Merge pull request #572 from gakada/perf

Add --max_batch_size and --batch_size auto:N

lm_eval/base.py

@@ -119,6 +119,12 @@ class LM(abc.ABC):
 
 
 class BaseLM(LM):
+    def __init__(self):
+        super().__init__()
+        self.batch_schedule = 1
+        self.batch_sizes = {}
+        self.max_batch_size = 512
+
     @property
     @abstractmethod
     def eot_token_id(self):
@@ -167,6 +173,26 @@ class BaseLM(LM):
         """
         pass
 
+    def _detect_batch_size(self, requests=None, pos=0):
+        if requests:
+            _, context_enc, continuation_enc = requests[pos]
+            max_length = len((context_enc + continuation_enc)[-(self.max_length + 1) :][:-1])
+        else:
+            max_length = self.max_length
+
+        # if OOM, then halves batch_size and tries again
+        @find_executable_batch_size(starting_batch_size=self.max_batch_size)
+        def forward_batch(batch_size):
+            test_batch = torch.ones((batch_size, max_length), device=self.device).long()
+            for _ in range(5):
+                _ = F.log_softmax(self._model_call(test_batch), dim=-1).cpu()
+            return batch_size
+
+        batch_size = forward_batch()
+        utils.clear_torch_cache()
+
+        return batch_size
+
     # subclass must implement properties vocab_size, eot_token_id, max_gen_toks, batch_size, device, max_length.
     # TODO: enforce this somehow
 
@@ -202,19 +228,7 @@ class BaseLM(LM):
         if self.batch_size == "auto":
             # using rolling window with maximum context
             print("Passed argument batch_size = auto. Detecting largest batch size")
-
-            @find_executable_batch_size(
-                starting_batch_size=512
-            )  # if OOM, then halves batch_size and tries again
-            def forward_batch(batch_size):
-                test_batch = torch.ones(
-                    (batch_size, self.max_length), device=self.device
-                ).long()
-                for _ in range(5):
-                    _ = F.log_softmax(self._model_call(test_batch), dim=-1).cpu()
-                return batch_size
-
-            batch_size = forward_batch()
+            batch_size = self._detect_batch_size()
             print(f"Determined Largest batch size: {batch_size}")
             adaptive_batch_size = batch_size
 
@@ -267,34 +281,24 @@ class BaseLM(LM):
 
         re_ord = utils.Reorderer(requests, _collate)
 
+        reordered_requests = re_ord.get_reordered()
+        n_reordered_requests = len(reordered_requests)
+
         # automatic (variable) batch size detection for vectorization
         # pull longest context sample from request
-        if len(re_ord.get_reordered()) > 0:
-            _, context_enc, continuation_enc = re_ord.get_reordered()[0]
-            max_context = len((context_enc + continuation_enc)[-(self.max_length + 1) :][:-1])
-            if (self.batch_size == 'auto'):
-
-                if override_bs is None:
-                    print('Passed argument batch_size = auto. Detecting largest batch size')
-                    @find_executable_batch_size(starting_batch_size=512) # if OOM, then halves batch_size and tries again
-                    def forward_batch(batch_size):
-                        test_batch = torch.ones((batch_size, max_context), device=self.device).long()
-                        for _ in range(5):
-                            out = F.log_softmax(self._model_call(test_batch), dim = -1).cpu()
-                        return batch_size
-
-                    batch_size = forward_batch()
-                    print(f"Determined largest batch size: {batch_size}")
-                    adaptive_batch_size = batch_size
-
-                else:
-                    adaptive_batch_size = override_bs
-        else:
-            adaptive_batch_size = 0 if override_bs is None else override_bs
+        def _batch_scheduler(pos):
+            sched = pos // int(n_reordered_requests / self.batch_schedule)
+            if sched in self.batch_sizes:
+                return self.batch_sizes[sched]
+            print(f"Passed argument batch_size = auto:{self.batch_schedule}. Detecting largest batch size")
+            self.batch_sizes[sched] = self._detect_batch_size(reordered_requests, pos)
+            print(f"Determined largest batch size: {self.batch_sizes[sched]}")
+            return self.batch_sizes[sched]
 
         for chunk in utils.chunks(
-            tqdm(re_ord.get_reordered(), disable=disable_tqdm),
-            self.batch_size if self.batch_size != "auto" else adaptive_batch_size,
+            tqdm(reordered_requests, disable=disable_tqdm),
+            n=self.batch_size if self.batch_size != "auto" else override_bs if override_bs is not None else 0,
+            fn=_batch_scheduler if self.batch_size == "auto" and n_reordered_requests > 0 else None,
         ):
             inps = []
             cont_toks_list = []

