more

Signed-off-by: SumanthRH <[email protected]>

Author: SumanthRH

.pre-commit-config.yaml

@@ -5,12 +5,12 @@ repos:
       - id: ruff
         args: [ --fix, --exit-non-zero-on-fix ]
         # NOTE (sumanthrh): Many of the files excluded here are used for validating code generation, and linters do not recognize some of the logic in these files. skythought/train is excluded for now because it's a fork of Llamafactory
-        exclude: (^skythought/train/.*|^skythought/skythought-rl/.*|tasks/taco/pyext2\.py|tasks/taco/taco_util\.py|tasks/apps/apps_util\.py|scripts/prompts\.py|skythought/test-time-scaling/.*)$
+        exclude: (^skythought/train/.*|^skythought/skythought-rl/.*|pyext2\.py|taco_util\.py|apps_util\.py|scripts/prompts\.py|skythought/test-time-scaling/.*)$
 
 
   # Black needs to be ran after ruff with --fix
   - repo: https://github.com/psf/black
     rev: 24.10.0
     hooks:
       - id: black
-        exclude: (^skythought/train/.*|^skythought/skythought-rl/.*|tasks/taco/pyext2\.py|skythought/test-time-scaling/.*)$
+        exclude: (^skythought/train/.*|^skythought/skythought-rl/.*|pyext2\.py|skythought/test-time-scaling/.*)$

recipes/sky-t1-preview/postprocess.py

@@ -1,14 +1,5 @@
-import copy
-import json
 from typing import Any, Dict
 
-import numpy as np
-import ray
-
-from skythought.evals.scoring.base import Scorer
-from skythought.evals.tasks.apps.apps_util import run_test as apps_run_test
-from skythought.evals.util.common import has_code
-
 STILL2_SYSTEM_PROMPT = "Your role as an assistant involves thoroughly exploring questions through a systematic long \
 thinking process before providing the final precise and accurate solutions. This requires \
 engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, \
@@ -30,41 +21,6 @@
 Now, try to solve the following question through the above guidelines:"
 
 
-class APPSScorer(Scorer):
-    def score(self, row: Dict[str, Any]):
-        TIMEOUT = 10
-        code_filter_result = has_code(row["response"])
-        if len(code_filter_result) == 0:
-            return False
-        else:
-            last_code = code_filter_result[-1]
-            problem_to_check = copy.deepcopy(row)
-            problem_to_check["input_output"] = json.loads(row["input_output"])
-            try:
-                problem_to_check["solutions"] = json.loads(row["solutions"])
-            except Exception:
-                problem_to_check["solutions"] = ""
-
-        @ray.remote
-        def _temp_run(problem, generation, debug):
-            try:
-                result = apps_run_test(problem=problem, test=generation, debug=debug)
-                return result
-            except Exception:
-                pass
-
-        result = ray.get(
-            _temp_run.remote(problem_to_check, last_code, False), timeout=TIMEOUT + 1
-        )
-
-        return bool(result and np.all(result[0]))
-
-
-class TACOScorer(Scorer):
-    def score(self, row: Dict[str, Any]):
-        return True
-
-
 def convert_to_sharegpt_format(row: Dict[str, Any]):
     prompt = row["user_input"]
     # accept

recipes/sky-t1-preview/prompts.py

@@ -1,4 +1,4 @@
-convert_prompt_example = (  # noqa: E501
+CONVERT_PROMPT_EXAMPLE = (  # noqa: E501
     "<|begin_of_thought|>\n\n"
     "Okay, so I've got this problem here. Mr. Wang leaves home at 6 AM, riding his bike at 12 km/h, "
     "and he stops to rest for 6 minutes after every 30 minutes of riding. Then, when he arrives at a park "
@@ -65,5 +65,5 @@
     "{example}\n"
     "Important: You should almost copy all the contents word-by-word of the original solution. Just convert them into two sections. "
     "Make sure you include: <|begin_of_slow_thought|>, <|end_of_slow_thought|>,  <|begin_of_solution|>,<|end_of_solution|>  These four headers explicitly. "
-    "Content to be converted: {{content}}".format(example=convert_prompt_example)
+    "Content to be converted: {content}"
 )

recipes/sky-t1-preview/recipe.py

@@ -13,11 +13,13 @@
     vLLMEngineProcessorConfig,
 )
 
+from skythought.evals.scoring.apps import APPSScorer
 from skythought.evals.scoring.math import MathEqualScorer
+from skythought.evals.scoring.taco import TACOScorer
 
-from .postprocess import APPSScorer, TACOScorer, convert_to_sharegpt_format
+from .postprocess import convert_to_sharegpt_format
 from .preprocess import APPSPreprocessor, NUMINAPreprocessor, TACOPreprocessor
-from .prompts import CONVERT_PROMPT
+from .prompts import CONVERT_PROMPT, CONVERT_PROMPT_EXAMPLE
 
 parser = argparse.ArgumentParser()
 parser.add_argument("--as-test", action="store_true")
@@ -26,11 +28,12 @@
 SYSTEM_PROMPT = "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step."  # noqa: E501
 
