Fix bug on function evaluation

Files changed (4) hide show

__pycache__/prompt.cpython-311.pyc ADDED Viewed

Binary file (9.24 kB). View file

src/evaluation/__pycache__/compare_result.cpython-311.pyc ADDED Viewed

Binary file (3.85 kB). View file

src/prompts/__pycache__/prompt.cpython-311.pyc ADDED Viewed

Binary file (9.25 kB). View file

test_pretrained.ipynb CHANGED Viewed

@@ -256,7 +256,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -275,7 +275,7 @@
     "        query_output = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)\n",
     "\n",
     "        # Evaluate model result\n",
-    "        valid, sql_matched, result_matched = compare_result(row[\"sql_query\"], row[\"result\"], query_output)\n",
     "        if valid:\n",
     "            num_valid += 1\n",
     "        if sql_matched:\n",

   },
   {
    "cell_type": "code",
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
     "        query_output = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)\n",
     "\n",
     "        # Evaluate model result\n",
+    "        valid, sql_matched, result_matched = compare_result(cursor, row[\"sql_query\"], row[\"result\"], query_output)\n",
     "        if valid:\n",
     "            num_valid += 1\n",
     "        if sql_matched:\n",