Convert Args to DictConfig (#9450) (#9717)

Summary:
1. Add structure to the complicated args
2. Convert args to DictConfig, to decouple the cli args
3. Pass needed sub configs to functions (instead of args)


Test Plan:
```
python3 -m examples.models.llama.export_llama -c stories110M.pt -p params.json -d fp32 -n tinyllama_xnnpack+custom_fp32_main.pte -kv -X --xnnpack-extended-ops -qmode 8da4w -G 128 --use_sdpa_with_kv_cache --output-dir tmp -E 8,64
```

Reviewed By: larryliu0820

Pulled By:
jackzhxng

jackzhxng

Differential Revision: D71557301

Author: jackzhxng

examples/models/llama/eval_llama.py

@@ -11,6 +11,7 @@
 import torch
 
 from .eval_llama_lib import (
+    _convert_cli_to_config_format,
     build_args_parser,
     eval_llama,
     eval_llama_with_attention_sink,
@@ -28,10 +29,11 @@ def main() -> None:
     args = parser.parse_args()
     # Overrides this arg, because evaluation requires full logits.
     args.generate_full_logits = True
+    config = _convert_cli_to_config_format(args)
     if args.use_attention_sink:
-        eval_llama_with_attention_sink(modelname, args)  # pyre-ignore
+        eval_llama_with_attention_sink(modelname, config)
     else:
-        eval_llama(modelname, args)  # pyre-ignore
+        eval_llama(modelname, config)
 
 
 if __name__ == "__main__":

examples/models/llama/eval_llama_lib.py

@@ -4,33 +4,26 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-
 import argparse
-
 from typing import Optional, Union
 
 import torch
-
-from datasets import load_dataset
 from executorch.examples.models.llama.export_llama_lib import (
+    _convert_args_to_config,
+    _prepare_for_llama_export,
+    build_args_parser as _build_args_parser,
     get_quantizer_and_quant_params,
 )
 
 from executorch.extension.llm.export.builder import LLMEdgeManager
 from lm_eval.evaluator import simple_evaluate
+from omegaconf import DictConfig, OmegaConf
 from pytorch_tokenizers import get_tokenizer
 from pytorch_tokenizers.llama2c import Llama2cTokenizer as SentencePieceTokenizer
 from pytorch_tokenizers.tiktoken import TiktokenTokenizer as Tiktoken
-from torch.nn import CrossEntropyLoss
-from tqdm import tqdm
 
 from .evaluate.eager_eval import EagerEvalWrapper
 
-from .export_llama_lib import (
-    _prepare_for_llama_export,
-    build_args_parser as _build_args_parser,
-)
-
 
 class GraphModuleEvalWrapper(EagerEvalWrapper):
     """
@@ -163,7 +156,7 @@ def _model_call(self, inps):
 
 def gen_eval_wrapper(
     model_name: str,
-    args: argparse.ArgumentParser,
+    config: DictConfig,
 ):
     """
     Generates a wrapper interface around the provided model and tokenizer for
@@ -172,17 +165,17 @@ def gen_eval_wrapper(
     Returns:
         eval_wrapper (LM): A wrapper interface for the lm-evaluation-harness library.
     """
-    tokenizer = get_tokenizer(args.tokenizer_path)  # pyre-ignore
+    tokenizer = get_tokenizer(config.model.tokenizer_path)
 
     # ExecuTorch Binary Evaluation
-    if (model := args.pte) is not None:  # pyre-ignore
-        if (tokenizer_bin := args.tokenizer_bin) is not None:  # pyre-ignore
+    if (model := config.eval.pte) is not None:
+        if (tokenizer_bin := config.eval.tokenizer_bin) is not None:
             # ETRunnerEvalWrapper: Create a wrapper around an ExecuTorch model, evaluated at runtime
             return ETRunnerEvalWrapper(
                 model=model,
                 tokenizer=tokenizer,
                 tokenizer_bin=tokenizer_bin,
-                max_seq_length=args.max_seq_length,  # pyre-ignore
+                max_seq_length=config.sequence.max_seq_length,
             )
 
         # ETPybindEvalWrapper: Create a wrapper around an ExecuTorch model, evaluated with pybindings
@@ -191,12 +184,12 @@ def gen_eval_wrapper(
             tokenizer=tokenizer,
             # Exported model takes at most (max_seq_length - 1) tokens.
             # Note that the eager model takes at most max_seq_length tokens.
-            max_seq_length=args.max_seq_length - 1,
+            max_seq_length=config.sequence.max_seq_length - 1,
         )
 
