{ "base_id": "huihui-ai/Qwen2.5-3B-Instruct-abliterated-SFT", "base_revision": "e21343fbcabbb6225d1cb9122061404f2d21f1ad", "trainable_parameters": 1843200, "training_rows": 165, "training_history": [ { "step": 1, "mean_loss": 1.38349 }, { "step": 20, "mean_loss": 1.05375 }, { "step": 40, "mean_loss": 0.56326 }, { "step": 60, "mean_loss": 0.21523 }, { "step": 80, "mean_loss": 0.31632 }, { "step": 100, "mean_loss": 0.26879 }, { "step": 120, "mean_loss": 0.28352 } ], "scratch": { "legacy_held_out": { "passed": 144, "total": 144, "outcomes": { "pass": 144 }, "example_failures": [], "exceeded_context": 0 }, "legacy_challenge": { "passed": 0, "total": 24, "outcomes": { "runtime_failure": 12, "invalid_syntax": 9, "wrong_result": 2, "outside_subset": 1 }, "example_failures": [ { "prompt": "def compose_0(a, b):\n \"\"\"Return the absolute value of the sum.\"\"\"\n", "response": "def compose_0(a, b):\n \"\"\"Return the absolute value of the sum.\"\"\"\n return abs(second - a) % 2 == 0\n", "outcome": "runtime_failure" }, { "prompt": "def compose_0(left, right):\n \"\"\"Return the absolute value of the sum.\"\"\"\n", "response": "def compose_0(left, right):\n \"\"\"Return the absolute value of the sum.\"\"\"\n return (second + right) % 2 ==\n", "outcome": "invalid_syntax" }, { "prompt": "def compose_0(x, y):\n \"\"\"Return the absolute value of the sum.\"\"\"\n", "response": "def compose_0(x, y):\n \"\"\"Return the absolute value of the sum.\"\"\"\n return m recond - y) > 0\n", "outcome": "invalid_syntax" } ], "exceeded_context": 0 }, "fresh_python": { "passed": 0, "total": 40, "outcomes": { "invalid_syntax": 30, "runtime_failure": 9, "wrong_result": 1 }, "example_failures": [ { "prompt": "def fresh_00_0(xs):\n \"\"\"Return differences of consecutive values, later minus earlier.\"\"\"\n", "response": "def fresh_00_0(xs):\n \"\"\"Return differences of consecutive values, later minus earlier.\"\"\"\n return (second < 0\n", "outcome": "invalid_syntax" }, { "prompt": "def fresh_00_1(xs):\n \"\"\"For every adjacent pair return the second item minus the first.\"\"\"\n", "response": "def fresh_00_1(xs):\n \"\"\"For every adjacent pair return the second item minus the first.\"\"\"\n return a > x\n", "outcome": "runtime_failure" }, { "prompt": "def fresh_01_0(xs):\n \"\"\"Keep only items that occur exactly once, preserving order.\"\"\"\n", "response": "def fresh_01_0(xs):\n \"\"\"Keep only items that occur exactly once, preserving order.\"\"\"\n returnd second\n", "outcome": "invalid_syntax" } ], "exceeded_context": 0 } }, "pretrained": { "legacy_held_out": { "passed": 124, "total": 144, "outcomes": { "pass": 124, "runtime_failure": 9, "wrong_result": 7, "outside_subset": 1, "invalid_syntax": 3 }, "example_failures": [ { "prompt": "def fn_080(q, first):\n \"\"\"Return the first input minus the second.\"\"\"\n", "response": "def fn_080(q, first):\n \"\"\"Return the first input minus the second.\"\"\"\n return first - q[1]", "outcome": "runtime_failure" }, { "prompt": "def fn_240(second, p):\n \"\"\"Subtract the second integer from the first.\"\"\"\n", "response": "def fn_240(second, p):\n \"\"\"Subtract the second integer from the first.\"\"\"\n return p - second", "outcome": "wrong_result" }, { "prompt": "def fn_280(a, first):\n \"\"\"Return the first input minus the second.\"\"\"\n", "response": "def fn_280(a, first):\n \"\"\"Return the first input minus the second.\"\"\"\n return first - a", "outcome": "wrong_result" } ] }, "legacy_challenge": { "passed": 24, "total": 24, "outcomes": { "pass": 24 }, "example_failures": [] }, "fresh_python": { "passed": 27, "total": 40, "outcomes": { "pass": 27, "wrong_result": 9, "outside_subset": 4 }, "example_failures": [ { "prompt": "def fresh_01_0(xs):\n \"\"\"Keep only items that occur exactly once, preserving order.\"\"\"\n", "response": "def fresh_01_0(xs):\n \"\"\"Keep only items that occur exactly once, preserving order.\"\"\"\n seen = {}\n result = []\n\n for item in xs:\n if item not in seen:\n seen[item] = 1\n result.append(item)\n\n return result", "outcome": "wrong_result" }, { "prompt": "def fresh_01_1(xs):\n \"\"\"Drop every occurrence of repeated values and preserve order.