Add models to benchmarks#40820
Open
ahadnagy wants to merge 4 commits into
Open
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
Member
|
cc @ArthurZucker maybe? Not sure who's working on the benchmarking |
ArthurZucker
left a comment
Collaborator
There was a problem hiding this comment.
Comments apply to all the files added here
Comment on lines
+30
to
+32
| def __init__(self, logger: logging.Logger): | ||
| super().__init__(logger) | ||
| self._default_prompt = "Why dogs are so cute?" |
Collaborator
There was a problem hiding this comment.
I think in general we'd want to run the models on the exact same prompts all the time no? it does not really matter anyways but I would decorelate this from the model
Comment on lines
+42
to
+55
| {"variant": "eager", "compile_mode": None, "use_cache": True, "description": "Eager execution with cache"}, | ||
| { | ||
| "variant": "compiled", | ||
| "compile_mode": "max-autotune", | ||
| "use_cache": True, | ||
| "description": "Compiled with max autotune", | ||
| }, | ||
| { | ||
| "variant": "kernelized", | ||
| "compile_mode": "max-autotune", | ||
| "use_cache": True, | ||
| "description": "Kernelized execution", | ||
| }, | ||
| ] |
Collaborator
There was a problem hiding this comment.
these are common IMO, not specific to bert, hsould not be seen here
Comment on lines
+57
to
+65
| def _is_kernelization_available(self) -> bool: | ||
| """Check if kernelization is available for BERT.""" | ||
| try: | ||
| from kernels import Mode, kernelize # noqa: F401 | ||
|
|
||
| return True | ||
| except ImportError: | ||
| self.logger.debug("Kernelization not available: kernels module not found") | ||
| return False |
Collaborator
There was a problem hiding this comment.
not really true, we should check if the model has a use_kernel_from_hub and have the kernel scenario activate only if model has them
Comment on lines
+67
to
+75
| def get_default_generation_config(self) -> dict[str, Any]: | ||
| """Get BERT-specific generation configuration.""" | ||
| return { | ||
| "do_sample": False, | ||
| "top_p": 1.0, | ||
| "temperature": 1.0, | ||
| "repetition_penalty": 1.0, | ||
| "max_new_tokens": None, | ||
| } |
Collaborator
There was a problem hiding this comment.
that's not default to bert
Comment on lines
+77
to
+91
| def get_model_init_kwargs(self, config) -> dict[str, Any]: | ||
| """Get BERT-specific model initialization kwargs.""" | ||
| return { | ||
| "torch_dtype": getattr(torch, config.torch_dtype), | ||
| "attn_implementation": config.attn_implementation, | ||
| "use_cache": True, | ||
| } | ||
|
|
||
| def get_default_torch_dtype(self) -> str: | ||
| """Get default torch dtype for BERT.""" | ||
| return "float16" | ||
|
|
||
| def get_default_device(self) -> str: | ||
| """Get default device for BERT.""" | ||
| return "cuda" |
Collaborator
There was a problem hiding this comment.
nothing specific here, should be in the base model AFAI can see
Comment on lines
+94
to
+117
| def run_bert(logger, output_dir, **kwargs): | ||
| """ | ||
| Run BERT benchmark with the given configuration. | ||
|
|
||
| Args: | ||
| logger: Logger instance | ||
| output_dir: Output directory for results | ||
| **kwargs: Additional configuration options | ||
|
|
||
| Returns: | ||
| Path to output file if successful | ||
| """ | ||
| from benchmark_framework import BenchmarkRunner | ||
|
|
||
| # Extract parameters with defaults | ||
| model_id = kwargs.get("model_id", "bert-base-uncased") | ||
| warmup_iterations = kwargs.get("warmup_iterations", 3) | ||
| measurement_iterations = kwargs.get("measurement_iterations", 5) | ||
| num_tokens_to_generate = kwargs.get("num_tokens_to_generate", 100) | ||
| include_sdpa_variants = kwargs.get("include_sdpa_variants", True) | ||
| device = kwargs.get("device", "cuda") | ||
| torch_dtype = kwargs.get("torch_dtype", "float16") | ||
| batch_size = kwargs.get("batch_size", 1) | ||
| commit_id = kwargs.get("commit_id") |
Collaborator
There was a problem hiding this comment.
this is also classic / should be shared by all model
Comment on lines
+125
to
+155
| # Create benchmark instance | ||
| benchmark = BERTBenchmark(logger) | ||
|
|
||
| # Create scenarios | ||
| scenarios = benchmark.create_scenarios( | ||
| model_id=model_id, | ||
| warmup_iterations=warmup_iterations, | ||
| measurement_iterations=measurement_iterations, | ||
| num_tokens_to_generate=num_tokens_to_generate, | ||
| include_sdpa_variants=include_sdpa_variants, | ||
| device=device, | ||
| torch_dtype=torch_dtype, | ||
| batch_size=batch_size, | ||
| ) | ||
|
|
||
| logger.info(f"Created {len(scenarios)} benchmark scenarios") | ||
|
|
||
| # Create runner and execute benchmarks | ||
| runner = BenchmarkRunner(logger, output_dir) | ||
| results = runner.run_benchmark(benchmark, scenarios, commit_id=commit_id) | ||
|
|
||
| if not results: | ||
| logger.warning("No successful benchmark results") | ||
| return None | ||
|
|
||
| # Save results | ||
| model_name = model_id.split("/")[-1] # Extract model name from ID | ||
| output_file = runner.save_results(model_name, results) | ||
|
|
||
| logger.info(f"BERT benchmark completed successfully. Results saved to: {output_file}") | ||
| return output_file |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do?
This PR adds bert, gemma3, gpt, mistral3 and qwen2 to the new benchmarking pipeline.
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.