|
|
|
|
|
import gradio as gr |
|
import json |
|
import os |
|
import re |
|
from get_llm_answer import get_model_response, parse_model_response, get_atla_response |
|
from jinja2 import Template |
|
|
|
def select_evaluators(criteria_group, df_state, prompt_state, save_prompt_button): |
|
with gr.Group(visible=True) as model_selection_group: |
|
select_evaluators_button = gr.Button("Select Evaluators", visible=False) |
|
|
|
|
|
def load_model_data(): |
|
model_data = {} |
|
try: |
|
script_dir = os.path.dirname(__file__) |
|
file_path = os.path.join(script_dir, "models.jsonl") |
|
with open(file_path, "r") as f: |
|
for line in f: |
|
model = json.loads(line) |
|
model_data[model["name"]] = { |
|
"organization": model["organization"], |
|
"license": model["license"], |
|
"api_model": model["api_model"], |
|
} |
|
except FileNotFoundError: |
|
print("Warning: models.jsonl not found") |
|
return {} |
|
return model_data |
|
|
|
model_data = load_model_data() |
|
model_choices = list(model_data.keys()) |
|
|
|
with gr.Row(visible=False) as evaluator_row: |
|
judge_a_dropdown = gr.Dropdown( |
|
choices=["Selene"], label="Judge A", value="Selene", interactive=False |
|
) |
|
judge_b_dropdown = gr.Dropdown( |
|
choices=model_choices, label="Judge B", value="Claude 3.5 Sonnet" |
|
) |
|
|
|
loading_spinner = gr.Markdown("Evaluation in progress...", visible=False) |
|
|
|
evaluation_result_df = gr.Dataframe( |
|
visible=False, |
|
label="Evaluation Results", |
|
elem_classes=["truncate_cells"] |
|
) |
|
|
|
with gr.Row(visible=False) as evaluation_nav_row: |
|
back_to_criteria_button = gr.Button("← Back to Criteria", visible=False) |
|
run_evaluation_button = gr.Button("Run Evaluation", visible=False) |
|
analyze_results_button = gr.Button("Analyze Results", visible=False) |
|
|
|
def show_evaluator_selection(current_df): |
|
updates = { |
|
criteria_group: gr.update(visible=False), |
|
save_prompt_button: gr.update(visible=False), |
|
evaluator_row: gr.update(visible=True), |
|
evaluation_nav_row: gr.update(visible=True), |
|
run_evaluation_button: gr.update(visible=True), |
|
back_to_criteria_button: gr.update(visible=True), |
|
analyze_results_button: gr.update(visible=False), |
|
evaluation_result_df: gr.update(visible=False), |
|
} |
|
if ( |
|
current_df.value is not None |
|
and hasattr(current_df.value, "attrs") |
|
and current_df.value.attrs.get("eval_done") |
|
): |
|
updates[loading_spinner] = gr.update(value="### Evaluation Complete", visible=True) |
|
updates[evaluation_result_df] = gr.update(value=current_df.value, visible=True) |
|
updates[analyze_results_button] = gr.update(visible=True) |
|
|
|
return updates |
|
|
|
save_prompt_button.click( |
|
fn=show_evaluator_selection, |
|
inputs=[df_state], |
|
outputs=[ |
|
save_prompt_button, |
|
criteria_group, |
|
evaluator_row, |
|
evaluation_nav_row, |
|
run_evaluation_button, |
|
back_to_criteria_button, |
|
loading_spinner, |
|
analyze_results_button, |
|
evaluation_result_df, |
|
], |
|
) |
|
|
|
def back_to_criteria(): |
|
return { |
|
save_prompt_button: gr.update(visible=True), |
|
criteria_group: gr.update(visible=True), |
|
evaluator_row: gr.update(visible=False), |
|
evaluation_nav_row: gr.update(visible=False), |
|
run_evaluation_button: gr.update(visible=False), |
|
loading_spinner: gr.update(visible=False), |
|
analyze_results_button: gr.update(visible=False), |
|