 # 1. Load datasets
-apps_ds = datasets.load_dataset("codeparrot/apps", split="test", streaming=True)
-taco_ds_medium = datasets.load_dataset(
-    "BAAI/TACO", split="test", name="MEDIUM", streaming=True
+apps_ds = datasets.load_dataset(
+    "codeparrot/apps",
+    split="test",
 )
-numina_ds = datasets.load_dataset("AI-MO/NuminaMath-CoT", split="train", streaming=True)
+taco_ds_medium = datasets.load_dataset("BAAI/TACO", split="test", name="MEDIUM")
+numina_ds = datasets.load_dataset("AI-MO/NuminaMath-CoT", split="train")
 
 # convert all to ray dataset
 apps_ds = ray.data.from_huggingface(apps_ds)
@@ -45,11 +48,11 @@
 
 
 if args.as_test:
-    apps_ds = apps_ds.limit(100)
-    taco_ds_medium = taco_ds_medium.limit(100)
-    numina_ds_amc_aime = numina_ds_amc_aime.limit(100)
-    numina_ds_olympiads = numina_ds_olympiads.limit(100)
-    numina_ds_math = numina_ds_math.limit(100)
+    apps_ds = apps_ds.limit(5)
+    taco_ds_medium = taco_ds_medium.limit(5)
+    numina_ds_amc_aime = numina_ds_amc_aime.limit(5)
+    numina_ds_olympiads = numina_ds_olympiads.limit(5)
+    numina_ds_math = numina_ds_math.limit(5)
 
 # 2. Get model responses for each of the datasets
 datasets = [
@@ -69,10 +72,20 @@
     NUMINAPreprocessor(),
 ]
 
+numina_scorer = MathEqualScorer(
+    response_column="formatted_response", answer_column="solution"
+)
+scorers = [
+    APPSScorer(response_column="formatted_response"),
+    TACOScorer(response_column="formatted_response"),
+    numina_scorer,
+    numina_scorer,
+    numina_scorer,
+]
+
 for i, ds in enumerate(datasets):
     datasets[i] = ds.map(preprocess_fns[i])
 
-    # our API
     config = vLLMEngineProcessorConfig(
         # model="Qwen/QwQ-32B-Preview",
         model="Qwen/Qwen2-0.5B-Instruct",
@@ -85,12 +98,11 @@
         batch_size=64,
     )
 
-    # our API
     processor = build_llm_processor(
         config,
         preprocess=lambda row: dict(
             messages=[
-                SYSTEM_PROMPT,
+                {"role": "system", "content": SYSTEM_PROMPT},
                 {"role": "user", "content": row["user_input"]},
             ],
             sampling_params=dict(
@@ -104,63 +116,56 @@
             **row,  # This will return all the original columns in the dataset.
         ),
     )
-    # our API
-    datasets[i] = processor(ds)
-
-# 3. Reformat the examples into a structured format
-# define a configuration for the reformatter
-config = HttpRequestProcessorConfig(
-    url="https://api.openai.com/v1/chat/completions",
-    headers={"Authorization": f"Bearer {os.environ['OPENAI_API_KEY']}"},
-    # number of processors to run in parallel
-    # Each handles a batch of requests
-    concurrency=1,
-)
-# define the reformatter
-reformatter = build_llm_processor(
-    config=config,
-    preprocess=lambda row: dict(
-        # define the payload / the exact arguments to the OpenAI chat completions API
-        payload=dict(
-            model="gpt-4o-mini",
-            messages=[
-                {"role": "system", "content": "You are a solution format convertor."},
-                {
-                    "role": "user",
-                    "content": CONVERT_PROMPT.format(
-                        content=f"{row['question']}\n{row['assistant_response']}"
-                    ),
-                },
-            ],
-            temperature=0.7,
-            max_tokens=16384,
+    datasets[i] = processor(datasets[i])
+
+    # 3. Reformat the examples into a structured format
+    # define a configuration for the reformatter
+    config = HttpRequestProcessorConfig(
+        url="https://api.openai.com/v1/chat/completions",
+        headers={"Authorization": f"Bearer {os.environ['OPENAI_API_KEY']}"},
+        # number of processors to run in parallel
+        # Each handles a batch of requests
+        concurrency=1,
+    )
+    # define the reformatter
+    reformatter = build_llm_processor(
+        config,
+        preprocess=lambda row: dict(
+            # define the payload / the exact arguments to the OpenAI chat completions API
+            payload=dict(
+                model="gpt-4o-mini",
+                messages=[
+                    {
+                        "role": "system",
+                        "content": "You are a solution format convertor.",
+                    },
+                    {
+                        "role": "user",
+                        "content": CONVERT_PROMPT.format(
+                            example=CONVERT_PROMPT_EXAMPLE,
+                            content=f"{row['question']}\n{row['assistant_response']}",
+                        ),
+                    },
+                ],
+                temperature=0.7,
+                max_tokens=2048,
+            ),
         ),
-    ),
-    postprocess=lambda row: dict(
-        formatted_response=row["http_response"]["choices"][0]["message"]["content"],
-    ),
-    batch_size=64,
-)
-
-for i, dataset in enumerate(datasets):
-    datasets[i] = reformatter(dataset)
-
-
-# 4. Rejection Sampling based on scoring
-# apps, taco, numina-amc-aime, numina-olympiads, numina-math
-numina_scorer = MathEqualScorer(
-    response_key="formatted_response", answer_key="solution"
-)
-scorers = [APPSScorer(), TACOScorer(), numina_scorer, numina_scorer, numina_scorer]
+        postprocess=lambda row: dict(
+            formatted_response=row["http_response"]["choices"][0]["message"]["content"],
+            **row,
+        ),
+    )
+    datasets[i] = reformatter(datasets[i])
 
