evaluating-llms-harness
npx machina-cli add skill Orchestra-Research/AI-Research-SKILLs/lm-evaluation-harness --openclawlm-evaluation-harness - LLM Benchmarking
Quick start
lm-evaluation-harness evaluates LLMs across 60+ academic benchmarks using standardized prompts and metrics.
Installation:
pip install lm-eval
Evaluate any HuggingFace model:
lm_eval --model hf \
--model_args pretrained=meta-llama/Llama-2-7b-hf \
--tasks mmlu,gsm8k,hellaswag \
--device cuda:0 \
--batch_size 8
View available tasks:
lm_eval --tasks list
Common workflows
Workflow 1: Standard benchmark evaluation
Evaluate model on core benchmarks (MMLU, GSM8K, HumanEval).
Copy this checklist:
Benchmark Evaluation:
- [ ] Step 1: Choose benchmark suite
- [ ] Step 2: Configure model
- [ ] Step 3: Run evaluation
- [ ] Step 4: Analyze results
Step 1: Choose benchmark suite
Core reasoning benchmarks:
- MMLU (Massive Multitask Language Understanding) - 57 subjects, multiple choice
- GSM8K - Grade school math word problems
- HellaSwag - Common sense reasoning
- TruthfulQA - Truthfulness and factuality
- ARC (AI2 Reasoning Challenge) - Science questions
Code benchmarks:
- HumanEval - Python code generation (164 problems)
- MBPP (Mostly Basic Python Problems) - Python coding
Standard suite (recommended for model releases):
--tasks mmlu,gsm8k,hellaswag,truthfulqa,arc_challenge
Step 2: Configure model
HuggingFace model:
lm_eval --model hf \
--model_args pretrained=meta-llama/Llama-2-7b-hf,dtype=bfloat16 \
--tasks mmlu \
--device cuda:0 \
--batch_size auto # Auto-detect optimal batch size
Quantized model (4-bit/8-bit):
lm_eval --model hf \
--model_args pretrained=meta-llama/Llama-2-7b-hf,load_in_4bit=True \
--tasks mmlu \
--device cuda:0
Custom checkpoint:
lm_eval --model hf \
--model_args pretrained=/path/to/my-model,tokenizer=/path/to/tokenizer \
--tasks mmlu \
--device cuda:0
Step 3: Run evaluation
# Full MMLU evaluation (57 subjects)
lm_eval --model hf \
--model_args pretrained=meta-llama/Llama-2-7b-hf \
--tasks mmlu \
--num_fewshot 5 \ # 5-shot evaluation (standard)
--batch_size 8 \
--output_path results/ \
--log_samples # Save individual predictions
# Multiple benchmarks at once
lm_eval --model hf \
--model_args pretrained=meta-llama/Llama-2-7b-hf \
--tasks mmlu,gsm8k,hellaswag,truthfulqa,arc_challenge \
--num_fewshot 5 \
--batch_size 8 \
--output_path results/llama2-7b-eval.json
Step 4: Analyze results
Results saved to results/llama2-7b-eval.json:
{
"results": {
"mmlu": {
"acc": 0.459,
"acc_stderr": 0.004
},
"gsm8k": {
"exact_match": 0.142,
"exact_match_stderr": 0.006
},
"hellaswag": {
"acc_norm": 0.765,
"acc_norm_stderr": 0.004
}
},
"config": {
"model": "hf",
"model_args": "pretrained=meta-llama/Llama-2-7b-hf",
"num_fewshot": 5
}
}
Workflow 2: Track training progress
Evaluate checkpoints during training.
