nvidia-vlmeval 26.3

OpenCompass VLM Evaluation Kit - packaged by NVIDIA

NVIDIA NeMo Evaluator

The goal of NVIDIA NeMo Evaluator is to advance and refine state-of-the-art methodologies for model evaluation, and deliver them as modular evaluation packages (evaluation containers and pip wheels) that teams can use as standardized building blocks.

Quick start guide

NVIDIA NeMo Evaluator provide you with evaluation clients, that are specifically built to evaluate model endpoints using our Standard API.

Launching an evaluation for an LLM

1. Install the package

   pip install nvidia-vlmeval
   

2. (Optional) Set a token to your API endpoint if it's protected

   export MY_API_KEY="your_api_key_here"
   

3. List the available evaluations:

   $ nemo-evaluator ls
   Available tasks:
   * ai2d_judge (in vlmevalkit)
   * chartqa (in vlmevalkit)
   * mathvista-mini (in vlmevalkit)
   * mmmu_judge (in vlmevalkit)
   * ocrbench (in vlmevalkit)
   * slidevqa (in vlmevalkit)
   ...
   

4. Run the evaluation of your choice:

   nemo-evaluator run_eval \
       --eval_type ocrbench \
       --model_id microsoft/phi-4-multimodal-instruct \
       --model_url https://integrate.api.nvidia.com/v1/chat/completions \
       --model_type chat \
       --api_key_name MY_API_KEY \
       --output_dir /workspace/results
   

5. Gather the results

   cat /workspace/results/results.yml
   

Command-Line Tool

Each package comes pre-installed with a set of command-line tools, designed to simplify the execution of evaluation tasks. Below are the available commands and their usage for the vlmevalkit:

Commands

1. List Evaluation Types

nemo-evaluator ls

Displays the evaluation types available within the harness.

2. Run an evaluation

The nemo-evaluator run_eval command executes the evaluation process. Below are the flags and their descriptions:

Required flags

• --eval_type <string> The type of evaluation to perform
• --model_id <string> The name or identifier of the model to evaluate.
• --model_url <url> The API endpoint where the model is accessible.
• --model_type <string> The type of the model to evaluate, currently either "chat" or "completions".
• --output_dir <directory> The directory to use as the working directory for the evaluation. The results, including the results.yml output file, will be saved here.

Optional flags

• --api_key_name <string> The name of the environment variable that stores the Bearer token for the API, if authentication is required.
• --run_config <path> Specifies the path to a YAML file containing the evaluation definition.

Example

nemo-evaluator run_eval \
    --eval_type ocrbench \
    --model_id my_model \
    --model_type chat \
    --model_url http://localhost:8000/v1/chat/completions \
    --output_dir ./evaluation_results

If the model API requires authentication, set the API key in an environment variable and reference it using the --api_key_name flag:

export MY_API_KEY="your_api_key_here"

nemo-evaluator run_eval \
    --eval_type ocrbench \
    --model_id my_model \
    --model_type chat \
    --model_url http://localhost:8000/v1/chat/completions \
    --api_key_name MY_API_KEY \
    --output_dir ./evaluation_results

Configuring evaluations via YAML

Evaluations in NVIDIA NeMo Evaluator are configured using YAML files that define the parameters and settings required for the evaluation process. These configuration files follow a standard API which ensures consistency across evaluations.

Example of a YAML config:

config:
  type: ocrbench
  params:
    parallelism: 50
    limit_samples: 20
target:
  api_endpoint:
    model_id: microsoft/phi-4-multimodal-instruct
    type: chat
    url: https://integrate.api.nvidia.com/v1/chat/completions
    api_key_name: NVIDIA_API_KEY

The priority of overrides is as follows:

1. command line arguments
2. user config (as seen above)
3. task defaults (defined per task type)
4. framework defaults

--dry_run option allows you to print the final run configuration and command without executing the evaluation.