-for i, dataset in enumerate(datasets):
-    fn = scorers[i]
-    datasets[i] = dataset.map(fn)
+    # 4. Rejection Sampling based on scoring
+    # apps, taco, numina-amc-aime, numina-olympiads, numina-math
+    datasets[i] = datasets[i].map(scorers[i])
+    # datasets[i] = datasets[i].filter(lambda x: x[scorers[i].SCORE_COLUMN])
 
-# 5. Convert to ShareGPT format
-for i, dataset in enumerate(datasets):
-    datasets[i] = dataset.map(convert_to_sharegpt_format)
+    # 5. Convert to ShareGPT format
+    datasets[i] = datasets[i].map(convert_to_sharegpt_format)
 
-# 6. Union + Save datasets
-datasets = datasets[0].union(*datasets[1:])
-datasets.write_parquet("sky-t1-preview.parquet")
+    # 6. Save datasets
+    dir_name = f"sky-t1-preview-{i}_parquet"
+    datasets[i].write_parquet(os.path.abspath(dir_name))

skythought/evals/scoring/__init__.py

@@ -1,7 +1,5 @@
 from .base import Scorer
 from .gsm8k import GSM8KScorer
-from .ifeval import IfEvalScorer
-from .livecodebench import LiveCodeBenchScorer
-from .math import MathScorer
+from .math import MathEqualScorer, MathVerifyScorer
 
-__all__ = ["Scorer", "MathScorer", "GSM8KScorer", "LiveCodeBenchScorer", "IfEvalScorer"]
+__all__ = ["Scorer", "MathEqualScorer", "MathVerifyScorer", "GSM8KScorer"]

skythought/evals/scoring/apps/__init__.py

@@ -0,0 +1,3 @@
+from .apps import APPSScorer
+
+__all__ = ["APPSScorer"]

skythought/evals/scoring/apps/apps.py

@@ -0,0 +1,61 @@
+import copy
+import json
+from typing import Any, Dict
+
+import numpy as np
+import ray
+from ray.exceptions import GetTimeoutError
+
+from skythought.evals.scoring.base import Scorer
+from skythought.evals.tasks.apps.apps_util import run_test as apps_run_test
+from skythought.evals.util.common import has_code
+
+
+class APPSScorer(Scorer):
+    SCORE_COLUMN = "apps_score"
+
+    def __init__(
+        self,
+        response_column="response",
+        answer_column="solutions",
+        input_column="input_output",
+    ) -> None:
+        super().__init__()
+        self.response_column = response_column
+        self.answer_column = answer_column
+        self.input_column = input_column
+
+    def score(self, row: Dict[str, Any]):
+        TIMEOUT = 10
+        code_filter_result = has_code(row[self.response_column])
+        if len(code_filter_result) == 0:
+            return {self.SCORE_COLUMN: False}
+        else:
+            last_code = code_filter_result[-1]
+            problem_to_check = copy.deepcopy(row)
+            problem_to_check[self.input_column] = json.loads(row[self.input_column])
+            try:
+                problem_to_check[self.answer_column] = json.loads(
+                    row[self.answer_column]
+                )
+            except Exception:
+                problem_to_check[self.answer_column] = ""
+
+        @ray.remote
+        def _temp_run(problem, generation, debug):
+            try:
+                result = apps_run_test(problem=problem, test=generation, debug=debug)
+                return result
+            except Exception:
+                pass
+
+        try:
+            result = ray.get(
+                _temp_run.remote(problem_to_check, last_code, False),
+                timeout=TIMEOUT + 1,
+            )
+        except GetTimeoutError:
+            result = []
+
+        score = bool(result and np.all(result[0]))
+        return {self.SCORE_COLUMN: score}

skythought/evals/scoring/ifeval/instructions.py

@@ -16,13 +16,13 @@
 """Library of instructions."""
 import collections
 import json
+import logging
 import random
 import re
 import string
 from typing import Dict, Optional, Sequence, Union
 
 import langdetect
-from absl import logging
 
 from . import instructions_util