evaluation_result_df: gr.update(visible=False), |
|
} |
|
|
|
back_to_criteria_button.click( |
|
fn=back_to_criteria, |
|
inputs=[], |
|
outputs=[ |
|
save_prompt_button, |
|
criteria_group, |
|
evaluator_row, |
|
evaluation_nav_row, |
|
run_evaluation_button, |
|
loading_spinner, |
|
analyze_results_button, |
|
evaluation_result_df |
|
], |
|
) |
|
|
|
|
|
def run_evaluation(judge_a, judge_b): |
|
|
|
yield { |
|
loading_spinner: gr.update(value="Evaluation in progress...", visible=True), |
|
evaluation_result_df: gr.update(visible=False), |
|
analyze_results_button: gr.update(visible=False), |
|
run_evaluation_button: gr.update(interactive=False), |
|
back_to_criteria_button: gr.update(interactive=False), |
|
} |
|
|
|
|
|
template_str = prompt_state.value['template'] |
|
mappings = prompt_state.value['mappings'] |
|
evaluation_criteria = mappings.get('evaluation_criteria') |
|
|
|
template = Template(template_str) |
|
|
|
for index, row in df_state.value.iterrows(): |
|
context = {} |
|
model_context = None |
|
expected_output = None |
|
|
|
for key, column in mappings.items(): |
|
if key == 'evaluation_criteria': |
|
continue |
|
elif column and column != 'None': |
|
context[key] = str(row[column]) |
|
if column == 'model_context': |
|
model_context = str(row[column]) |
|
elif column == 'expected_model_output': |
|
expected_output = str(row[column]) |
|
|
|
context['evaluation_criteria'] = evaluation_criteria |
|
|
|
|
|
current_prompt = template.render(**context) |
|
print(f"\nDEBUG - Final Prompt sent to Model B:\n{current_prompt}\n") |
|
response_a = get_atla_response( |
|
"atla-selene", |
|
model_input=context.get('model_input'), |
|
model_output=context.get('model_output'), |
|
model_context=model_context, |
|
expected_output=expected_output, |
|
evaluation_criteria=evaluation_criteria |
|
) |
|
response_b = get_model_response( |
|
judge_b, |
|
model_data.get(judge_b), |
|
current_prompt |
|
) |
|
|
|
|
|
if isinstance(response_a, dict): |
|
score_a, critique_a = response_a['score'], response_a['critique'] |
|
else: |
|
score_a, critique_a = "Error", response_a |
|
|
|
score_b, critique_b = parse_model_response(response_b) |
|
|
|
|
|
model_b_snake = judge_b.lower().replace(' ', '_').replace('-', '_').replace('.', '_') |
|
|
|
df_state.value.loc[index, 'score_selene'] = score_a |
|
df_state.value.loc[index, 'critique_selene'] = critique_a |
|
df_state.value.loc[index, f'score_{model_b_snake}'] = score_b |
|
df_state.value.loc[index, f'critique_{model_b_snake}'] = critique_b |
|
|
|
import time |
|
time.sleep(2) |
|
|
|
|
|
yield {loading_spinner: gr.update(visible=False)} |
|
|
|
|
|
yield { |
|
loading_spinner: gr.update(value="### Evaluation Complete", visible=True), |
|
evaluation_result_df: gr.update(value=df_state.value, visible=True), |
|
analyze_results_button: gr.update(visible=True), |
|
run_evaluation_button: gr.update(interactive=True), |
|
back_to_criteria_button: gr.update(interactive=True), |
|
} |
|
|
|
if hasattr(df_state.value, "attrs"): |
|
df_state.value.attrs["eval_done"] = True |
|
|
|
|
|
run_evaluation_button.click( |
|
fn=run_evaluation, |
|
inputs=[judge_a_dropdown, judge_b_dropdown], |
|
outputs=[ |
|
loading_spinner, |
|
evaluation_result_df, |
|
analyze_results_button, |
|
run_evaluation_button, |
|
back_to_criteria_button, |
|
], |
|
) |
|
|
|
return model_selection_group, df_state, analyze_results_button |