Rename evaluator/ to eval/ and remove legacy evaluator.py files

README.md

@@ -76,13 +76,13 @@ export OPENAI_API_KEY="<your-key>"
 # Try the circle packing benchmark
 uv sync --extra math
 uv run skydiscover-run benchmarks/math/circle_packing/initial_program.py \
-  benchmarks/math/circle_packing/evaluator.py \
+  benchmarks/math/circle_packing/eval \
   --config benchmarks/math/circle_packing/config.yaml \
   --search evox \
   --iterations 100
 
 uv run skydiscover-run benchmarks/math/circle_packing/initial_program.py \
-  benchmarks/math/circle_packing/evaluator.py \
+  benchmarks/math/circle_packing/eval \
   --config benchmarks/math/circle_packing/config.yaml \
   --search adaevolve \
   --iterations 100

benchmarks/ADRS/README.md

@@ -46,15 +46,15 @@ Given a set of database transactions with read/write dependencies on shared keys
 
 Each benchmark directory contains:
 - `initial_program.py` — the seed solution for evolution
-- `evaluator.py` — the scoring function
+- `eval/` — containerized evaluator directory
 - `config.yaml` — run configuration
 
 Run any benchmark from the repo root:
 
 ```bash
 uv run skydiscover-run \
   benchmarks/ADRS/cloudcast/initial_program.py \
-  benchmarks/ADRS/cloudcast/evaluator.py \
+  benchmarks/ADRS/cloudcast/eval \
   -c benchmarks/ADRS/cloudcast/config.yaml \
   -s [your_algorithm] \
   -i 100

benchmarks/ADRS/cloudcast/README.md

@@ -31,7 +31,7 @@ From the repo root:
 ```bash
 uv run skydiscover-run \
   benchmarks/ADRS/cloudcast/initial_program.py \
-  benchmarks/ADRS/cloudcast/evaluator.py \
+  benchmarks/ADRS/cloudcast/eval \
   -c benchmarks/ADRS/cloudcast/config.yaml \
   -s [your_algorithm] \
   -i 100
@@ -42,7 +42,7 @@ uv run skydiscover-run \
 | File | Description |
 |------|-------------|
 | `initial_program.py` | Baseline `search_algorithm` function to evolve |
-| `evaluator.py` | Scores programs on total transfer cost across 5 network configs |
+| `eval/` | Containerized evaluator — scores programs on total transfer cost across 5 network configs |
 | `config.yaml` | Task-specific config (LLM, evaluator timeout, system prompt) |
 | `simulator.py` | Broadcast cost simulator |
 | `broadcast.py` | `BroadCastTopology` data structure |

benchmarks/ADRS/cloudcast/config.yaml

@@ -1,5 +1,5 @@
 # CloudCast — Cloud Broadcast Optimization (NSDI'24)
-# Usage: skydiscover-run initial_program.py evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.py eval -c config.yaml -s <strategy>
 language: python
 diff_based_generation: true
 max_iterations: 100

benchmarks/ADRS/eplb/README.md

@@ -30,7 +30,7 @@ From the repo root:
 ```bash
 uv run skydiscover-run \
   benchmarks/ADRS/eplb/initial_program.py \
-  benchmarks/ADRS/eplb/evaluator.py \
+  benchmarks/ADRS/eplb/eval \
   -c benchmarks/ADRS/eplb/config.yaml \
   -s [your_algorithm] \
   -i 100 \
@@ -40,7 +40,7 @@ uv run skydiscover-run \
 Or from this directory:
 
 ```bash
-uv run skydiscover-run initial_program.py evaluator.py \
+uv run skydiscover-run initial_program.py eval \
   -c config.yaml \
   -s [your_algorithm] \
   -i 100
@@ -57,7 +57,7 @@ python evaluate_best_program.py
 | File | Description |
 |------|-------------|
 | `initial_program.py` | Baseline `rebalance_experts` function to evolve |
-| `evaluator.py` | Scores programs on load-balance quality and execution speed |
+| `eval/` | Containerized evaluator — scores programs on load-balance quality and execution speed |
 | `config.yaml` | Task-specific config (LLM, evaluator timeout, system prompt) |
 | `evaluate_best_program.py` | Standalone script to evaluate a saved best program |
 | `expert-load.json` | Workload data (must be downloaded — see Setup) |

benchmarks/ADRS/eplb/config.yaml

@@ -1,5 +1,5 @@
 # Expert Parallelism Load Balancer (EPLB) — MoE Expert Rearrangement
-# Usage: skydiscover-run initial_program.py evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.py eval -c config.yaml -s <strategy>
 # NOTE: Requires expert-load.json — see README.md for download instructions.
 language: python
 diff_based_generation: true

