vllm-judge 0.1.4__py3-none-any.whl → 0.1.5__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
vllm_judge/models.py CHANGED
@@ -99,7 +99,8 @@ class Metric:
99
99
  system_prompt: Optional[str] = None,
100
100
  template_vars: Optional[Dict[str, Any]] = None,
101
101
  required_vars: Optional[List[str]] = None,
102
- template_engine: Union[str, TemplateEngine] = TemplateEngine.FORMAT
102
+ template_engine: Union[str, TemplateEngine] = TemplateEngine.FORMAT,
103
+ additional_instructions: Optional[str] = None
103
104
  ):
104
105
  """
105
106
  Initialize a reusable metric.
@@ -125,7 +126,7 @@ class Metric:
125
126
  self.template_vars = template_vars or {}
126
127
  self.required_vars = required_vars or []
127
128
  self.template_engine = TemplateEngine(template_engine)
128
-
129
+ self.additional_instructions = additional_instructions
129
130
  # Auto-detect required variables if not specified
130
131
  if not self.required_vars and self.template_engine == TemplateEngine.FORMAT:
131
132
  self._auto_detect_required_vars()
vllm_judge/prompts.py CHANGED
@@ -48,7 +48,7 @@ class PromptBuilder:
48
48
  # Output format instructions
49
49
  system_prompt+="\nYou must respond in JSON format:\n"
50
50
  system_prompt+="""{
51
- "decision": <your judgment - string|number|boolean>,
51
+ "decision": <your judgment - string|boolean>,
52
52
  "reasoning": "<concise explanation of your judgment>",
53
53
  "score": <numeric score if requested, otherwise null>
54
54
  }"""
@@ -105,16 +105,18 @@ class PromptBuilder:
105
105
  parts.append(f"\nResponse B:\n{content['b']}")
106
106
  else:
107
107
  if input:
108
- parts.append(f"Evaluate how well this response addresses the input for: {criteria}")
108
+ parts.append(f"Evaluate how well this content addresses the input for: {criteria}")
109
109
  else:
110
- parts.append(f"Evaluate the following response based on: {criteria}")
110
+ parts.append(f"Evaluate the following content based on: {criteria}")
111
111
  if context:
112
112
  parts.append(f"\nContext: {context}")
113
- parts.append(f"\nResponse to evaluate:\n{content}")
113
+ parts.append(f"\nContent to evaluate:\n{content}")
114
114
 
115
+ parts.append(f"\nYou must return a decision label/class (your judgement) for the `decision` field and a concise explanation for the `reasoning` field.")
116
+
115
117
  # Add scale and rubric
116
118
  if scale:
117
- parts.append(f"\nProvide a score from {scale[0]} to {scale[1]}")
119
+ parts.append(f"\nIn addition to these, provide a score from {scale[0]} to {scale[1]}")
118
120
 
119
121
  if isinstance(rubric, dict):
120
122
  parts.append("\nScoring guide:")
@@ -137,7 +139,7 @@ class PromptBuilder:
137
139
  if "input" in ex:
138
140
  parts.append(f"Input: {ex['input']}")
139
141
  if "content" in ex:
140
- parts.append(f"Response: {ex['content']}")
142
+ parts.append(f"Content: {ex['content']}")
141
143
  elif "text" in ex:
142
144
  parts.append(f"Text: {ex['text']}")
143
145
 
