updated lm-eval task

lm-eval-task/_ds1000_base.yml

@@ -0,0 +1,24 @@
+dataset_path: PNYX/ds1000_pnyx
+unsafe_code: true
+output_type: generate_until
+test_split: test
+doc_to_text: "{{user_chat_prompt}}"
+doc_to_target: "{{test_code}}"
+process_results: !function utils.pass_at_k_process
+metric_list:
+  - metric: pass@1
+    aggregation: mean
+    higher_is_better: true
+generation_kwargs:
+  until: []
+  max_gen_toks: 30000
+  do_sample: false
+repeats: 1
+num_fewshot: 0
+filter_list:
+  - name: "create_test"
+    filter:
+      - function: "custom"
+        filter_fn: !function utils.build_predictions_chat_ds1000
+metadata:
+  version: 1.0

lm-eval-task/numpy_chat.yaml

@@ -1,26 +1,5 @@
+include: _ds1000_base.yml
 task: ds1000_numpy_sandbox_chat
-dataset_path: PNYX/ds1000_pnyx
 dataset_name: Numpy
-unsafe_code: true
-output_type: generate_until
-test_split: test
-doc_to_text: "{{user_chat_prompt}}"
-doc_to_target: "{{test_code}}"
-metric_list:
-  - metric: !function utils.pass_at_k
-    aggregation: mean
-    higher_is_better: true
-    k: [1]
-generation_kwargs:
-  until: []
-  max_gen_toks: 8000
-  do_sample: false
-repeats: 1
-num_fewshot: 0
-filter_list:
-  - name: "create_test"
-    filter:
-      - function: "custom"
-        filter_fn: !function utils.build_predictions_chat_ds1000
 metadata:
   version: 1.0

lm-eval-task/pandas_chat.yaml

@@ -1,3 +1,5 @@
-include: numpy_chat.yaml
+include: _ds1000_base.yml
 task: ds1000_pandas_sandbox_chat
-dataset_name: Pandas
+dataset_name: Pandas
+metadata:
+  version: 1.0

lm-eval-task/pytorch_chat.yaml

@@ -1,3 +1,5 @@
-include: numpy_chat.yaml
+include: _ds1000_base.yml
 task: ds1000_pytorch_sandbox_chat
-dataset_name: Pytorch
+dataset_name: Pytorch
+metadata:
+  version: 1.0

lm-eval-task/scipy_chat.yaml

@@ -1,3 +1,5 @@
-include: numpy_chat.yaml
+include: _ds1000_base.yml
 task: ds1000_scipy_sandbox_chat
-dataset_name: Scipy
+dataset_name: Scipy
+metadata:
+  version: 1.0

lm-eval-task/sklearn_chat.yaml

@@ -1,3 +1,5 @@
-include: numpy_chat.yaml
+include: _ds1000_base.yml
 task: ds1000_sklearn_sandbox_chat
-dataset_name: Sklearn
+dataset_name: Sklearn
+metadata:
+  version: 1.0

lm-eval-task/tensorflow_chat.yaml

@@ -1,3 +1,5 @@
-include: numpy_chat.yaml
+include: _ds1000_base.yml
 task: ds1000_tensorflow_sandbox_chat
-dataset_name: Tensorflow
+dataset_name: Tensorflow
+metadata:
+  version: 1.0

lm-eval-task/utils.py

@@ -11,18 +11,28 @@ except Exception as e:
     raise e
 
 
-def pass_at_k(references: list[str], predictions: list[list[str]], k: list[int] = None):
+
+def pass_at_k_process(doc, results):
     global compute_
-    assert k is not None
-    if isinstance(k, int):
-        k = [k]
+    
+    result = {"pass@1": 0}
+
+    # We support a single prediction currently
+    prediction = results[0][0]
+    if prediction.strip() == "" or prediction is None:
+        return result
+    
+    assert doc["test_code"] is not None
+
+    # Compute
     res = compute_.compute(
-        references=references,
-        predictions=predictions,
-        k=k,
+        references=[doc["test_code"]],
+        predictions=[[prediction]],
+        k=[1],
     )
-    return res[0]
+    result["pass@1"] = res[0]['pass@1']
 
+    return result
 
 def build_predictions_chat_ds1000(
     resps: list[list[str]], docs: list[dict]