Enable tie_word_embeddings config setting to enable / disable weigh…

…t tied embeddings (#728) * enable disabling embed weight tying * fix bug * updt with descriptive var names * fix hf config * move comment with code * bug fix * add _tie_weights method * undo mcli yaml change * refactor * add tests * Update llmfoundry/models/mpt/modeling_mpt.py Co-authored-by: Sasha Doubov <[email protected]> * pr comments * updt tests to guard against numerical issues --------- Co-authored-by: Sasha Doubov <[email protected]>
mosaicml · Nov 13, 2023 · 7899178 · 7899178
1 parent d11ba82
commit 7899178
Show file tree

Hide file tree

Showing 6 changed files with 169 additions and 60 deletions.
diff --git a/llmfoundry/models/mpt/configuration_mpt.py b/llmfoundry/models/mpt/configuration_mpt.py
@@ -59,6 +59,7 @@ def __init__(
         use_cache: bool = False,
         init_config: Dict = init_config_defaults,
         fc_type: str = 'torch',
+        tie_word_embeddings: bool = True,
         verbose: Optional[int] = None,
         **kwargs: Any,
     ):
@@ -128,6 +129,7 @@ def __init__(
                 ---
                 See llmfoundry.models.utils.param_init_fns.py for info on other param init config options
             fc_type (str): choose fc layer implementation. Options: torch and te. te layers support fp8 when using H100 GPUs.
+            tie_word_embeddings (bool): Whether to tie the input embedding and output layers.
         """
         self.d_model = d_model
         self.n_heads = n_heads
@@ -164,7 +166,11 @@ def __init__(
             warnings.warn(
                 f'alibi or rope is turned on, setting `learned_pos_emb` to `False.`'
             )
-        super().__init__(**kwargs)
+        # tie_word_embeddings is set in Huggingface's PretrainedConfig __init__
+        super().__init__(
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
 
         self._validate_config()
 

diff --git a/llmfoundry/models/mpt/modeling_mpt.py b/llmfoundry/models/mpt/modeling_mpt.py
@@ -231,10 +231,11 @@ def __init__(self, config: MPTConfig):
         log.debug(self)
         log.debug(f'Using {self.config.init_config["name"]} initialization.')
 
-    def get_input_embeddings(self) -> nn.Embedding:
+    def get_input_embeddings(self) -> Union[SharedEmbedding, nn.Embedding]:
         return self.wte
 
-    def set_input_embeddings(self, value: nn.Embedding) -> None:
+    def set_input_embeddings(
+            self, value: Union[SharedEmbedding, nn.Embedding]) -> None:
         self.wte = value
 
     @torch.no_grad()
@@ -574,14 +575,20 @@ class MPTForCausalLM(MPTPreTrainedModel):
 
     def __init__(self, config: MPTConfig):
         super().__init__(config)
-        if not config.tie_word_embeddings:
-            raise ValueError(
-                'MPTForCausalLM only supports tied word embeddings')
-
         log.info(f'Instantiating an MPTForCausalLM model from {__file__}')
 
         self.transformer: MPTModel = MPTModel(config)
 
+        self.lm_head = None
+        if not config.tie_word_embeddings:
+            self.lm_head = nn.Linear(
+                config.d_model,
+                config.vocab_size,
+                bias=False,
+                device=config.init_device,
+            )
+            self.lm_head._fsdp_wrap = True
+
         for child in self.transformer.children():
             if isinstance(child, torch.nn.ModuleList):
                 continue
@@ -602,19 +609,38 @@ def __init__(self, config: MPTConfig):
                     )
             self.logit_scale = logit_scale
 
-    def get_input_embeddings(self) -> nn.Embedding:
-        return self.transformer.wte
+    def get_input_embeddings(self) -> Union[SharedEmbedding, nn.Embedding]:
+        return self.transformer.get_input_embeddings()
 
     def set_input_embeddings(
             self, value: Union[SharedEmbedding, nn.Embedding]) -> None:
-        self.transformer.wte = value
+        self.transformer.set_input_embeddings(value)
 