benchmarks/ADRS/llm_sql/README.md

@@ -31,7 +31,7 @@ From the repo root:
 ```bash
 uv run skydiscover-run \
   benchmarks/ADRS/llm_sql/initial_program.py \
-  benchmarks/ADRS/llm_sql/evaluator.py \
+  benchmarks/ADRS/llm_sql/eval \
   -c benchmarks/ADRS/llm_sql/config.yaml \
   -s [your_algorithm] \
   -i 100
@@ -49,7 +49,7 @@ Combined score: `0.95 * average_hit_rate + 0.05 * (12 - min(12, avg_runtime)) /
 | File | Description |
 |------|-------------|
 | `initial_program.py` | Baseline `Evolved` class with `reorder()` method to evolve |
-| `evaluator.py` | Scores programs on prefix hit rate and runtime across 5 datasets |
+| `eval/` | Containerized evaluator — scores programs on prefix hit rate and runtime across 5 datasets |
 | `config.yaml` | Task-specific config (LLM, evaluator timeout, system prompt) |
 | `solver.py` | Base `Algorithm` class and greedy baseline |
 | `utils.py` | Prefix hit count evaluation utilities |

benchmarks/ADRS/llm_sql/config.yaml

@@ -1,5 +1,5 @@
 # LLM SQL — Prompt Caching Column Reordering Optimization
-# Usage: skydiscover-run initial_program.py evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.py eval -c config.yaml -s <strategy>
 language: python
 diff_based_generation: true
 max_iterations: 100

benchmarks/ADRS/prism/config.yaml

@@ -1,5 +1,5 @@
 # Prism (GPU Model Placement) — Prompt Caching Column Reordering Optimization
-# Usage: skydiscover-run initial_program.py evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.py eval -c config.yaml -s <strategy>
 language: python
 diff_based_generation: true
 max_iterations: 100

benchmarks/ADRS/txn_scheduling/config.yaml

@@ -1,5 +1,5 @@
 # Transaction Scheduling — Minimize makespan for database workloads
-# Usage: skydiscover-run initial_program.py evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.py eval -c config.yaml -s <strategy>
 language: python
 diff_based_generation: true
 max_iterations: 100

benchmarks/README.md

@@ -31,13 +31,13 @@ export OPENAI_API_KEY="..."
 
 # Containerized benchmark (recommended — evaluator runs in Docker)
 uv run skydiscover-run benchmarks/math/circle_packing_rect/initial_program.py \
-  benchmarks/math/circle_packing_rect/evaluator \
+  benchmarks/math/circle_packing_rect/eval \
   -c benchmarks/math/circle_packing_rect/config.yaml \
   -s best_of_n -i 50
 
 # Plain Python evaluator (runs on host)
 uv run skydiscover-run benchmarks/math/circle_packing/initial_program.py \
-  benchmarks/math/circle_packing/evaluator.py \
+  benchmarks/math/circle_packing/eval \
   -c benchmarks/math/circle_packing/config.yaml \
   -s best_of_n -i 100
 ```
@@ -67,15 +67,15 @@ There are two ways to set up a benchmark: a **containerized evaluator** (recomme
 <task>/
 ├── initial_program.py       # Starting solution
 ├── config.yaml              # System prompt + search/evaluator settings
-└── evaluator/               # Self-contained Docker benchmark
+└── eval/                    # Self-contained Docker benchmark
     ├── Dockerfile
     ├── evaluate.sh          # Entrypoint (receives solution path + mode)
     ├── evaluator.py         # Scoring logic
     ├── requirements.txt     # Python dependencies
     └── ...                  # Any other data/files the evaluator needs
 ```
 
-The `evaluator/` directory is the Docker build context. Everything inside it gets copied into the image — data files, model weights, test fixtures, etc. SkyDiscover auto-detects this layout when `evaluation_file` points to a directory containing a `Dockerfile` and `evaluate.sh`.
+The `eval/` directory is the Docker build context. Everything inside it gets copied into the image — data files, model weights, test fixtures, etc. SkyDiscover auto-detects this layout when `evaluation_file` points to a directory containing a `Dockerfile` and `evaluate.sh`.
 
 ### Plain Python evaluator
 
@@ -154,7 +154,7 @@ ENTRYPOINT ["./evaluate.sh"]
 
 If you have an existing `evaluate(program_path) -> dict` function, you can wrap it with the backwards-compatibility wrapper:
 
-1. Copy `skydiscover/evaluation/wrapper.py` into your `evaluator/` directory.
+1. Copy `skydiscover/evaluation/wrapper.py` into your `eval/` directory.
 2. Add this to the bottom of your `evaluator.py`:
 