\"\"\"\n", "response": "def fresh_01_1(xs):\n \"\"\"Drop every occurrence of repeated values and preserve order.\"\"\"\n result = []\n seen = set()\n\n for item in xs:\n if item not in seen:\n result.append(item)\n seen.add(item)\n\n return result", "outcome": "wrong_result" }, { "prompt": "def fresh_05_0(xs):\n \"\"\"Return the alternating sum, starting by adding the first item.\"\"\"\n", "response": "def fresh_05_0(xs):\n \"\"\"Return the alternating sum, starting by adding the first item.\"\"\"\n if len(xs) == 0:\n return 0\n elif len(xs) == 1:\n return xs[0]\n else:\n return xs[0] - xs[1] + fresh_05_0(xs[2:])", "outcome": "outside_subset" } ] } }, "fine_tuned": { "legacy_held_out": { "passed": 125, "total": 144, "outcomes": { "pass": 125, "wrong_result": 8, "runtime_failure": 8, "invalid_syntax": 2, "outside_subset": 1 }, "example_failures": [ { "prompt": "def fn_080(q, first):\n \"\"\"Return the first input minus the second.\"\"\"\n", "response": "def fn_080(q, first):\n \"\"\"Return the first input minus the second.\"\"\"\n return first - q\n", "outcome": "wrong_result" }, { "prompt": "def fn_240(second, p):\n \"\"\"Subtract the second integer from the first.\"\"\"\n", "response": "def fn_240(second, p):\n \"\"\"Subtract the second integer from the first.\"\"\"\n return p - second\n", "outcome": "wrong_result" }, { "prompt": "def fn_280(a, first):\n \"\"\"Return the first input minus the second.\"\"\"\n", "response": "def fn_280(a, first):\n \"\"\"Return the first input minus the second.\"\"\"\n return first - a\n", "outcome": "wrong_result" } ] }, "legacy_challenge": { "passed": 24, "total": 24, "outcomes": { "pass": 24 }, "example_failures": [] }, "fresh_python": { "passed": 24, "total": 40, "outcomes": { "pass": 24, "wrong_result": 12, "outside_subset": 3, "invalid_syntax": 1 }, "example_failures": [ { "prompt": "def fresh_01_0(xs):\n \"\"\"Keep only items that occur exactly once, preserving order.\"\"\"\n", "response": "def fresh_01_0(xs):\n \"\"\"Keep only items that occur exactly once, preserving order.\"\"\"\n seen = {}\n result = []\n\n for item in xs:\n if item not in seen:\n seen[item] = 1\n result.append(item)\n\n return result\n", "outcome": "wrong_result" }, { "prompt": "def fresh_01_1(xs):\n \"\"\"Drop every occurrence of repeated values and preserve order.\"\"\"\n", "response": "def fresh_01_1(xs):\n \"\"\"Drop every occurrence of repeated values and preserve order.\"\"\"\n seen = []\n result = []\n\n for item in xs:\n if item not in seen:\n seen.append(item)\n result.append(item)\n\n return result\n", "outcome": "wrong_result" }, { "prompt": "def fresh_05_1(xs):\n \"\"\"Add values at even indices and subtract values at odd indices.\"\"\"\n", "response": "def fresh_05_1(xs):\n \"\"\"Add values at even indices and subtract values at odd indices.\"\"\"\n return [x * 2 if i % 2 == 0 else -x for i, x in enumerate(xs)]\n", "outcome": "wrong_result" } ] } }, "development_before": { "new_wording": { "passed": 28, "total": 30, "outcomes": { "wrong_result": 1, "pass": 28, "outside_subset": 1 }, "example_failures": [ { "prompt": "def training_00_1(xs):\n \"\"\"Add the items that are strictly positive.\"\"\"\n", "response": "def training_00_1(xs):\n \"\"\"Add the items that are strictly positive.\"\"\"\n result = []\n for x in xs:\n if x > 0:\n result.append(x)\n return result", "outcome": "wrong_result" }, { "prompt": "def training_23_1(s):\n \"\"\"Build a character frequency map.\"\"\"\n", "response": "def training_23_1(s):\n \"\"\"Build a character frequency map.\"\"\"\n freq_map = {}\n \n for char in s:\n if char in freq_map:\n freq_map[char] += 1\n else:\n freq_map[char] = 1\n \n return freq_map", "outcome": "outside_subset" } ] }, "new_families": { "passed": 10, "total": 22, "outcomes": { "pass": 10, "runtime_failure": 2, "outside_subset": 4, "wrong_result": 6 }, "example_failures": [ { "prompt": "def development_01_0(xs):\n \"\"\"Return the last index of the maximum, or -1 if empty.\"\"\"\n", "response": "def development_01_0(xs):\n \"\"\"Return the last index of the maximum, or -1 if empty.