Spaces:
Sleeping
Sleeping
File size: 6,760 Bytes
9ceb38c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 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 | import gradio as gr
from src.api_clients import openai_client, claude_client, deepseek_client
from src.summary_generator import get_openrouter_models, generate_summary_from_openrouter, is_prompt_valid_for_summary
from src.evaluation import evaluate
from src.comparison import get_last_eval_data, run_comparison, import_model_metrics
from src.utils import preset_vals
from config.css import CSS
with gr.Blocks(css=CSS) as demo:
with gr.Tabs():
# ▸ Tab 1: Unified Summary + Evaluation
with gr.Tab("🧠 Generate + Evaluate"):
gr.Markdown("## Unified Summary Generator and Evaluator")
article = gr.Textbox(label="Paste article", lines=10)
auto_gen_toggle = gr.Checkbox(label="Enable Auto-Generation if Summary is Empty", value=True)
summary = gr.Textbox(label="Paste summary", lines=6, visible=False, interactive=True)
prompt = gr.Textbox(label="Prompt for summary (only if generating)", lines=2, visible=True, interactive=True)
model = gr.Dropdown(choices=get_openrouter_models(), label="Model (for generation)", visible=True, interactive=True)
variant = gr.Radio(["Twin-Lock","Judge-Lock","ParallelX-TJ"], value="Twin-Lock", label="Variant", elem_id="variant-group")
back = gr.CheckboxGroup(["OpenAI","DeepSeek","Claude"], value=["OpenAI","DeepSeek","Claude"], label="Back-ends", elem_id="backend-group")
p0 = preset_vals("Twin-Lock")
w_cov = gr.Slider(0,1,p0[0],step=0.01,label="Coverage", elem_classes=["metric-slider"])
w_align= gr.Slider(0,1,p0[1],step=0.01,label="Alignment", elem_classes=["metric-slider"])
w_hall = gr.Slider(0,1,p0[2],step=0.01,label="Hallucination", elem_classes=["metric-slider"])
w_rel = gr.Slider(0,1,p0[3],step=0.01,label="Relevance", elem_classes=["metric-slider"])
w_bias = gr.Slider(0,1,p0[4],step=0.01,label="Bias/Toxicity", elem_classes=["metric-slider"])
temp = gr.Slider(0,1,0,step=0.01,label="temperature")
show_ev = gr.Checkbox(True,label="Show evidence spans")
run = gr.Button("🔁 Generate & Evaluate")
gen_sum = gr.Textbox(label="Generated Summary", lines=6, visible=True)
table = gr.DataFrame(label="Metrics")
comm = gr.JSON(label="Comments JSON")
score = gr.JSON(label="Average score")
tokbox = gr.JSON(label="Token usage")
csv_dl = gr.File(label="CSV download")
zip_dl = gr.File(label="Raw JSON zip")
def toggle_ui(auto):
return (
gr.update(visible=not auto, interactive=not auto),
gr.update(visible=auto, interactive=auto),
gr.update(visible=auto, interactive=auto),
gr.update(value=("🔁 Generate & Evaluate" if auto else "✅ Only Evaluate")),
gr.update(visible=auto) # toggle gen_sum
)
auto_gen_toggle.change(
toggle_ui,
auto_gen_toggle,
[summary, prompt, model, run, gen_sum]
)
def unified_run(article, prompt, model, summary, auto_flag, variant, active_back, temp,
w_cov, w_align, w_hall, w_rel, w_bias, show_ev):
if auto_flag and not summary:
if not is_prompt_valid_for_summary(prompt):
return "⛔ Prompt rejected: not summarization-related.", None, None, None, None, None, None
summary = generate_summary_from_openrouter(article, prompt, model)
elif not auto_flag and not summary:
return "⚠️ Please provide a summary or enable auto-generation.", None, None, None, None, None, None
return summary, *evaluate(article, summary, variant, active_back, temp, w_cov, w_align, w_hall, w_rel, w_bias, show_ev)
run.click(
unified_run,
[article, prompt, model, summary, auto_gen_toggle, variant, back, temp,
w_cov,w_align,w_hall,w_rel,w_bias,show_ev],
[gen_sum, table, comm, score, tokbox, csv_dl, zip_dl]
)
# ▸ Tab 3: Comparison
# ▸ Tab 3: Comparison (manual inputs for both human and model, no import button)
with gr.Tab("📊 Comparison"):
with gr.Column():
gr.Markdown("## Compare Human vs Model Evaluations")
gr.Markdown("### 🧍 Human Evaluation")
with gr.Row():
hc = gr.Slider(1, 10, 1, step=1, label="Coverage")
ha = gr.Slider(1, 10, 1, step=1, label="Alignment")
hh = gr.Slider(1, 10, 1, step=1, label="Hallucination")
hr = gr.Slider(1, 10, 1, step=1, label="Relevance")
hb = gr.Slider(1, 10, 1, step=1, label="Bias/Toxicity")
human_comments = gr.Textbox(label="Human Comments")
gr.Markdown("### 🤖 Model Evaluation")
with gr.Row():
mc = gr.Slider(1, 10, 1, step=1, label="Coverage")
ma = gr.Slider(1, 10, 1, step=1, label="Alignment")
mh = gr.Slider(1, 10, 1, step=1, label="Hallucination")
mr = gr.Slider(1, 10, 1, step=1, label="Relevance")
mb = gr.Slider(1, 10, 1, step=1, label="Bias/Toxicity")
model_comments = gr.Textbox(label="Model Comments")
def compare_structured(hc, ha, hh, hr, hb, human_comments, mc, ma, mh, mr, mb, model_comments):
human_scores = {
"coverage": hc,
"alignment": ha,
"hallucination": hh,
"relevance": hr,
"bias_toxicity": hb
}
model_scores = {
"coverage": mc,
"alignment": ma,
"hallucination": mh,
"relevance": mr,
"bias_toxicity": mb
}
return run_comparison(human_scores, human_comments, model_scores, model_comments)
compare_btn = gr.Button("🔍 Compare")
output_analysis = gr.Textbox(label="Analysis", lines=10)
compare_btn.click(
compare_structured,
[hc, ha, hh, hr, hb, human_comments, mc, ma, mh, mr, mb, model_comments],
output_analysis
)
demo.launch(share=True, show_error=True) |