 ```python
@@ -167,11 +167,11 @@ The wrapper handles stdout redirection (so debug prints don't corrupt JSON), err
 
 #### Running a containerized benchmark
 
-Point `evaluation_file` at the `evaluator/` directory:
+Point `evaluation_file` at the `eval/` directory:
 
 ```bash
 skydiscover-run benchmarks/math/circle_packing_rect/initial_program.py \
-  benchmarks/math/circle_packing_rect/evaluator \
+  benchmarks/math/circle_packing_rect/eval \
   -c benchmarks/math/circle_packing_rect/config.yaml \
   -s best_of_n -i 50
 ```

benchmarks/ale_bench/README.md

@@ -24,7 +24,7 @@ Run evolution on a single problem:
 ```bash
 uv run skydiscover-run \
   benchmarks/ale_bench/ale-bench-lite-problems/ahc025/initial_program.cpp \
-  benchmarks/ale_bench/ale-bench-lite-problems/ahc025/evaluator.py \
+  benchmarks/ale_bench/ale-bench-lite-problems/ahc025/eval \
   -c benchmarks/ale_bench/ale-bench-lite-problems/ahc025/config.yaml \
   --search evox \
   -i 100
@@ -59,7 +59,7 @@ ale_bench/
 ├── ale-bench-lite-problems/
 │   └── ahcXXX/
 │       ├── initial_program.cpp   # Starting C++ solution
-│       ├── evaluator.py          # Runs 50 public cases via ale_bench
+│       ├── eval/             # Containerized evaluator (runs 50 public cases via ale_bench)
 │       └── config.yaml           # Search config (cpp, diff-based, 100 iterations)
 ├── ale_agent_best/
 │   └── ahcXXX.cpp               # Best known solutions (reference)

benchmarks/ale_bench/ale-bench-lite-problems/ahc008/config.yaml

@@ -1,5 +1,5 @@
 # ALE-Bench ahc008 — AtCoder Heuristic Contest
-# Usage: skydiscover-run initial_program.cpp evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.cpp eval -c config.yaml -s <strategy>
 language: cpp
 diff_based_generation: true
 max_iterations: 100

benchmarks/ale_bench/ale-bench-lite-problems/ahc008/eval/Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.12-slim
+
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /benchmark
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+RUN chmod +x evaluate.sh
+ENTRYPOINT ["./evaluate.sh"]

benchmarks/ale_bench/ale-bench-lite-problems/ahc008/eval/evaluate.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+set -euo pipefail
+python /benchmark/evaluator.py "$1"

benchmarks/ale_bench/ale-bench-lite-problems/ahc008/evaluator.py → benchmarks/ale_bench/ale-bench-lite-problems/ahc008/eval/evaluator.py

@@ -62,4 +62,7 @@ def evaluate(program_path):
         return {
             "overall_score": 0.0,
             "error": str(e),
-        }
+        }
+if __name__ == "__main__":
+    from wrapper import run
+    run(evaluate)

benchmarks/ale_bench/ale-bench-lite-problems/ahc008/eval/requirements.txt

@@ -0,0 +1,2 @@
+ale-bench
+ale-bench-eval

benchmarks/ale_bench/ale-bench-lite-problems/ahc011/config.yaml

@@ -1,5 +1,5 @@
 # ALE-Bench ahc011 — AtCoder Heuristic Contest
-# Usage: skydiscover-run initial_program.cpp evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.cpp eval -c config.yaml -s <strategy>
 language: cpp
 diff_based_generation: true
 max_iterations: 100

benchmarks/ale_bench/ale-bench-lite-problems/ahc011/eval/Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.12-slim
+
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /benchmark
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+RUN chmod +x evaluate.sh
+ENTRYPOINT ["./evaluate.sh"]

benchmarks/ale_bench/ale-bench-lite-problems/ahc011/eval/evaluate.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+set -euo pipefail
+python /benchmark/evaluator.py "$1"

benchmarks/ale_bench/ale-bench-lite-problems/ahc011/evaluator.py → benchmarks/ale_bench/ale-bench-lite-problems/ahc011/eval/evaluator.py

@@ -62,4 +62,7 @@ def evaluate(program_path):
         return {
             "overall_score": 0.0,
             "error": str(e),
-        }
+        }
+if __name__ == "__main__":
+    from wrapper import run
+    run(evaluate)

benchmarks/ale_bench/ale-bench-lite-problems/ahc011/eval/requirements.txt