\"\"\"\n max_val = max(xs)\n max_index = [i for i, val in enumerate(xs) if val == max_val]\n return max_index[0] if max_index else -1", "outcome": "runtime_failure" }, { "prompt": "def development_01_1(xs):\n \"\"\"Find the position of the final occurrence of the largest item.\"\"\"\n", "response": "def development_01_1(xs):\n \"\"\"Find the position of the final occurrence of the largest item.\"\"\"\n max_value = max(xs)\n max_index = xs.index(max_value)\n return max_index", "outcome": "runtime_failure" }, { "prompt": "def development_02_0(xs):\n \"\"\"Return the number of repeated occurrences beyond the first occurrence of each value.\"\"\"\n", "response": "def development_02_0(xs):\n \"\"\"Return the number of repeated occurrences beyond the first occurrence of each value.\"\"\"\n repeated_counts = {}\n for i in range(1, len(xs)):\n if xs[i] in repeated_counts:\n repeated_counts[xs[i]] += 1\n else:\n repeated_counts[xs[i]] = 1\n return repeated_counts", "outcome": "outside_subset" } ] } }, "development_after": { "0.5": { "new_wording": { "passed": 27, "total": 30, "outcomes": { "wrong_result": 1, "pass": 27, "invalid_syntax": 1, "outside_subset": 1 }, "example_failures": [ { "prompt": "def training_00_1(xs):\n \"\"\"Add the items that are strictly positive.\"\"\"\n", "response": "def training_00_1(xs):\n \"\"\"Add the items that are strictly positive.\"\"\"\n return [x for x in xs if x > 0]", "outcome": "wrong_result" }, { "prompt": "def training_06_1(xs):\n \"\"\"Check each item is at least as large as the previous one.\"\"\"\n", "response": "def training_06_1(xs):\n \"\"\"Check each item is at least as large as the previous one.\"\"\"\n return all(x >= y for x, y in zip(xs, xs[1:])\n\n# Test cases\nprint(training_06_1([1, 2, 3])) # True\nprint(training_06_1([1, 2, 1])) # False", "outcome": "invalid_syntax" }, { "prompt": "def training_23_1(s):\n \"\"\"Build a character frequency map.\"\"\"\n", "response": "def training_23_1(s):\n \"\"\"Build a character frequency map.\"\"\"\n freq_map = {}\n \n for char in s:\n if char in freq_map:\n freq_map[char] += 1\n else:\n freq_map[char] = 1\n \n return freq_map", "outcome": "outside_subset" } ] }, "new_families": { "passed": 12, "total": 22, "outcomes": { "pass": 12, "runtime_failure": 2, "outside_subset": 4, "wrong_result": 4 }, "example_failures": [ { "prompt": "def development_01_0(xs):\n \"\"\"Return the last index of the maximum, or -1 if empty.\"\"\"\n", "response": "def development_01_0(xs):\n \"\"\"Return the last index of the maximum, or -1 if empty.\"\"\"\n max_val = max(xs)\n max_index = [i for i, val in enumerate(xs) if val == max_val]\n return max_index[0] if max_index else -1", "outcome": "runtime_failure" }, { "prompt": "def development_01_1(xs):\n \"\"\"Find the position of the final occurrence of the largest item.\"\"\"\n", "response": "def development_01_1(xs):\n \"\"\"Find the position of the final occurrence of the largest item.\"\"\"\n max_value = max(xs)\n max_index = xs.index(max_value)\n return max_index", "outcome": "runtime_failure" }, { "prompt": "def development_02_0(xs):\n \"\"\"Return the number of repeated occurrences beyond the first occurrence of each value.\"\"\"\n", "response": "def development_02_0(xs):\n \"\"\"Return the number of repeated occurrences beyond the first occurrence of each value.\"\"\"\n repeated_counts = {}\n for i in range(1, len(xs)):\n if xs[i] in repeated_counts:\n repeated_counts[xs[i]] += 1\n else:\n repeated_counts[xs[i]] = 1\n return repeated_counts\n", "outcome": "outside_subset" } ] } }, "1.0": { "new_wording": { "passed": 27, "total": 30, "outcomes": { "wrong_result": 2, "pass": 27, "outside_subset": 1 }, "example_failures": [ { "prompt": "def training_00_1(xs):\n \"\"\"Add the items that are strictly positive.\"\"\"\n", "response": "def training_00_1(xs):\n \"\"\"Add the items that are strictly positive.\"\"\"\n return [x for x in xs if x > 0]", "outcome": "wrong_result" }, { "prompt": "def training_06_1(xs):\n \"\"\"Check each item is at least as large as the previous one.\"\"\"\n", "response": "def training_06_1(xs):\n \"\"\"Check each item is at least as large as the previous one.