Example:

nemo-evaluator run_eval \
    --eval_type ocrbench \
    --model_id my_model \
    --model_type chat \
    --model_url http://localhost:8000/v1/chat/completions \
    --output_dir .evaluation_results \
    --dry_run

Output:

Rendered config:

command: "cat > {{config.output_dir}}/vlmeval_config.json << 'EOF'\n{\n  \"model\"\
  : {\n    \"{{target.api_endpoint.model_id.split('/')[-1]}}\": {\n      \"class\"\
  : \"CustomOAIEndpoint\",\n      \"model\": \"{{target.api_endpoint.model_id}}\"\
  ,\n      \"api_base\": \"{{target.api_endpoint.url}}\",\n      \"api_key_var_name\"\
  : \"{{target.api_endpoint.api_key_name}}\",\n      \"max_tokens\": {{config.params.max_new_tokens}},\n\
  \      \"temperature\": {{config.params.temperature}},{% if config.params.top_p\
  \ is not none %}\n      \"top_p\": {{config.params.top_p}},{% endif %}\n      \"\
  retry\": {{config.params.max_retries}},\n      \"timeout\": {{config.params.request_timeout}}{%\
  \ if config.params.extra.wait is defined %},\n      \"wait\": {{config.params.extra.wait}}{%\
  \ endif %}{% if config.params.extra.img_size is defined %},\n      \"img_size\"\
  : {{config.params.extra.img_size}}{% endif %}{% if config.params.extra.img_detail\
  \ is defined %},\n      \"img_detail\": \"{{config.params.extra.img_detail}}\"{%\
  \ endif %}{% if config.params.extra.system_prompt is defined %},\n      \"system_prompt\"\
  : \"{{config.params.extra.system_prompt}}\"{% endif %}{% if config.params.extra.verbose\
  \ is defined %},\n      \"verbose\": {{config.params.extra.verbose}}{% endif %}\n\
  \    }\n  },\n  \"data\": {\n    \"{{config.params.extra.dataset.name}}\": {\n \
  \     \"class\": \"{{config.params.extra.dataset.class}}\",\n      \"dataset\":\
  \ \"{{config.params.extra.dataset.name}}\",\n      \"model\": \"{{target.api_endpoint.model_id}}\"\
  \n    }\n  }\n}\nEOF\npython -m vlmeval.run \\\n  --config {{config.output_dir}}/vlmeval_config.json\
  \ \\\n  --work-dir {{config.output_dir}} \\\n  --api-nproc {{config.params.parallelism}}\
  \ \\\n  {%- if config.params.extra.judge is defined %}\n  --judge {{config.params.extra.judge.model}}\
  \ \\\n  --judge-args '{{config.params.extra.judge.args}}' \\\n  {%- endif %}\n \
  \ {% if config.params.limit_samples is not none %}--first-n {{config.params.limit_samples}}{%\
  \ endif %}\n"
framework_name: vlmevalkit
pkg_name: vlmeval
config:
  output_dir: .evaluation_results
  params:
    limit_samples: null
    max_new_tokens: 2048
    max_retries: 5
    parallelism: 4
    task: null
    temperature: 0.0
    request_timeout: 60
    top_p: null
    extra:
      dataset:
        name: OCRBench
        class: OCRBench
  supported_endpoint_types:
  - chat
  type: ocrbench
target:
  api_endpoint:
    api_key_name: null
    model_id: my_model
    stream: null
    type: chat
    url: http://localhost:8000/v1/chat/completions


Rendered command:

cat > .evaluation_results/vlmeval_config.json << 'EOF'
{
  "model": {
    "my_model": {
      "class": "CustomOAIEndpoint",
      "model": "my_model",
      "api_base": "http://localhost:8000/v1/chat/completions",
      "api_key_var_name": "None",
      "max_tokens": 2048,
      "temperature": 0.0,
      "retry": 5,
      "timeout": 60
    }
  },
  "data": {
    "OCRBench": {
      "class": "OCRBench",
      "dataset": "OCRBench",
      "model": "my_model"
    }
  }
}
EOF
python -m vlmeval.run \
  --config .evaluation_results/vlmeval_config.json \
  --work-dir .evaluation_results \
  --api-nproc 4 \

FAQ

Deploying a model as an endpoint

NVIDIA NeMo Evaluator utilize a client-server communication architecture to interact with the model. As a prerequisite, the model must be deployed as an endpoint with a NIM-compatible API.

Users have the flexibility to deploy their model using their own infrastructure and tooling.

Servers with APIs that conform to the OpenAI/NIM API standard are expected to work seamlessly out of the box.