src/content.py

LOGO = '<img src="https://huggingface.co/spaces/optimum/llm-perf-leaderboard/resolve/main/logo.png">'

TITLE = """<h1 align="center" id="space-title">🤗 LLM-Perf Leaderboard 🏋️</h1>"""

INTRODUCTION = """
The 🤗 LLM-Perf Leaderboard 🏋️ aims to benchmark the performance (latency, throughput, memory & energy) of Large Language Models (LLMs) with different hardwares, backends and optimizations using [Optimum-Benchmark](https://github.com/huggingface/optimum-benchmark) and [Optimum](https://github.com/huggingface/optimum) flavors.

Anyone from the community can request a model or a hardware/backend/optimization configuration for automated benchmarking:
- Model evaluation requests should be made in the [🤗 Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard) and will be added to the [🤗 LLM Performance Leaderboard 🏋️](https://huggingface.co/spaces/optimum/llm-perf-leaderboard) automatically.
- Hardware/Backend/Optimization performance requests should be made in the [community discussions](https://huggingface.co/spaces/optimum/llm-perf-leaderboard/discussions) to assess their relevance and feasibility.
"""

ABOUT = """<h3>About the 🤗 LLM-Perf Leaderboard 🏋️</h3>
<ul>
    <li>To avoid communication-dependent results, only one GPU is used.</li>
    <li>Score is the average evaluation score obtained from the <a href="https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard">🤗 Open LLM Leaderboard</a>.</li>
    <li>LLMs are running on a singleton batch with a prompt size of 256 and generating a 256 tokens.</li>
    <li>Energy consumption is measured in kWh using CodeCarbon and taking into consideration the GPU, CPU, RAM and location of the machine.</li>
    <li>We measure three types of memory: Max Allocated Memory, Max Reserved Memory and Max Used Memory. The first two being reported by PyTorch and the last one being observed using PyNVML.</li>
</ul>
"""

EXAMPLE_CONFIG = """
Here's an example of the configuration file used to benchmark the models with Optimum-Benchmark:
```yaml
defaults:
  - backend: pytorch
  - _base_ # inheriting from base config
  - _self_ # for hydra 1.1 compatibility

experiment_name: pytorch+cuda+float16+gptq-4bit+exllama-v1
device: cuda

backend:
  no_weights: true
  torch_dtype: float16
  quantization_scheme: gptq
  quantization_config:
    bits: 4
    use_cuda_fp16: false
    use_exllama: true
    exllama_config:
      version: 1
```

Where the base config is:
```yaml
defaults:
  - benchmark: inference # default benchmark
  - launcher: process # isolated process launcher
  - experiment # inheriting from experiment config
  - _self_ # for hydra 1.1 compatibility
  - override hydra/job_logging: colorlog # colorful logging
  - override hydra/hydra_logging: colorlog # colorful logging

hydra:
  run:
    dir: dataset/${oc.env:HOSTNAME}/${experiment_name}/${model}
  job:
    chdir: true
    env_set:
      COUNTRY_ISO_CODE: FRA
      OVERRIDE_BENCHMARKS: 0
      CUDA_VISIBLE_DEVICES: 0
      CUDA_DEVICE_ORDER: PCI_BUS_ID

backend:
  continuous_isolation: true

benchmark:
  duration: 10
  memory: true
  energy: true

  input_shapes:
    batch_size: 1
    sequence_length: 256

  new_tokens: 256

hub_kwargs:
  trust_remote_code: true
```
"""


CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results."
CITATION_BUTTON = r"""@misc{llm-perf-leaderboard,
  author = {Ilyas Moutawwakil, Régis Pierrard},
  title = {LLM-Perf Leaderboard},
  year = {2023},
  publisher = {Hugging Face},
  howpublished = "\url{https://huggingface.co/spaces/optimum/llm-perf-leaderboard}",
@software{optimum-benchmark,
  author = {Ilyas Moutawwakil, Régis Pierrard},
  publisher = {Hugging Face},
  title = {Optimum-Benchmark: A framework for benchmarking the performance of Transformers models with different hardwares, backends and optimizations.},
}
"""