-    pt2e_quant_params, quantizers, quant_dtype = get_quantizer_and_quant_params(args)
+    pt2e_quant_params, quantizers, quant_dtype = get_quantizer_and_quant_params(config)
     # GPTFastEvalWrapper: Create a wrapper around a pre-exported model
-    manager: LLMEdgeManager = _prepare_for_llama_export(args)
+    manager: LLMEdgeManager = _prepare_for_llama_export(config)
 
     if len(quantizers) != 0:
         manager = manager.export().pt2e_quantize(quantizers)
@@ -208,9 +201,9 @@ def gen_eval_wrapper(
         return GraphModuleEvalWrapper(
             model=model,
             tokenizer=tokenizer,
-            max_seq_length=args.max_seq_length,
-            use_kv_cache=args.use_kv_cache,  # pyre-ignore
-            enable_dynamic_shape=args.enable_dynamic_shape,  # pyre-ignore
+            max_seq_length=config.sequence.max_seq_length,
+            use_kv_cache=config.kv_cache.use_kv_cache,
+            enable_dynamic_shape=config.misc.enable_dynamic_shape,
         )
     else:
         # TODO: use manager.pre_autograd_graph_module for the eval to remove the if-else branch
@@ -228,18 +221,94 @@ def gen_eval_wrapper(
         # that is not available in this eval_llama. We save the checkpoint
         # here for consistency with eval_llama. The accuracy results we
         # get from eval_llama can be used as a reference to other evaluations.
-        if args.output_eager_checkpoint_file is not None:  # pyre-ignore
-            torch.save(model, args.output_eager_checkpoint_file)
+        if config.eval.output_eager_checkpoint_file is not None:
+            torch.save(model, config.eval.output_eager_checkpoint_file)
 
         return EagerEvalWrapper(
             model=model,
             tokenizer=tokenizer,
-            max_seq_length=args.max_seq_length,
-            use_kv_cache=args.use_kv_cache,
+            max_seq_length=config.sequence.max_seq_length,
+            use_kv_cache=config.kv_cache.use_kv_cache,
+        )
+
+
+def eval_llama(
+    model_name: str,
+    config: DictConfig,
+) -> None:
+    # Generate the eval wrapper
+    eval_wrapper = gen_eval_wrapper(model_name, config)
+
+    # Needed for loading mmlu dataset.
+    # See https://github.com/EleutherAI/lm-evaluation-harness/pull/1998/files
+    if config.eval.tasks and "mmlu" in config.eval.tasks:
+        import datasets
+
+        datasets.config.HF_DATASETS_TRUST_REMOTE_CODE = True
+
+    # Evaluate the model
+    tasks = (
+        None if config.eval.tasks is None else OmegaConf.to_container(config.eval.tasks)
+    )
+    with torch.no_grad():
+        eval_results = simple_evaluate(
+            model=eval_wrapper,
+            tasks=tasks,
+            num_fewshot=config.eval.num_fewshot,
+            limit=config.eval.limit,
         )
 
+    for task, res in eval_results["results"].items():
+        print(f"{task}: {res}")
+
+
+def eval_llama_with_attention_sink(
+    model_name: str,
+    config: DictConfig,
+) -> None:
+    # Generate the eval wrapper
+    eval_wrapper = gen_eval_wrapper(model_name, config)
+
+    # Needed for loading mmlu dataset.
+    # See https://github.com/EleutherAI/lm-evaluation-harness/pull/1998/files
+    if config.eval.tasks and "mmlu" in config.eval.tasks:
+        import datasets
+
+        datasets.config.HF_DATASETS_TRUST_REMOTE_CODE = True
+
+    # Evaluate the model
+    with torch.no_grad():
+        eval_results = simple_evaluate(
+            model=eval_wrapper,
+            tasks=OmegaConf.to_container(config.eval.tasks),
+            num_fewshot=config.eval.num_fewshot,
+            limit=config.eval.limit,
+        )
+
+    for task, res in eval_results["results"].items():
+        print(f"{task}: {res}")
+
+
+def _convert_cli_to_config_format(args) -> DictConfig:
+    """Convert CLI arguments to config format."""
+    # First convert common args using the shared function
+    config = _convert_args_to_config(args)
+
+    # Add evaluation-specific settings
+    config.eval = OmegaConf.create()
+    config.eval.tasks = args.tasks
+    config.eval.limit = args.limit
+    config.eval.num_fewshot = args.num_fewshot
+    config.eval.pte = args.pte
+    config.eval.tokenizer_bin = args.tokenizer_bin
+    config.eval.output_eager_checkpoint_file = args.output_eager_checkpoint_file
+    config.eval.attention_sink_eval_tokens = args.attention_sink_eval_tokens
+
+    return config
+
 
 def build_args_parser() -> argparse.ArgumentParser:
+    """Build argument parser for evaluation, extending the export parser with eval-specific args."""
     # Start with arg parser from export_llama_lib
     parser = _build_args_parser()
 
