
|
113 |
+
"return_full_text": False,
|
114 |
+
"temperature": temperature,
|
115 |
+
"seed": 42
|
116 |
+
}
|
117 |
+
}
|
118 |
+
|
119 |
+
response = requests.post(
|
120 |
+
"https://azk0vbxyrc64s2v2.us-east-1.aws.endpoints.huggingface.cloud",
|
121 |
+
headers=headers,
|
122 |
+
json=payload
|
123 |
+
)
|
124 |
+
return response.json()[0]["generated_text"]
|
125 |
+
except Exception as e:
|
126 |
+
return f"Error with Atla model {model_name}: {str(e)}"
|
127 |
+
|
128 |
def get_cohere_response(model_name, prompt, system_prompt=JUDGE_SYSTEM_PROMPT, max_tokens=500, temperature=0):
|
129 |
"""Get response from Cohere API"""
|
130 |
try:
|
|
|
161 |
api_model = model_info["api_model"]
|
162 |
organization = model_info["organization"]
|
163 |
|
164 |
+
# Determine if model is Prometheus or Atla
|
165 |
is_prometheus = (organization == "Prometheus")
|
166 |
+
is_atla = (organization == "Atla")
|
167 |
|
168 |
+
# For non-Prometheus/Atla models, use the Judge system prompt
|
169 |
+
system_prompt = None if (is_prometheus or is_atla) else JUDGE_SYSTEM_PROMPT
|
170 |
|
171 |
# Select the appropriate base prompt
|
172 |
+
if is_atla:
|
173 |
+
base_prompt = ATLA_PROMPT_WITH_REFERENCE if use_reference else ATLA_PROMPT
|
174 |
+
elif use_reference:
|
175 |
base_prompt = PROMETHEUS_PROMPT_WITH_REFERENCE
|
176 |
else:
|
177 |
base_prompt = PROMETHEUS_PROMPT
|
178 |
|
179 |
+
# For non-Prometheus/non-Atla models, replace the specific instruction
|
180 |
+
if not (is_prometheus or is_atla):
|
181 |
base_prompt = base_prompt.replace(
|
182 |
'3. The output format should look as follows: "Feedback: (write a feedback for criteria) [RESULT] (an integer number between 1 and 5)"',
|
183 |
'3. Your output format should strictly adhere to JSON as follows: {{"feedback": "<write feedback>", "result": <numerical score>}}. Ensure the output is valid JSON, without additional formatting or explanations.'
|
|
|
209 |
api_model, final_prompt, system_prompt, max_tokens, temperature
|
210 |
)
|
211 |
elif organization == "Prometheus":
|
212 |
+
return get_prometheus_response(
|
213 |
+
api_model, final_prompt, max_tokens, temperature = 0.01
|
214 |
+
)
|
215 |
+
elif organization == "Atla":
|
216 |
+
return get_atla_response(
|
217 |
+
api_model, final_prompt, max_tokens, temperature = 0.01
|
218 |
)
|
219 |
elif organization == "Cohere":
|
220 |
return get_cohere_response(
|
|
|
305 |
|
306 |
except Exception as e:
|
307 |
print(f"Failed to parse response: {str(e)}")
|
308 |
+
return "Error", f"Exception during parsing: {str(e)}"
|
309 |
+
|
310 |
+
def atla_parse_model_response(output):
|
311 |
+
"""Parse response from ATLA model"""
|
312 |
+
try:
|
313 |
+
print(f"Raw Atla model response: {output}")
|
314 |
+
output = output.strip()
|
315 |
+
|
316 |
+
# Look for the Reasoning and Result sections
|
317 |
+
reasoning_match = re.search(r'\*\*Reasoning:\*\*(.*?)(?=\*\*Result:|$)', output, re.DOTALL)
|
318 |
+
result_match = re.search(r'\*\*Result:\*\*\s*(\d+)', output)
|
319 |
+
|
320 |
+
if reasoning_match and result_match:
|
321 |
+
feedback = reasoning_match.group(1).strip()
|
322 |
+
score = result_match.group(1)
|
323 |
+
return str(score), feedback
|
324 |
+
|
325 |
+
return "Error", f"Failed to parse ATLA response format: {output}"
|
326 |
+
|
327 |
+
except Exception as e:
|
328 |
+
print(f"Failed to parse ATLA response: {str(e)}")
|
329 |
return "Error", f"Exception during parsing: {str(e)}"
|