Training Progress Tracking:
- [ ] Step 1: Set up periodic evaluation
- [ ] Step 2: Choose quick benchmarks
- [ ] Step 3: Automate evaluation
- [ ] Step 4: Plot learning curves
Step 1: Set up periodic evaluation
Evaluate every N training steps:
#!/bin/bash
# eval_checkpoint.sh
CHECKPOINT_DIR=$1
STEP=$2
lm_eval --model hf \
--model_args pretrained=$CHECKPOINT_DIR/checkpoint-$STEP \
--tasks gsm8k,hellaswag \
--num_fewshot 0 \ # 0-shot for speed
--batch_size 16 \
--output_path results/step-$STEP.json
Step 2: Choose quick benchmarks
Fast benchmarks for frequent evaluation:
- HellaSwag: ~10 minutes on 1 GPU
- GSM8K: ~5 minutes
- PIQA: ~2 minutes
Avoid for frequent eval (too slow):
- MMLU: ~2 hours (57 subjects)
- HumanEval: Requires code execution
Step 3: Automate evaluation
Integrate with training script:
# In training loop
if step % eval_interval == 0:
model.save_pretrained(f"checkpoints/step-{step}")
# Run evaluation
os.system(f"./eval_checkpoint.sh checkpoints step-{step}")
Or use PyTorch Lightning callbacks:
from pytorch_lightning import Callback
class EvalHarnessCallback(Callback):
def on_validation_epoch_end(self, trainer, pl_module):
step = trainer.global_step
checkpoint_path = f"checkpoints/step-{step}"
# Save checkpoint
trainer.save_checkpoint(checkpoint_path)
# Run lm-eval
os.system(f"lm_eval --model hf --model_args pretrained={checkpoint_path} ...")
Step 4: Plot learning curves
import json
import matplotlib.pyplot as plt
# Load all results
steps = []
mmlu_scores = []
for file in sorted(glob.glob("results/step-*.json")):
with open(file) as f:
data = json.load(f)
step = int(file.split("-")[1].split(".")[0])
steps.append(step)
mmlu_scores.append(data["results"]["mmlu"]["acc"])
# Plot
plt.plot(steps, mmlu_scores)
plt.xlabel("Training Step")
plt.ylabel("MMLU Accuracy")
plt.title("Training Progress")
plt.savefig("training_curve.png")
Workflow 3: Compare multiple models
Benchmark suite for model comparison.
Model Comparison:
- [ ] Step 1: Define model list
- [ ] Step 2: Run evaluations
- [ ] Step 3: Generate comparison table
Step 1: Define model list
# models.txt
meta-llama/Llama-2-7b-hf
meta-llama/Llama-2-13b-hf
mistralai/Mistral-7B-v0.1
microsoft/phi-2
Step 2: Run evaluations
#!/bin/bash
# eval_all_models.sh
TASKS="mmlu,gsm8k,hellaswag,truthfulqa"
while read model; do
echo "Evaluating $model"
# Extract model name for output file
model_name=$(echo $model | sed 's/\//-/g')
lm_eval --model hf \
--model_args pretrained=$model,dtype=bfloat16 \
--tasks $TASKS \
--num_fewshot 5 \
--batch_size auto \
--output_path results/$model_name.json
done < models.txt
Step 3: Generate comparison table
import json
import pandas as pd
models = [
"meta-llama-Llama-2-7b-hf",
"meta-llama-Llama-2-13b-hf",
"mistralai-Mistral-7B-v0.1",
"microsoft-phi-2"
]
tasks = ["mmlu", "gsm8k", "hellaswag", "truthfulqa"]
results = []
for model in models:
with open(f"results/{model}.json") as f:
data = json.load(f)
row = {"Model": model.replace("-", "/")}
for task in tasks:
# Get primary metric for each task
metrics = data["results"][task]
if "acc" in metrics:
row[task.upper()] = f"{metrics['acc']:.3f}"
elif "exact_match" in metrics:
row[task.upper()] = f"{metrics['exact_match']:.3f}"
results.append(row)
df = pd.DataFrame(results)
print(df.to_markdown(index=False))
Output:
| Model | MMLU | GSM8K | HELLASWAG | TRUTHFULQA |
|------------------------|-------|-------|-----------|------------|
| meta-llama/Llama-2-7b | 0.459 | 0.142 | 0.765 | 0.391 |
| meta-llama/Llama-2-13b | 0.549 | 0.287 | 0.801 | 0.430 |
| mistralai/Mistral-7B | 0.626 | 0.395 | 0.812 | 0.428 |
| microsoft/phi-2 | 0.560 | 0.613 | 0.682 | 0.447 |
Workflow 4: Evaluate with vLLM (faster inference)
Use vLLM backend for 5-10x faster evaluation.