@@ -286,92 +355,7 @@ def build_args_parser() -> argparse.ArgumentParser:
         help="Save the checkpoint after source transformations, for other evaluation platform to run the same checkpoint.",
     )
 
-    # Set of parameters secpific to AttentionSink.
+    # Set of parameters specific to AttentionSink.
     parser.add_argument("--attention_sink_eval_tokens", type=int, default=0)
 
     return parser
-
-
-def eval_llama(
-    model_name: str,
-    args: argparse.ArgumentParser,
-) -> None:
-    # Generate the eval wrapper
-    eval_wrapper = gen_eval_wrapper(model_name, args)
-
-    # Needed for loading mmlu dataset.
-    # See https://github.com/EleutherAI/lm-evaluation-harness/pull/1998/files
-    # pyre-ignore: Undefined attribute [16]: `argparse.ArgumentParser` has no attribute `tasks`
-    if args.tasks and "mmlu" in args.tasks:
-        import datasets
-
-        datasets.config.HF_DATASETS_TRUST_REMOTE_CODE = True
-
-    # Evaluate the model
-    with torch.no_grad():
-        eval_results = simple_evaluate(
-            model=eval_wrapper,
-            tasks=args.tasks,
-            num_fewshot=args.num_fewshot,  # pyre-ignore: Undefined attribute [16]: `argparse.ArgumentParser` has no attribute `num_fewshot`
-            limit=args.limit,  # pyre-ignore: Undefined attribute [16]: `argparse.ArgumentParser` has no attribute `limit`
-        )
-
-    for task, res in eval_results["results"].items():
-        print(f"{task}: {res}")
-
-
-def eval_llama_with_attention_sink(model_name: str, args: argparse.ArgumentParser):
-    """
-    Evaluate the model's perplexity when AttentionSink is enabled.
-
-    This is mostly copied from https://github.com/mit-han-lab/streaming-llm/blob/main/examples/eval_long_ppl.py
-    """
-    assert args.use_attention_sink is not None  # pyre-ignore [16]
-    assert args.attention_sink_eval_tokens > 0  # pyre-ignore [16]
-    attention_sink_params = args.use_attention_sink.split(",")
-    assert len(attention_sink_params) == 3
-    sink_size = int(attention_sink_params[0])
-    window_size = int(attention_sink_params[1])
-
-    assert args.max_seq_length == sink_size + window_size  # pyre-ignore [16]
-
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    manager: LLMEdgeManager = _prepare_for_llama_export(args)
-    model = manager.model.eval().to(device=device)
-    tokenizer = get_tokenizer(args.tokenizer_path)  # pyre-ignore [16]
-
-    eval_data = load_dataset("wikitext", "wikitext-2-raw-v1", split="test")
-
-    nlls = []
-    loss_fn = CrossEntropyLoss(reduction="none")
-    progress_bar = tqdm(total=args.attention_sink_eval_tokens)
-    input_pos = 0
-    while input_pos < args.attention_sink_eval_tokens:
-        for text in eval_data["text"]:  # pyre-ignore [16]
-            tokens = tokenizer.encode(text, bos=False, eos=False)
-            if len(tokens) <= 0:
-                continue
-            with torch.no_grad():
-                num_tokens = min(
-                    len(tokens) - 1, args.attention_sink_eval_tokens - input_pos
-                )
-                logits = model(
-                    torch.tensor(
-                        [tokens[:num_tokens]], dtype=torch.int64, device=device
-                    ),
-                    torch.tensor([input_pos], dtype=torch.int64, device=device),
-                ).squeeze(dim=0)
-                neg_log_likelihood = loss_fn(
-                    logits,
-                    torch.tensor(
-                        [tokens[1 : num_tokens + 1]], dtype=torch.int64, device=device
-                    ).view(-1),
-                )
-                nlls.append(neg_log_likelihood)
-                input_pos += num_tokens
-                progress_bar.update(num_tokens)
-            if input_pos >= args.attention_sink_eval_tokens:
-                break
-    ppl = torch.exp(torch.cat(nlls).mean())
-    print(f"Perplexity: {ppl.item()}")
-    return ppl.item()