\"\"\"\n return all(x >= y for x, y in zip(xs, xs[1:]))\n", "outcome": "wrong_result" }, { "prompt": "def training_23_1(s):\n \"\"\"Build a character frequency map.\"\"\"\n", "response": "def training_23_1(s):\n \"\"\"Build a character frequency map.\"\"\"\n freq_map = {}\n \n for char in s:\n if char in freq_map:\n freq_map[char] += 1\n else:\n freq_map[char] = 1\n \n return freq_map", "outcome": "outside_subset" } ] }, "new_families": { "passed": 13, "total": 22, "outcomes": { "pass": 13, "runtime_failure": 1, "outside_subset": 5, "wrong_result": 3 }, "example_failures": [ { "prompt": "def development_01_0(xs):\n \"\"\"Return the last index of the maximum, or -1 if empty.\"\"\"\n", "response": "def development_01_0(xs):\n \"\"\"Return the last index of the maximum, or -1 if empty.\"\"\"\n max_val = max(xs)\n max_index = [i for i, val in enumerate(xs) if val == max_val][-1] if max_val in xs else -1\n return max_index\n", "outcome": "runtime_failure" }, { "prompt": "def development_01_1(xs):\n \"\"\"Find the position of the final occurrence of the largest item.\"\"\"\n", "response": "def development_01_1(xs):\n \"\"\"Find the position of the final occurrence of the largest item.\"\"\"\n try:\n return max(xs[::-1]).index(xs[-1])\n except ValueError:\n return -1", "outcome": "outside_subset" }, { "prompt": "def development_02_0(xs):\n \"\"\"Return the number of repeated occurrences beyond the first occurrence of each value.\"\"\"\n", "response": "def development_02_0(xs):\n \"\"\"Return the number of repeated occurrences beyond the first occurrence of each value.\"\"\"\n repeated_counts = {}\n for i in range(1, len(xs)):\n if xs[i] in repeated_counts:\n repeated_counts[xs[i]] += 1\n else:\n repeated_counts[xs[i]] = 1\n return repeated_counts\n", "outcome": "outside_subset" } ] } }, "2.0": { "new_wording": { "passed": 26, "total": 30, "outcomes": { "wrong_result": 2, "pass": 26, "runtime_failure": 1, "outside_subset": 1 }, "example_failures": [ { "prompt": "def training_00_1(xs):\n \"\"\"Add the items that are strictly positive.\"\"\"\n", "response": "def training_00_1(xs):\n \"\"\"Add the items that are strictly positive.\"\"\"\n return [x for x in xs if x > 0]\n", "outcome": "wrong_result" }, { "prompt": "def training_06_1(xs):\n \"\"\"Check each item is at least as large as the previous one.\"\"\"\n", "response": "def training_06_1(xs):\n \"\"\"Check each item is at least as large as the previous one.\"\"\"\n return all(x >= y for x, y in zip(xs, xs[1:]))\n", "outcome": "wrong_result" }, { "prompt": "def training_07_1(xs):\n \"\"\"What is the range of the values? Use zero for an empty list.\"\"\"\n", "response": "def training_07_1(xs):\n \"\"\"What is the range of the values? Use zero for an empty list.\"\"\"\n return [max(xs) - min(xs) for xs in [xs]]\n", "outcome": "runtime_failure" } ] }, "new_families": { "passed": 15, "total": 22, "outcomes": { "pass": 15, "wrong_result": 3, "outside_subset": 4 }, "example_failures": [ { "prompt": "def development_01_0(xs):\n \"\"\"Return the last index of the maximum, or -1 if empty.\"\"\"\n", "response": "def development_01_0(xs):\n \"\"\"Return the last index of the maximum, or -1 if empty.\"\"\"\n return max([x for x in reversed(xs)], default=-1)\n", "outcome": "wrong_result" }, { "prompt": "def development_01_1(xs):\n \"\"\"Find the position of the final occurrence of the largest item.\"\"\"\n", "response": "def development_01_1(xs):\n \"\"\"Find the position of the final occurrence of the largest item.\"\"\"\n try:\n return max(xs[::-1].index(x) for x in xs)\n except ValueError:\n return -1\n", "outcome": "outside_subset" }, { "prompt": "def development_02_0(xs):\n \"\"\"Return the number of repeated occurrences beyond the first occurrence of each value.\"\"\"\n", "response": "def development_02_0(xs):\n \"\"\"Return the number of repeated occurrences beyond the first occurrence of each value.\"\"\"\n return {x: sum(1 for y in xs if y == x and y != xs[0]) for x in set(xs) if xs.count(x) > 1}\n", "outcome": "wrong_result" } ] } } }, "prefill_signature": true, "adapter_strength": 2.0 }