vLLM Evaluation:
- [ ] Step 1: Install vLLM
- [ ] Step 2: Configure vLLM backend
- [ ] Step 3: Run evaluation
Step 1: Install vLLM
pip install vllm
Step 2: Configure vLLM backend
lm_eval --model vllm \
--model_args pretrained=meta-llama/Llama-2-7b-hf,tensor_parallel_size=1,dtype=auto,gpu_memory_utilization=0.8 \
--tasks mmlu \
--batch_size auto
Step 3: Run evaluation
vLLM is 5-10× faster than standard HuggingFace:
# Standard HF: ~2 hours for MMLU on 7B model
lm_eval --model hf \
--model_args pretrained=meta-llama/Llama-2-7b-hf \
--tasks mmlu \
--batch_size 8
# vLLM: ~15-20 minutes for MMLU on 7B model
lm_eval --model vllm \
--model_args pretrained=meta-llama/Llama-2-7b-hf,tensor_parallel_size=2 \
--tasks mmlu \
--batch_size auto
When to use vs alternatives
Use lm-evaluation-harness when:
- Benchmarking models for academic papers
- Comparing model quality across standard tasks
- Tracking training progress
- Reporting standardized metrics (everyone uses same prompts)
- Need reproducible evaluation
Use alternatives instead:
- HELM (Stanford): Broader evaluation (fairness, efficiency, calibration)
- AlpacaEval: Instruction-following evaluation with LLM judges
- MT-Bench: Conversational multi-turn evaluation
- Custom scripts: Domain-specific evaluation
Common issues
Issue: Evaluation too slow
Use vLLM backend:
lm_eval --model vllm \
--model_args pretrained=model-name,tensor_parallel_size=2
Or reduce fewshot examples:
--num_fewshot 0 # Instead of 5
Or evaluate subset of MMLU:
--tasks mmlu_stem # Only STEM subjects
Issue: Out of memory
Reduce batch size:
--batch_size 1 # Or --batch_size auto
Use quantization:
--model_args pretrained=model-name,load_in_8bit=True
Enable CPU offloading:
--model_args pretrained=model-name,device_map=auto,offload_folder=offload
Issue: Different results than reported
Check fewshot count:
--num_fewshot 5 # Most papers use 5-shot
Check exact task name:
--tasks mmlu # Not mmlu_direct or mmlu_fewshot
Verify model and tokenizer match:
--model_args pretrained=model-name,tokenizer=same-model-name
Issue: HumanEval not executing code
Install execution dependencies:
pip install human-eval
Enable code execution:
lm_eval --model hf \
--model_args pretrained=model-name \
--tasks humaneval \
--allow_code_execution # Required for HumanEval
Advanced topics
Benchmark descriptions: See references/benchmark-guide.md for detailed description of all 60+ tasks, what they measure, and interpretation.
Custom tasks: See references/custom-tasks.md for creating domain-specific evaluation tasks.
API evaluation: See references/api-evaluation.md for evaluating OpenAI, Anthropic, and other API models.
Multi-GPU strategies: See references/distributed-eval.md for data parallel and tensor parallel evaluation.
