Update llmeval.py
Browse files- llmeval.py +13 -27
llmeval.py
CHANGED
@@ -42,40 +42,36 @@ class LLM_as_Evaluator():
|
|
42 |
|
43 |
SYSTEM='''
|
44 |
Task:
|
45 |
-
Evaluate the biological quality of a prompt
|
46 |
|
47 |
Goal:
|
48 |
Assess:
|
49 |
|
50 |
-
Whether the Prompt is clear, biologically specific, and aligned with the Research Data.
|
51 |
|
52 |
-
Whether the
|
|
|
|
|
53 |
|
54 |
Scoring Guide (0–1 continuous scale):
|
55 |
|
56 |
Score 1.0 if:
|
57 |
|
58 |
-
Prompt is clear, biologically detailed, and
|
59 |
|
60 |
-
|
61 |
|
62 |
Lower scores if:
|
63 |
|
64 |
-
The prompt is vague or misaligned.
|
|
|
|
|
65 |
|
66 |
-
The
|
67 |
|
68 |
-
EXAMPLE:
|
69 |
-
Input:
|
70 |
-
Prompt: Identify a paradigm explaining the functional impact of BRCA1 mutations in ovarian cancer, focusing on DNA repair mechanisms.
|
71 |
-
Research Data: BRCA1 loss-of-function mutations are associated with impaired homologous recombination repair, leading to genomic instability in ovarian epithelial cells.
|
72 |
-
Agent's Response: BRCA1 mutations inhibit non-homologous end joining, which causes increased apoptosis in neurons, suggesting a neurodegeneration model.
|
73 |
|
74 |
Your output must begin with Score: and contain only two fields: Score: and Reasoning:. No extra commentary, no markdown, no explanations before or after.:
|
75 |
-
|
76 |
-
Reasoning: The prompt and research data focus on ovarian cancer and homologous recombination, but the response incorrectly shifts to neurons and the wrong DNA repair pathway (non-homologous end joining instead of homologous recombination). Misalignment between response and biological context.
|
77 |
-
|
78 |
-
|
79 |
Think step by step
|
80 |
'''
|
81 |
|
@@ -101,7 +97,7 @@ Think step by step
|
|
101 |
def Observation_LLM_Evaluator(self,promptversion):
|
102 |
SYSTEM='''
|
103 |
Task:
|
104 |
-
Evaluate the biological quality of a prompt
|
105 |
|
106 |
Goal:
|
107 |
Assess:
|
@@ -124,18 +120,8 @@ The prompt is vague or overly generic.
|
|
124 |
|
125 |
The response includes irrelevant, biologically implausible, contradictory, or trivial observations.
|
126 |
|
127 |
-
EXAMPLE:
|
128 |
-
Input:
|
129 |
-
Prompt: Generate diverse biological observations derived from the functional consequences of TP53 R175H mutations in epithelial tumors.
|
130 |
-
Research Data: TP53 R175H mutants lose sequence-specific DNA binding, form dominant-negative complexes with wild-type TP53, and lead to unchecked cell proliferation.
|
131 |
-
Agent's Response: TP53 R175H mutations increase glucose uptake in muscle cells and promote heart tissue regeneration.
|
132 |
-
|
133 |
Your output must begin with Score: and contain only two fields: Score: and Reasoning: No extra commentary, no markdown, no explanations before or after.
|
134 |
|
135 |
-
Output:
|
136 |
-
Score: 0.2
|
137 |
-
Reasoning: The response introduces observations unrelated to epithelial tumors or TP53's DNA binding function. The mention of muscle and heart tissue is off-context, and the observations are biologically implausible in this setting.
|
138 |
-
|
139 |
'''
|
140 |
data_to_evaluate=dbe.GetData(promptversion)
|
141 |
messages =[
|
|
|
42 |
|
43 |
SYSTEM='''
|
44 |
Task:
|
45 |
+
Evaluate the biological quality of a prompt, research data, paradigm list, and the selected paradigm on a 0–1 continuous scale.
|
46 |
|
47 |
Goal:
|
48 |
Assess:
|
49 |
|
50 |
+
Whether the Prompt is clear, biologically specific, and aligned with the Research Data and the Paradigm List.
|
51 |
|
52 |
+
Whether the selected Paradigm is biologically relevant, mechanistically coherent, and experimentally actionable based on the Research Data.
|
53 |
+
|
54 |
+
Whether the selected Paradigm is correctly chosen from the Paradigm List in light of the Research Data.
|
55 |
|
56 |
Scoring Guide (0–1 continuous scale):
|
57 |
|
58 |
Score 1.0 if:
|
59 |
|
60 |
+
The Prompt is clear, biologically detailed, and well-aligned to the Research Data and Paradigm List.
|
61 |
|
62 |
+
The selected Paradigm correctly reflects a biologically valid interpretation of the Research Data and is appropriately drawn from the Paradigm List.
|
63 |
|
64 |
Lower scores if:
|
65 |
|
66 |
+
The prompt is vague or misaligned with the research context.
|
67 |
+
|
68 |
+
The selected paradigm is biologically irrelevant, mechanistically incoherent, or mismatched with the Research Data.
|
69 |
|
70 |
+
The selected paradigm is not the most plausible or supported choice from the Paradigm List.
|
71 |
|
|
|
|
|
|
|
|
|
|
|
72 |
|
73 |
Your output must begin with Score: and contain only two fields: Score: and Reasoning:. No extra commentary, no markdown, no explanations before or after.:
|
74 |
+
|
|
|
|
|
|
|
75 |
Think step by step
|
76 |
'''
|
77 |
|
|
|
97 |
def Observation_LLM_Evaluator(self,promptversion):
|
98 |
SYSTEM='''
|
99 |
Task:
|
100 |
+
Evaluate the biological quality of a prompt , research data . response triplet from an Observations Generator Agent on a 0–1 continuous scale.
|
101 |
|
102 |
Goal:
|
103 |
Assess:
|
|
|
120 |
|
121 |
The response includes irrelevant, biologically implausible, contradictory, or trivial observations.
|
122 |
|
|
|
|
|
|
|
|
|
|
|
|
|
123 |
Your output must begin with Score: and contain only two fields: Score: and Reasoning: No extra commentary, no markdown, no explanations before or after.
|
124 |
|
|
|
|
|
|
|
|
|
125 |
'''
|
126 |
data_to_evaluate=dbe.GetData(promptversion)
|
127 |
messages =[
|