@@ -0,0 +1,2 @@
+ale-bench
+ale-bench-eval

benchmarks/ale_bench/ale-bench-lite-problems/ahc015/config.yaml

@@ -1,5 +1,5 @@
 # ALE-Bench ahc015 — AtCoder Heuristic Contest
-# Usage: skydiscover-run initial_program.cpp evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.cpp eval -c config.yaml -s <strategy>
 language: cpp
 diff_based_generation: true
 max_iterations: 100

benchmarks/ale_bench/ale-bench-lite-problems/ahc015/eval/Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.12-slim
+
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /benchmark
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+RUN chmod +x evaluate.sh
+ENTRYPOINT ["./evaluate.sh"]

benchmarks/ale_bench/ale-bench-lite-problems/ahc015/eval/evaluate.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+set -euo pipefail
+python /benchmark/evaluator.py "$1"

benchmarks/ale_bench/ale-bench-lite-problems/ahc015/evaluator.py → benchmarks/ale_bench/ale-bench-lite-problems/ahc015/eval/evaluator.py

@@ -62,4 +62,7 @@ def evaluate(program_path):
         return {
             "overall_score": 0.0,
             "error": str(e),
-        }
+        }
+if __name__ == "__main__":
+    from wrapper import run
+    run(evaluate)

benchmarks/ale_bench/ale-bench-lite-problems/ahc015/eval/requirements.txt

@@ -0,0 +1,2 @@
+ale-bench
+ale-bench-eval

benchmarks/ale_bench/ale-bench-lite-problems/ahc016/config.yaml

@@ -1,5 +1,5 @@
 # ALE-Bench ahc016 — AtCoder Heuristic Contest
-# Usage: skydiscover-run initial_program.cpp evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.cpp eval -c config.yaml -s <strategy>
 language: cpp
 diff_based_generation: true
 max_iterations: 100

benchmarks/ale_bench/ale-bench-lite-problems/ahc016/eval/Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.12-slim
+
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /benchmark
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+RUN chmod +x evaluate.sh
+ENTRYPOINT ["./evaluate.sh"]

benchmarks/ale_bench/ale-bench-lite-problems/ahc016/eval/evaluate.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+set -euo pipefail
+python /benchmark/evaluator.py "$1"

benchmarks/ale_bench/ale-bench-lite-problems/ahc016/evaluator.py → benchmarks/ale_bench/ale-bench-lite-problems/ahc016/eval/evaluator.py

@@ -62,4 +62,7 @@ def evaluate(program_path):
         return {
             "overall_score": 0.0,
             "error": str(e),
-        }
+        }
+if __name__ == "__main__":
+    from wrapper import run
+    run(evaluate)

benchmarks/ale_bench/ale-bench-lite-problems/ahc016/eval/requirements.txt

@@ -0,0 +1,2 @@
+ale-bench
+ale-bench-eval

benchmarks/ale_bench/ale-bench-lite-problems/ahc024/config.yaml

@@ -1,5 +1,5 @@
 # ALE-Bench ahc024 — AtCoder Heuristic Contest
-# Usage: skydiscover-run initial_program.cpp evaluator.py -c config.yaml -s <strategy>
+# Usage: skydiscover-run initial_program.cpp eval -c config.yaml -s <strategy>
 language: cpp
 diff_based_generation: true
 max_iterations: 100

benchmarks/ale_bench/ale-bench-lite-problems/ahc024/eval/Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.12-slim
+
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /benchmark
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+RUN chmod +x evaluate.sh
+ENTRYPOINT ["./evaluate.sh"]

benchmarks/ale_bench/ale-bench-lite-problems/ahc024/eval/evaluate.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+set -euo pipefail
+python /benchmark/evaluator.py "$1"

benchmarks/ale_bench/ale-bench-lite-problems/ahc024/evaluator.py → benchmarks/ale_bench/ale-bench-lite-problems/ahc024/eval/evaluator.py

@@ -62,4 +62,7 @@ def evaluate(program_path):
         return {
             "overall_score": 0.0,
             "error": str(e),
-        }
+        }
+if __name__ == "__main__":
+    from wrapper import run
+    run(evaluate)

benchmarks/ale_bench/ale-bench-lite-problems/ahc024/eval/requirements.txt

@@ -0,0 +1,2 @@
+ale-bench
+ale-bench-eval