File size: 4,181 Bytes
c7a6fe6 | 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 | import json
import requests
import os
# --- Configuration ---
DEV_SET_PATH = "/home/mshahidul/readctrl/data/new_exp/test_health_literacy_data.json"
FEW_SHOT_SET_PATH = "/home/mshahidul/readctrl/data/new_exp/final_prompt_template_info.json" # Using the one with reasoning
LOCAL_API_URL = "http://172.16.34.29:8004/v1/chat/completions"
LOCAL_MODEL_NAME = "Qwen/Qwen3-30B-A3B-Instruct-2507"
# Define the range of few-shots per label you want to test
# e.g., [0, 1, 2, 3] will test 0-shot, 1-shot (3 total), 2-shot (6 total), etc.
SHOTS_TO_EVALUATE = [0, 1, 2, 3,4,5,6]
# --- Core Functions ---
def build_dynamic_prompt(few_shot_data, k_per_label):
"""Constructs a prompt with k examples per literacy category."""
instruction = (
"You are an expert in health communication. Your task is to judge the health literacy level of the provided text.\n"
"Classify the text into: low_health_literacy, intermediate_health_literacy, or proficient_health_literacy.\n\n"
)
if k_per_label == 0:
return instruction + "### Task:\nTarget Text: \"{input_text}\"\nReasoning:"
# Organize few-shot data by label
categorized = {}
for entry in few_shot_data:
label = entry['label']
categorized.setdefault(label, []).append(entry)
few_shot_blocks = "### Examples:\n"
labels = ["low_health_literacy", "intermediate_health_literacy", "proficient_health_literacy"]
for label in labels:
examples = categorized.get(label, [])[:k_per_label]
for ex in examples:
few_shot_blocks += f"Target Text: \"{ex['gen_text']}\"\n"
few_shot_blocks += f"Reasoning: {ex['reasoning']}\n"
few_shot_blocks += f"Label: {label}\n"
few_shot_blocks += "-" * 30 + "\n"
return instruction + few_shot_blocks + "\n### Task:\nTarget Text: \"{input_text}\"\nReasoning:"
def get_prediction(prompt_template, input_text):
"""Sends the formatted prompt to the local LLM."""
final_prompt = prompt_template.format(input_text=input_text)
payload = {
"model": LOCAL_MODEL_NAME,
"messages": [{"role": "user", "content": final_prompt}],
"temperature": 0
}
try:
response = requests.post(LOCAL_API_URL, json=payload, timeout=30)
return response.json()['choices'][0]['message']['content'].strip()
except Exception:
return "Error"
def parse_label(text):
"""Normalizes LLM output to match dataset labels."""
text = text.lower()
if "low" in text: return "low_health_literacy"
if "intermediate" in text: return "intermediate_health_literacy"
if "proficient" in text: return "proficient_health_literacy"
return "unknown"
# --- Main Execution ---
# 1. Load Data
with open(DEV_SET_PATH, 'r') as f:
dev_set = json.load(f)
with open(FEW_SHOT_SET_PATH, 'r') as f:
few_shot_pool = json.load(f)
# 2. Filter Dev Set
# Ensure no overlap between few-shot examples and dev set
shot_ids = {item['doc_id'] for item in few_shot_pool}
clean_dev_set = [item for item in dev_set if item['doc_id'] not in shot_ids]
results_summary = []
print(f"Starting Evaluation on {len(clean_dev_set)} samples...\n")
# 3. Loop through shot counts
for k in SHOTS_TO_EVALUATE:
print(f"Evaluating {k}-shot per label (Total {k*3} examples)...")
current_template = build_dynamic_prompt(few_shot_pool, k)
correct = 0
for case in clean_dev_set:
raw_output = get_prediction(current_template, case['gen_text'])
pred = parse_label(raw_output)
actual = parse_label(case['label'])
if pred == actual:
correct += 1
accuracy = (correct / len(clean_dev_set)) * 100
results_summary.append({"shots_per_label": k, "accuracy": accuracy})
print(f"-> Accuracy: {accuracy:.2f}%\n")
# --- Final Report ---
print("-" * 30)
print(f"{'Shots/Label':<15} | {'Accuracy':<10}")
print("-" * 30)
for res in results_summary:
print(f"{res['shots_per_label']:<15} | {res['accuracy']:.2f}%")
with open("/home/mshahidul/readctrl/data/new_exp/few_shot_evaluation_summary.json", 'w') as f:
json.dump(results_summary, f, indent=4) |