olmocr/scripts/compare_vllm.sh
2025-08-06 16:31:08 +00:00

201 lines
6.1 KiB
Bash
Executable File

#!/bin/bash
# Compares VLM inference between vLLM and HuggingFace checkpoints
# Usage: ./scripts/compare_vllm.sh <model_path> [--max-tokens N] [--num-prompts N] [--prob-threshold F] [--seed N]
set -e
# Default values
DEFAULT_MAX_TOKENS=1000
DEFAULT_NUM_PROMPTS=100
DEFAULT_PROB_THRESHOLD=0.20
DEFAULT_SEED=42
# Parse arguments
if [ $# -lt 1 ]; then
echo "Usage: $0 <model_path> [--max-tokens N] [--num-prompts N] [--prob-threshold F] [--seed N]"
echo "Example: $0 Qwen/Qwen2.5-VL-7B-Instruct"
echo "Example: $0 s3://ai2-oe-data/jakep/olmocr/model --max-tokens 50 --num-prompts 200"
exit 1
fi
MODEL_PATH="$1"
MAX_TOKENS="$DEFAULT_MAX_TOKENS"
NUM_PROMPTS="$DEFAULT_NUM_PROMPTS"
PROB_THRESHOLD="$DEFAULT_PROB_THRESHOLD"
SEED="$DEFAULT_SEED"
# Parse optional arguments
shift 1
while [[ $# -gt 0 ]]; do
case $1 in
--max-tokens)
MAX_TOKENS="$2"
shift 2
;;
--num-prompts)
NUM_PROMPTS="$2"
shift 2
;;
--prob-threshold)
PROB_THRESHOLD="$2"
shift 2
;;
--seed)
SEED="$2"
shift 2
;;
*)
echo "Unknown option: $1"
exit 1
;;
esac
done
# Check for uncommitted changes
if ! git diff-index --quiet HEAD --; then
echo "Error: There are uncommitted changes in the repository."
echo "Please commit or stash your changes before running the comparison."
echo ""
echo "Uncommitted changes:"
git status --short
exit 1
fi
# Use conda environment Python if available, otherwise use system Python
if [ -n "$CONDA_PREFIX" ]; then
PYTHON="$CONDA_PREFIX/bin/python"
echo "Using conda Python from: $CONDA_PREFIX"
else
PYTHON="python"
echo "Warning: No conda environment detected, using system Python"
fi
# Get version from version.py
VERSION=$($PYTHON -c 'import olmocr.version; print(olmocr.version.VERSION)')
echo "OlmOCR version: $VERSION"
# Get first 10 characters of git hash
GIT_HASH=$(git rev-parse HEAD | cut -c1-10)
echo "Git hash: $GIT_HASH"
# Get current git branch name
GIT_BRANCH=$(git rev-parse --abbrev-ref HEAD)
echo "Git branch: $GIT_BRANCH"
# Create full image tag
IMAGE_TAG="olmocr-compare-vllm-${VERSION}-${GIT_HASH}"
echo "Building Docker image with tag: $IMAGE_TAG"
# Build the Docker image
echo "Building Docker image..."
docker build --platform linux/amd64 -f ./Dockerfile -t $IMAGE_TAG .
# Get Beaker username
BEAKER_USER=$(beaker account whoami --format json | jq -r '.[0].name')
echo "Beaker user: $BEAKER_USER"
# Push image to beaker
echo "Trying to push image to Beaker..."
if ! beaker image create --workspace ai2/oe-data-pdf --name $IMAGE_TAG $IMAGE_TAG 2>/dev/null; then
echo "Warning: Beaker image with tag $IMAGE_TAG already exists. Using existing image."
fi
# Create Python script to run beaker experiment
cat << 'EOF' > /tmp/run_compare_vllm_experiment.py
import sys
from beaker import Beaker, ExperimentSpec, TaskSpec, TaskContext, ResultSpec, TaskResources, ImageSource, Priority, Constraints, EnvVar, DataMount
# Get parameters from command line
image_tag = sys.argv[1]
beaker_user = sys.argv[2]
git_branch = sys.argv[3]
git_hash = sys.argv[4]
model_path = sys.argv[5]
max_tokens = sys.argv[6]
num_prompts = sys.argv[7]
prob_threshold = sys.argv[8]
seed = sys.argv[9]
# Initialize Beaker client
b = Beaker.from_env(default_workspace="ai2/olmocr")
# Check if AWS credentials secret exists
aws_creds_secret = f"{beaker_user}-AWS_CREDENTIALS_FILE"
try:
# Try to get the secret to see if it exists
b.secret.get(aws_creds_secret, workspace="ai2/olmocr")
has_aws_creds = True
print(f"Found AWS credentials secret: {aws_creds_secret}")
except:
has_aws_creds = False
print(f"AWS credentials secret not found: {aws_creds_secret}")
# Build commands for comparison job
commands = []
if has_aws_creds:
commands.extend([
"mkdir -p ~/.aws",
'echo "$AWS_CREDENTIALS_FILE" > ~/.aws/credentials'
])
commands.extend([
# Install accelerate
"pip install accelerate",
# Run comparison
f'python -m olmocr.train.compare_vllm_checkpoint --model {model_path} --max-tokens {max_tokens} --num-prompts {num_prompts} --prob-threshold {prob_threshold} --seed {seed}'
])
# Build task spec with optional env vars
task_spec_args = {
"name": "olmocr-compare-vllm",
"image": ImageSource(beaker=f"{beaker_user}/{image_tag}"),
"command": [
"bash", "-c",
" && ".join(commands)
],
"context": TaskContext(
priority=Priority.normal,
preemptible=True,
),
"resources": TaskResources(gpu_count=1),
"constraints": Constraints(cluster=["ai2/ceres-cirrascale", "ai2/jupiter-cirrascale-2"]),
"result": ResultSpec(path="/noop-results"),
"datasets": [
DataMount.new(mount_path="/weka/oe-data-default", weka="oe-data-default"),
DataMount.new(mount_path="/weka/oe-training-default", weka="oe-training-default"),
]
}
# Add env vars if AWS credentials exist
if has_aws_creds:
task_spec_args["env_vars"] = [
EnvVar(name="AWS_CREDENTIALS_FILE", secret=aws_creds_secret)
]
# Create experiment spec
experiment_spec = ExperimentSpec(
description=f"OlmOCR vLLM vs HF Comparison - Branch: {git_branch}, Commit: {git_hash}, Model: {model_path}",
budget="ai2/oe-base",
tasks=[TaskSpec(**task_spec_args)],
)
# Create the experiment
experiment = b.experiment.create(spec=experiment_spec, workspace="ai2/olmocr")
print(f"Created comparison experiment: {experiment.id}")
print(f"View at: https://beaker.org/ex/{experiment.id}")
EOF
# Run the Python script to create the experiment
echo "Creating Beaker experiment..."
echo "Comparing model: $MODEL_PATH"
echo "Max tokens: $MAX_TOKENS"
echo "Number of prompts: $NUM_PROMPTS"
echo "Probability threshold: $PROB_THRESHOLD"
echo "Random seed: $SEED"
$PYTHON /tmp/run_compare_vllm_experiment.py $IMAGE_TAG $BEAKER_USER $GIT_BRANCH $GIT_HASH "$MODEL_PATH" "$MAX_TOKENS" "$NUM_PROMPTS" "$PROB_THRESHOLD" "$SEED"
# Clean up temporary file
rm /tmp/run_compare_vllm_experiment.py
echo "Comparison experiment submitted successfully!"