-    def get_output_embeddings(self) -> nn.Embedding:
-        return self.transformer.wte
+    def get_output_embeddings(
+            self) -> Union[SharedEmbedding, nn.Embedding, nn.Linear]:
+        if self.lm_head is not None:
+            return self.lm_head
+        return self.transformer.get_input_embeddings()
 
     def set_output_embeddings(
-            self, new_embeddings: Union[SharedEmbedding, nn.Embedding]) -> None:
-        self.transformer.wte = new_embeddings
+        self, new_embeddings: Union[SharedEmbedding, nn.Embedding,
+                                    nn.Linear]) -> None:
+        if self.lm_head is not None:
+            self.lm_head = new_embeddings
+        else:
+            if not isinstance(new_embeddings, (SharedEmbedding, nn.Embedding)):
+                raise ValueError(
+                    'new_embeddings must be an instance of SharedEmbedding ' +
+                    f'or nn.Embedding, but got {type(new_embeddings)}.')
+            warnings.warn(
+                'Using `set_output_embeddings` to set the embedding layer of ' +
+                'MPTForCausalLM with tied weights. Given weights are tied, ' +
+                'using `set_input_embeddings` is recommended over using ' +
+                '`set_output_embeddings`.')
+            self.transformer.set_input_embeddings(new_embeddings)
+
+    def tie_weights(self) -> None:
+        self.lm_head = None
 
     def set_decoder(self, decoder: MPTModel) -> None:
         self.transformer = decoder
@@ -658,12 +684,14 @@ def forward(
             use_cache=use_cache,
         )
 
-        # move outputs to same device as weights for token embedding
-        # needed to support HF `device_map`
-        logits = self.transformer.wte(
-            outputs.last_hidden_state.to(self.transformer.wte.weight.device),
-            True,
-        )
+        if self.lm_head is not None:
+            logits = self.lm_head(outputs.last_hidden_state)
+        else:
+            # move outputs to same device as weights for token embedding
+            # needed to support HF `device_map`
+            out = outputs.last_hidden_state
+            out = out.to(self.transformer.wte.weight.device)
+            logits = self.transformer.wte(out, True)
 
         if self.logit_scale is not None:
             if self.logit_scale == 0:
@@ -859,7 +887,11 @@ def flops_per_batch(self, batch: Mapping) -> int:
         # assume the backward pass is approximately 2x the forward pass
 
         bs, msl = batch['input_ids'].shape[0:2]
-        params_flops_per_token = 2 * self.n_active_params
+        params = self.n_active_params
+        if not self.model.transformer.config.tie_word_embeddings:
+            # embedding layers are lookup tables, therefore are not counted in the FLOP computation
+            params -= self.model.transformer.wte.weight.numel()
+        params_flops_per_token = 2 * params
         params_flops_per_seq = params_flops_per_token * msl
         attn_flops_per_seq = (self.model.config.n_layers * 2 * 2 *
                               (self.model.config.d_model * (msl**2)))

diff --git a/tests/test_hf_conversion_script.py b/tests/test_hf_conversion_script.py
@@ -248,20 +248,21 @@ def test_callback_inits_with_defaults():
 
 @pytest.mark.world_size(2)
 @pytest.mark.gpu
-@pytest.mark.parametrize('model', ['mpt', 'neo', 'llama2'])
+@pytest.mark.parametrize(
+    'model,tie_word_embeddings',
+    [('mpt', True), ('mpt', False), ('neo', None), ('llama2', None)],
+)
 @pytest.mark.parametrize('fsdp_state_dict_type', ['full', 'sharded', None])
 @pytest.mark.parametrize('log_to_mlflow', [True, False])
 @pytest.mark.parametrize(
     'hf_save_interval,save_interval,max_duration,expected_hf_checkpoints,expected_normal_checkpoints',
     [('3ba', '2ba', '7ba', 3, 4), ('1dur', '2ba', '1ep', 1, 4)])
 @patch('os.cpu_count', MagicMock(return_value=None))
-def test_huggingface_conversion_callback(model: str, tmp_path: pathlib.Path,
-                                         fsdp_state_dict_type: Optional[str],
-                                         log_to_mlflow: bool,
-                                         hf_save_interval: str,
-                                         save_interval: str, max_duration: str,
-                                         expected_hf_checkpoints: int,
-                                         expected_normal_checkpoints: int):
+def test_huggingface_conversion_callback(
+        model: str, tmp_path: pathlib.Path, tie_word_embeddings: bool,
+        fsdp_state_dict_type: Optional[str], log_to_mlflow: bool,
+        hf_save_interval: str, save_interval: str, max_duration: str,
+        expected_hf_checkpoints: int, expected_normal_checkpoints: int):
     delete_transformers_cache()
 
