restore files

llmfoundry/models/hf/hf_causal_lm.py

@@ -5,16 +5,18 @@
 
 import logging
 import os
+import warnings
 from typing import Mapping, Union
 
 # required for loading a python model into composer
 import transformers
-from composer.metrics.nlp import (
-    InContextLearningCodeEvalAccuracy,
-    InContextLearningLMAccuracy, InContextLearningLMExpectedCalibrationError,
-    InContextLearningMCExpectedCalibrationError,
-    InContextLearningMultipleChoiceAccuracy, InContextLearningQAAccuracy,
-    LanguageCrossEntropy, LanguagePerplexity)
+from composer.metrics.nlp import (InContextLearningCodeEvalAccuracy,
+                                  InContextLearningLMAccuracy,
+                                  InContextLearningLMExpectedCalibrationError,
+                                  InContextLearningMCExpectedCalibrationError,
+                                  InContextLearningMultipleChoiceAccuracy,
+                                  InContextLearningQAAccuracy,
+                                  LanguageCrossEntropy, LanguagePerplexity)
 from composer.utils import dist
 from omegaconf import DictConfig
 from torch import nn
@@ -156,6 +158,24 @@ def __init__(self, om_model_config: Union[DictConfig,
             if dist.get_local_rank() != 0 and init_device == 'mixed':
                 om_model_config.pretrained = False
 
+            # If the HuggingFace model is coming from a local folder, Hugging Face copies the modules into the
+            # transformers modules cache. On particular systems, this operation seems to cause contention between
+            # the different processes. To avoid this contention, we first create the model (on meta device) on local rank
+            # zero. This will set up the transformers model cache and avoid the future contention.
+            if dist.get_local_rank() == 0 and os.path.isdir(
+                    om_model_config.pretrained_model_name_or_path):
+                with init_empty_weights(include_buffers=False):
+                    with warnings.catch_warnings():
+                        warnings.simplefilter('ignore', UserWarning)
+                        AutoModelForCausalLM.from_pretrained(
+                            om_model_config.pretrained_model_name_or_path,
+                            trust_remote_code=trust_remote_code,
+                            use_auth_token=use_auth_token,
+                            config=config,
+                        )
+
+            dist.barrier()
+
             # initialize the model on the correct device
             if resolved_init_device == 'cpu':
                 if om_model_config.pretrained:

mcli/mcli-hf-eval.yaml

@@ -1,7 +1,7 @@
 integrations:
 - integration_type: git_repo
   git_repo: mosaicml/llm-foundry
-  git_branch:  execution_prediction # v0.2.0
+  git_branch: v0.3.0
   # git_commit: # OR use your commit hash
   pip_install: -e ".[gpu]"
   ssh_clone: false # Should be true if using a private repo
@@ -11,10 +11,10 @@ command: |
   composer eval/eval.py /mnt/config/parameters.yaml
 
 # Mosaic Cloud will use run_name (with a unique suffix) to populate the env var $RUN_NAME
-run_name: mpt-gauntlet-v0.1
+run_name: mpt-eval
 gpu_num: 8
-gpu_type: a100_80gb
-cluster:  r1z1 # replace with your cluster here!
+# gpu_type:
+# cluster:  # replace with your cluster here!
 
 image: mosaicml/llm-foundry:2.0.1_cu118-latest
 
@@ -31,7 +31,7 @@ parameters:
     model_name: mosaicml/mpt-7b-instruct
     # Tokenizer
     tokenizer:
-      name: mosaicml/mpt-7b-instruct
+      name: EleutherAI/gpt-neox-20b
       kwargs:
         model_max_length: ${max_seq_len}
 
@@ -41,14 +41,14 @@ parameters:
       init_device: mixed
       pretrained: true
       use_auth_token: false
-
 
   # FSDP config for model sharding
   fsdp_config:
     sharding_strategy: FULL_SHARD
-    mixed_precision: PURE
+    mixed_precision: FULL
     forward_prefetch: True
     limit_all_gathers: True
 
+
   icl_tasks: 'eval/yamls/tasks.yaml'
   eval_gauntlet: 'eval/yamls/eval_gauntlet.yaml'

scripts/eval/eval.py

@@ -117,6 +117,7 @@ def evaluate_model(
     tokenizer_name = tokenizer_cfg['name']
     tokenizer_kwargs = tokenizer_cfg.get('kwargs', {})
     tokenizer = build_tokenizer(tokenizer_name, tokenizer_kwargs)
+
     evaluators, logger_keys, eval_gauntlet_callback = build_icl_data_and_gauntlet(
         icl_tasks, eval_gauntlet_config, tokenizer, device_eval_batch_size,
         max_seq_len, icl_subset_num_batches)
@@ -173,6 +174,7 @@ def evaluate_model(
         dist_timeout=dist_timeout,
         python_log_level=python_log_level,
     )
+
     if torch.cuda.is_available():
         torch.cuda.synchronize()
     a = time.time()
@@ -315,7 +317,6 @@ def main(cfg: DictConfig):
             row = {'model_name': model_cfg['model_name']}
             row.update(
                 {k.split('/')[-1]: v for k, v in composite_scores.items()})
-
             eval_gauntlet_df = pd.concat(
                 [eval_gauntlet_df, pd.DataFrame([row])], ignore_index=True)