@@ -156,7 +158,7 @@ class PromptBuilder:
156
158
  # Output format instructions
157
159
  parts.append("\nYou must respond in JSON format:")
158
160
  parts.append("""{
159
- "decision": <your judgment - string|number|boolean>,
161
+ "decision": <your judgment - string|boolean>,
160
162
  "reasoning": "<concise explanation of your judgment>",
161
163
  "score": <numeric score if requested, otherwise null>
162
164
  }""")
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.1
2
2
  Name: vllm_judge
3
- Version: 0.1.4
3
+ Version: 0.1.5
4
4
  Summary: LLM-as-a-Judge evaluations for vLLM hosted models
5
5
  Author: TrustyAI team
6
6
  Author-email: Sai Chandra Pandraju <saichandrapandraju@gmail.com>
@@ -1,20 +1,20 @@
1
- vllm_judge/__init__.py,sha256=RsdlyvZ78SR3E9ytzQcdurgP-8jh_nlyw355WgUcR7M,2469
1
+ vllm_judge/__init__.py,sha256=6OKo_RbNOov83pZIPfg12ITxiE6UZh2_UOTjQsgWbFY,2792
2
2
  vllm_judge/batch.py,sha256=3zkatZxQESCjYz99qfLhxl2Dq2tHAfhtdTiXxjVqUxE,4836
3
3
  vllm_judge/cli.py,sha256=tnMqJ2RvCFaXUY4ok4IO-d9IRNJhEck60AJNzdCaqhg,13679
4
4
  vllm_judge/client.py,sha256=QPz64q9-7XEOOJiKQU7FBkGFWocJ-WGUmpETKSLQYDI,8386
5
5
  vllm_judge/exceptions.py,sha256=X9YxnukDuI3RwJPkabj3pl6v0JIbflvhUaWrdAW4RTM,1066
6
6
  vllm_judge/judge.py,sha256=SDT_cGDZzHu8NOjG6eqHQsYqIuXR12j7ocpyrVDhHrQ,16939
7
- vllm_judge/metrics.py,sha256=kH5Zb5Z6bIVa26qROe1PscBMnBX98ueKMbweLhhfM9o,25646
8
- vllm_judge/models.py,sha256=aEXZmP2sM-9aetstzHE3ngZwvCcvnrqzcj-8oV0NCJA,7889
9
- vllm_judge/prompts.py,sha256=kNswJPsJtdweV-yItggsYF0FV6FWP71fREmxZFy8sjg,7085
7
+ vllm_judge/metrics.py,sha256=WwtR6Bb4cc0gDplhZnysNzD1EfOMCEzFc8-3hJMqnJs,48709
8
+ vllm_judge/models.py,sha256=o4OdRtRdsz9n5RhHrz-uA9ylG0cGQg99NJYay0RaeDE,7998
9
+ vllm_judge/prompts.py,sha256=KC8AfiIgKKxQuhT1bnnyYXrSBbcU2-RnkSLqDJfrt8o,7251
10
10
  vllm_judge/templating.py,sha256=LjVFXFcwHl8xnBLLVr_IIqtN-EbLp0HZ5ndNbBpcJTQ,6998
11
11
  vllm_judge/utils.py,sha256=lhByBIMS_1EwvxEe31jFgVcTwcFwm5mWoJDXG4TnbvQ,509
12
12
  vllm_judge/api/__init__.py,sha256=aPQ1o7_ZzbJJpm2UyX3H35snbOGbgQJoglJjzdnc1LU,762
13
13
  vllm_judge/api/client.py,sha256=l46IpQHJxmbDfXpyCOXfir70c_3hPaIr6OEiOzOMk5Q,12449
14
14
  vllm_judge/api/models.py,sha256=GXj3slwytJWg5M4f5MPZ8Ft_hrkEEAZh0qgpYDy-Qe4,5102
15
15
  vllm_judge/api/server.py,sha256=1UQMV6MRdlqHS6NYdrQI41bi_wNb0QC8RZD4jCEeTkU,17888
16
- vllm_judge-0.1.4.dist-info/METADATA,sha256=KaiXUiIsEYbBbc4bdP1yvMwugXKPDRBoGal-Q-8ADTc,4251
17
- vllm_judge-0.1.4.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
18
- vllm_judge-0.1.4.dist-info/entry_points.txt,sha256=F3plmbMXOQ0pBIh0clqWPVIJWl20_1LZ7QHxC2XF5Lg,51
19
- vllm_judge-0.1.4.dist-info/top_level.txt,sha256=bqtMvn2y13cHSz_1-HKCBMzYSTfDHsTQBG6U5STHvwM,11
20
- vllm_judge-0.1.4.dist-info/RECORD,,
16
+ vllm_judge-0.1.5.dist-info/METADATA,sha256=5UXUqyckWp9fGLQXcBxkI6ejmFfWpCjjpyIeMx96zTI,4251
17
+ vllm_judge-0.1.5.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
18
+ vllm_judge-0.1.5.dist-info/entry_points.txt,sha256=F3plmbMXOQ0pBIh0clqWPVIJWl20_1LZ7QHxC2XF5Lg,51
19
+ vllm_judge-0.1.5.dist-info/top_level.txt,sha256=bqtMvn2y13cHSz_1-HKCBMzYSTfDHsTQBG6U5STHvwM,11
20
+ vllm_judge-0.1.5.dist-info/RECORD,,