     dist.initialize_dist(get_device('gpu'))
@@ -298,9 +299,11 @@ def test_huggingface_conversion_callback(model: str, tmp_path: pathlib.Path,
                 'attn_impl': 'torch',
             },
             'loss_fn': 'torch_crossentropy',
+            'tie_word_embeddings': tie_word_embeddings,
         }
         tokenizer_name = 'EleutherAI/gpt-neox-20b'
     elif model == 'neo':
+        assert tie_word_embeddings is None
         model_cfg = {
             'name': 'hf_causal_lm',
             'pretrained_model_name_or_path': 'EleutherAI/gpt-neo-125M',
@@ -313,6 +316,7 @@ def test_huggingface_conversion_callback(model: str, tmp_path: pathlib.Path,
         }
         tokenizer_name = 'EleutherAI/gpt-neo-125M'
     elif model == 'llama2':
+        assert tie_word_embeddings is None
         if 'HUGGING_FACE_HUB_TOKEN' not in os.environ:
             pytest.skip(
                 'The CI cluster does not have access to the Llama models, so skip this test.'
@@ -489,19 +493,26 @@ def test_huggingface_conversion_callback(model: str, tmp_path: pathlib.Path,
     delete_transformers_cache()
 
 
-@pytest.mark.parametrize('model', ['mpt', 'neo', 'llama2'])
-def test_convert_and_generate(model: str, tmp_path: pathlib.Path):
+@pytest.mark.parametrize(
+    'model,tie_word_embeddings',
+    [('mpt', True), ('mpt', False), ('neo', None), ('llama2', None)],
+)
+def test_convert_and_generate(model: str, tie_word_embeddings: bool,
+                              tmp_path: pathlib.Path):
     delete_transformers_cache()
 
     om_cfg = None
     if model == 'mpt':
         om_cfg = get_config(
             conf_path='scripts/train/yamls/pretrain/testing.yaml')
+        om_cfg['tie_word_embeddings'] = tie_word_embeddings
     elif model == 'neo':
+        assert tie_word_embeddings is None
         om_cfg = get_config(
             conf_path='scripts/train/yamls/pretrain/gpt-neo-125m.yaml')
         om_cfg['model']['config_overrides']['hidden_size'] = 36
     elif model == 'llama2':
+        assert tie_word_embeddings is None
         if 'HUGGING_FACE_HUB_TOKEN' not in os.environ:
             pytest.skip(
                 'The CI cluster does not have access to the Llama models, so skip this test.'
@@ -562,11 +573,14 @@ def test_convert_and_generate(model: str, tmp_path: pathlib.Path):
 
 
 @pytest.mark.gpu
-def test_convert_and_generate_triton(tmp_path: pathlib.Path):
+@pytest.mark.parametrize('tie_word_embeddings', [True, False])
+def test_convert_and_generate_triton(tie_word_embeddings: str,
+                                     tmp_path: pathlib.Path):
     delete_transformers_cache()
 
     cfg = get_config()
     cfg['model']['init_device'] = 'cpu'
+    cfg['tie_word_embeddings'] = tie_word_embeddings
     tokenizer = transformers.AutoTokenizer.from_pretrained(
         'EleutherAI/gpt-neox-20b')
     model = ComposerMPTCausalLM(cfg['model'], tokenizer)
@@ -602,7 +616,9 @@ def test_convert_and_generate_triton(tmp_path: pathlib.Path):
     delete_transformers_cache()
 
 
-def test_convert_and_generate_meta(tmp_path: pathlib.Path):
+@pytest.mark.parametrize('tie_word_embeddings', [True, False])
+def test_convert_and_generate_meta(tie_word_embeddings: str,
+                                   tmp_path: pathlib.Path):
     delete_transformers_cache()
 
     from composer.utils import dist
@@ -612,6 +628,7 @@ def test_convert_and_generate_meta(tmp_path: pathlib.Path):
     om_cfg = get_config(conf_path='scripts/train/yamls/pretrain/testing.yaml')
 
     om_cfg['model']['init_device'] = 'cpu'
+    om_cfg['tie_word_embeddings'] = tie_word_embeddings
     tokenizer = transformers.AutoTokenizer.from_pretrained(
         om_cfg.tokenizer.name)
     original_model = COMPOSER_MODEL_REGISTRY[om_cfg['model'].name](