Hardware requirements
- GPU: NVIDIA (CUDA 11.8+), works on CPU (very slow)
- VRAM:
- 7B model: 16GB (bf16) or 8GB (8-bit)
- 13B model: 28GB (bf16) or 14GB (8-bit)
- 70B model: Requires multi-GPU or quantization
- Time (7B model, single A100):
- HellaSwag: 10 minutes
- GSM8K: 5 minutes
- MMLU (full): 2 hours
- HumanEval: 20 minutes
Resources
- GitHub: https://github.com/EleutherAI/lm-evaluation-harness
- Docs: https://github.com/EleutherAI/lm-evaluation-harness/tree/main/docs
- Task library: 60+ tasks including MMLU, GSM8K, HumanEval, TruthfulQA, HellaSwag, ARC, WinoGrande, etc.
- Leaderboard: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard (uses this harness)
Source
git clone https://github.com/Orchestra-Research/AI-Research-SKILLs/blob/main/11-evaluation/lm-evaluation-harness/SKILL.mdView on GitHub Overview
Evaluates LLMs using standardized prompts and metrics across 60+ benchmarks (e.g., MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). This makes it easy to benchmark model quality, compare models, report academic results, and track training progress. It is the industry standard used by EleutherAI, HuggingFace, and major labs, with support for HuggingFace, vLLM, and APIs.
How This Skill Works
Install the harness (pip install lm-eval), configure a model backend (e.g., HuggingFace HF models or other backends), select tasks (mmlu, gsm8k, hellaswag, etc.), run the evaluation to produce per-task metrics and a combined result JSON, and review config metadata. Results include per-task scores and a saved config block to help reproduce experiments.
When to Use It
- Benchmark model quality on standard suites
- Compare multiple models side-by-side in a single run
- Publish or report academic results with per-task metrics
- Track training progress via periodic evaluations
- Validate performance after fine-tuning or architecture changes
Quick Start
- Step 1: Install lm-eval (pip install lm-eval)
- Step 2: Run a core benchmark on a HuggingFace model using the example: lm_eval --model hf --model_args pretrained=MODEL --tasks mmlu,gsm8k,hellaswag --device cuda:0 --batch_size 8
- Step 3: View and analyze results saved in a JSON file or results directory
Best Practices
- Use the Standard suite for model releases: mmlu, gsm8k, hellaswag, truthfulqa, arc_challenge
- Auto-detect batch size when possible and choose appropriate few-shot settings
- Record and share the run configuration (model_args, tasks, device) for reproducibility
- Save full results JSON and summaries to facilitate reporting and paper drafts
- Run multiple seeds or task subsets to assess stability and variance
Example Use Cases
- Benchmarking a HuggingFace model (e.g., Llama-2-7b-hf) across mmlu, gsm8k, hellaswag to compare with peers
- Tracking training progress by evaluating checkpoints at regular intervals
- Publishing a model-paper with per-task accuracies and confidence intervals
- Generating and sharing a results JSON file like results/llama2-7b-eval.json
- Cross-lab benchmarking to validate industry-standard metrics across models
Frequently Asked Questions
Related Skills
phoenix-observability
Orchestra-Research/AI-Research-SKILLs
Open-source AI observability platform for LLM tracing, evaluation, and monitoring. Use when debugging LLM applications with detailed traces, running evaluations on datasets, or monitoring production AI systems with real-time insights.
evaluating-code-models
Orchestra-Research/AI-Research-SKILLs
Evaluates code generation models across HumanEval, MBPP, MultiPL-E, and 15+ benchmarks with pass@k metrics. Use when benchmarking code models, comparing coding abilities, testing multi-language support, or measuring code generation quality. Industry standard from BigCode Project used by HuggingFace leaderboards.
langsmith-observability
Orchestra-Research/AI-Research-SKILLs
LLM observability platform for tracing, evaluation, and monitoring. Use when debugging LLM applications, evaluating model outputs against datasets, monitoring production systems, or building systematic testing pipelines for AI applications.
nemo-evaluator-sdk
Orchestra-Research/AI-Research-SKILLs
Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.