lm_eval/evaluator.py

@@ -16,6 +16,7 @@ def simple_evaluate(
     tasks=[],
     num_fewshot=0,
     batch_size=None,
+    max_batch_size=None,
     device=None,
     no_cache=False,
     limit=None,
@@ -37,8 +38,10 @@ def simple_evaluate(
         List of task names or Task objects. Task objects will be taken to have name task.EVAL_HARNESS_NAME if defined and type(task).__name__ otherwise.
     :param num_fewshot: int
         Number of examples in few-shot context
-    :param batch_size: int, optional
+    :param batch_size: int or str, optional
         Batch size for model
+    :param max_batch_size: int, optional
+        Maximal batch size to try with automatic batch size detection
     :param device: str, optional
         PyTorch device (e.g. "cpu" or "cuda:0") for running models
     :param no_cache: bool
@@ -67,7 +70,7 @@ def simple_evaluate(
         if model_args is None:
             model_args = ""
         lm = lm_eval.models.get_model(model).create_from_arg_string(
-            model_args, {"batch_size": batch_size, "device": device}
+            model_args, {"batch_size": batch_size, "max_batch_size": max_batch_size, "device": device}
         )
     else:
         assert isinstance(model, lm_eval.base.LM)
@@ -106,6 +109,7 @@ def simple_evaluate(
         "model_args": model_args,
         "num_fewshot": num_fewshot,
         "batch_size": batch_size,
+        "batch_sizes": list(lm.batch_sizes.values()),
         "device": device,
         "no_cache": no_cache,
         "limit": limit,

lm_eval/models/huggingface.py

@@ -9,7 +9,6 @@ from typing import List, Mapping, NewType, Optional, Tuple, Union
 from tqdm import tqdm
 
 from transformers import BatchEncoding
-from accelerate import find_executable_batch_size
 
 from lm_eval import utils
 from lm_eval.base import BaseLM
@@ -76,6 +75,7 @@ class HuggingFaceAutoLM(BaseLM):
         subfolder: Optional[str] = None,
         revision: Optional[str] = "main",
         batch_size: Optional[Union[int, str]] = 1,
+        max_batch_size: Optional[int] = 512,
         max_gen_toks: Optional[int] = 256,
         max_length: Optional[int] = None,
         add_special_tokens: Optional[bool] = None,
@@ -172,10 +172,13 @@ class HuggingFaceAutoLM(BaseLM):
             ), "Evaluating causal models with `add_special_tokens=True` is currently not supported."
 
         # setup for automatic batch size detection
-        if batch_size == "auto":
-            self._batch_size = batch_size
+        if str(batch_size).startswith("auto"):
+            batch_size = batch_size.split(":")
+            self._batch_size = batch_size[0]
+            self.batch_schedule = float(batch_size[1]) if len(batch_size) > 1 else 1
         else:
             self._batch_size = int(batch_size)
+        self.max_batch_size = max_batch_size
 
         self._max_gen_toks = max_gen_toks
         self._max_length = max_length
@@ -411,19 +414,7 @@ class HuggingFaceAutoLM(BaseLM):
         if self.batch_size == "auto":
             # using rolling window with maximum context
             print("Passed argument batch_size = auto. Detecting largest batch size")
-
-            @find_executable_batch_size(
-                starting_batch_size=512
-            )  # if OOM, then halves batch_size and tries again
-            def forward_batch(batch_size):
-                test_batch = torch.ones(
-                    (batch_size, self.max_length), device=self.device
-                ).long()
-                for _ in range(5):
-                    _ = F.log_softmax(self._model_call(test_batch), dim=-1).cpu()
-                return batch_size
-
-            batch_size = forward_batch()
+            batch_size = self._detect_batch_size()
             print(f"Determined Largest batch size: {batch_size}")
             adaptive_batch_size = batch_size
 

lm_eval/utils.py

@@ -8,6 +8,7 @@ import sys
 import fnmatch
 from typing import List, Union
 
+import gc
 import torch
 
 from omegaconf import OmegaConf
@@ -64,11 +65,11 @@ def join_iters(iters):
         yield from iter
 
 
-def chunks(iter, n):
+def chunks(iter, n=0, fn=None):
     arr = []
-    for x in iter:
+    for i, x in enumerate(iter):
         arr.append(x)
-        if len(arr) == n:
+        if len(arr) == (fn(i) if fn else n):
             yield arr
             arr = []
 
@@ -283,3 +284,8 @@ def run_task_tests(task_list: List[str]):
         raise ValueError(
             f"Not all tests for the specified tasks ({task_list}) ran successfully! Error code: {pytest_return_val}"
         )
+
+
+def clear_torch_cache():
+    gc.collect()
+    torch.cuda.empty_cache()

main.py

@@ -16,6 +16,8 @@ def parse_args():
     parser.add_argument("--provide_description", action="store_true")
     parser.add_argument("--num_fewshot", type=int, default=0)
     parser.add_argument("--batch_size", type=str, default=None)
+    parser.add_argument("--max_batch_size", type=int, default=None,
+                        help="Maximal batch size to try with --batch_size auto")
     parser.add_argument("--device", type=str, default=None)
     parser.add_argument("--output_path", default=None)
     parser.add_argument("--limit", type=float, default=None,
@@ -60,6 +62,7 @@ def main():
         tasks=task_names,
         num_fewshot=args.num_fewshot,
         batch_size=args.batch_size,
+        max_batch_size=args.max_batch_size,
         device=args.device,
         no_cache=args.no_cache,
         limit=args.limit,
@@ -78,9 +81,10 @@ def main():
         with open(args.output_path, "w") as f:
             f.write(dumped)
 
+    batch_sizes = ",".join(map(str, results["config"]["batch_sizes"]))
     print(
         f"{args.model} ({args.model_args}), limit: {args.limit}, provide_description: {args.provide_description}, "
-        f"num_fewshot: {args.num_fewshot}, batch_size: {args.batch_size}"
+        f"num_fewshot: {args.num_fewshot}, batch_size: {args.batch_size}{f' ({batch_sizes})' if batch_sizes else ''}"
     )
     print(evaluator.make_table(results))