Llama-Recipe make use of lm-evaluation-harness for evaluating our fine-tuned Llama2 model. It also can serve as a tool to evaluate quantized model to ensure the quality in lower precision or other optimization applied to the model that might need evaluation.

lm-evaluation-harness provide a wide range of features:

• Over 60 standard academic benchmarks for LLMs, with hundreds of subtasks and variants implemented.
• Support for models loaded via transformers (including quantization via AutoGPTQ), GPT-NeoX, and Megatron-DeepSpeed, with a flexible tokenization-agnostic interface.
• Support for fast and memory-efficient inference with vLLM.
• Support for commercial APIs including OpenAI, and TextSynth.
• Support for evaluation on adapters (e.g. LoRA) supported in Hugging Face's PEFT library.
• Support for local models and benchmarks.

The Language Model Evaluation Harness is also the backend for 🤗 Hugging Face's (HF) popular Open LLM Leaderboard.

Before running the evaluation script, ensure you have all the necessary dependencies installed.

• Python 3.8+
• Your language model's dependencies

Clone the lm-evaluation-harness repository and install it:

git clone https://github.com/matthoffner/lm-evaluation-harness.git
cd lm-evaluation-harness
pip install -e .

To run evaluation for Hugging Face Llama2 7B model on a single GPU please run the following,

python eval.py --model hf --model_args pretrained=meta-llama/Llama-2-7b-chat-hf --tasks hellaswag --device cuda:0   --batch_size 8

Tasks can be extended by using , between them for example --tasks hellaswag,arc.

To set the number of shots you can use --num_fewshot to set the number for few shot evaluation.

In case you have fine-tuned your model using PEFT you can set the PATH to the PEFT checkpoints using PEFT as part of model_args as shown below:

python eval.py --model hf --model_args pretrained=meta-llama/Llama-2-7b-hf,dtype="float",peft=../peft_output --tasks hellaswag --num_fewshot 10  --device cuda:0 --batch_size 8 

There has been an study from IBM on efficient benchmarking of LLMs, with main take a way that to identify if a model is performing poorly, benchmarking on wider range of tasks is more important than the number example in each task. This means you could run the evaluation harness with fewer number of example to have initial decision if the performance got worse from the base line. To limit the number of example here, it can be set using --limit flag with actual desired number. But for the full assessment you would need to run the full evaluation. Please read more in the paper linked above.

python eval.py --model hf --model_args pretrained=meta-llama/Llama-2-7b-hf,dtype="float",peft=../peft_output --tasks hellaswag --num_fewshot 10  --device cuda:0 --batch_size 8 --limit 100

Here, we provided a list of tasks from Open-LLM-Leaderboard which can be used by passing --open-llm-leaderboard-tasks instead of tasks to the eval.py.

NOTE Make sure to run the bash script below, that will set the include paths in the config files. The script will prompt you to enter the path to the cloned lm-evaluation-harness repo.You would need this step only for the first time.

bash open_llm_eval_prep.sh

Now we can run the eval benchmark:

python eval.py --model hf --model_args pretrained=meta-llama/Llama-2-7b-hf,dtype="float",peft=../peft_output --num_fewshot 10  --device cuda:0 --batch_size 8 --limit 100 --open_llm_leaderboard_tasks

In the HF leaderboard, the LLMs are evaluated on 7 benchmarks from Language Model Evaluation Harness as described below:

• AI2 Reasoning Challenge (25-shot) - a set of grade-school science questions.
• HellaSwag (10-shot) - a test of commonsense inference, which is easy for humans (~95%) but challenging for SOTA models.
• MMLU (5-shot) - a test to measure a text model's multitask accuracy. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more.
• TruthfulQA (0-shot) - a test to measure a model's propensity to reproduce falsehoods commonly found online. Note: TruthfulQA in the Harness is actually a minima a 6-shots task, as it is prepended by 6 examples systematically, even when launched using 0 for the number of few-shot examples.
• Winogrande (5-shot) - an adversarial and difficult Winograd benchmark at scale, for commonsense reasoning.
• GSM8k (5-shot) - diverse grade school math word problems to measure a model's ability to solve multi-step mathematical reasoning problems. For all these evaluations, a higher score is a better score. We chose these benchmarks as they test a variety of reasoning and general knowledge across a wide variety of fields in 0-shot and few-shot settings.

In case you have customized the Llama model, for example a quantized version of model where it has different model loading from normal HF model, you can follow this guide to add your model to the eval.py and run the eval benchmarks.

You can also find full task list here.

Hugging Face's accelerate 🚀 library can be used for multi-GPU evaluation as it is supported by lm-evaluation-harness.

To perform data-parallel evaluation (where each GPU loads a separate full copy of the model), leverage the accelerate launcher as follows:

accelerate config

accelerate launch eval.py --model hf --model_args "pretrained=meta-llama/Llama-2-7b-chat-hf" --limit 100 --open-llm-leaderboard-tasks --output_path ./results.json --log_samples 

In case your model can fit on a single GPU, this allows you to evaluate on K GPUs K times faster than on one.

WARNING: This setup does not work with FSDP model sharding, so in accelerate config FSDP must be disabled, or the NO_SHARD FSDP option must be used.

In case your model is too large to fit on a single GPU.

In this setting, run the library outside of the accelerate launcher, but passing parallelize=True to --model_args as follows:

python eval.py --model hf --model_args "pretrained=meta-llama/Llama-2-7b-chat-hf,parallelize=True" --limit 100 --open_llm_leaderboard_tasks --output_path ./results.json --log_samples 

This means that your model's weights will be split across all available GPUs.

For more advanced users or even larger models, we allow for the following arguments when parallelize=True as well:

• device_map_option: How to split model weights across available GPUs. defaults to "auto".
• max_memory_per_gpu: the max GPU memory to use per GPU in loading the model.
• max_cpu_memory: the max amount of CPU memory to use when offloading the model weights to RAM.
• offload_folder: a folder where model weights will be offloaded to disk if needed.

These two options (accelerate launch and parallelize=True) are mutually exclusive.

Also lm-evaluation-harness supports vLLM for faster inference on supported model types, especially faster when splitting a model across multiple GPUs. For single-GPU or multi-GPU — tensor parallel, data parallel, or a combination of both — inference, for example:

python eval.py --model vllm --model_args "pretrained=meta-llama/Llama-2-7b-chat-hf,tensor_parallel_size=1,dtype=auto,gpu_memory_utilization=0.8,data_parallel_size=2" --limit 100 --open_llm_leaderboard_tasks --output_path ./results.json --log_samples --batch_size auto

For a full list of supported vLLM configurations, please to here.

Note from lm-evaluation-harness vLLM occasionally differs in output from Hugging Face. We treat Hugging Face as the reference implementation, and provide a script for checking the validity of vllm results against HF.