Merge branch 'main' into main

mosaicml · Sep 21, 2023 · 9853135 · 9853135
2 parents 507b4df + 299e737
commit 9853135
Show file tree

Hide file tree

Showing 6 changed files with 14 additions and 8 deletions.
diff --git a/.github/workflows/pytest-cpu.yaml b/.github/workflows/pytest-cpu.yaml
@@ -27,7 +27,7 @@ jobs:
         set -ex
         export PATH=/composer-python:$PATH
         python -m pip install --upgrade 'pip<23' wheel
-        python -m pip install --upgrade .[dev]
+        python -m pip install --upgrade .[all-cpu]
     - name: Run Tests
       id: tests
       run: |

diff --git a/scripts/eval/yamls/hf_8bit_eval.yaml b/scripts/eval/yamls/hf_8bit_eval.yaml
@@ -10,7 +10,7 @@ models:
   model:
     name: hf_causal_lm
     pretrained_model_name_or_path: ${model_name_or_path}
-    init_device: cpu
+    init_device: mixed
     pretrained: true
     load_in_8bit: true
   tokenizer:

diff --git a/scripts/eval/yamls/hf_eval.yaml b/scripts/eval/yamls/hf_eval.yaml
@@ -12,7 +12,7 @@ models:
   model:
     name: hf_causal_lm
     pretrained_model_name_or_path: ${model_name_or_path}
-    init_device: cpu
+    init_device: mixed
     pretrained: true
   tokenizer:
     name: ${model_name_or_path}

diff --git a/scripts/eval/yamls/hf_lora_eval.yml b/scripts/eval/yamls/hf_lora_eval.yml
@@ -5,7 +5,9 @@ precision: amp_fp16
 # If you are using one model, put it here:
 model_name_or_path: EleutherAI/gpt-neo-125m
 # If you are using a seperated lora weight, put it here:
-lora_id_or_path: nathan0/lora-gpt-neo-125m-alpaca
+# lora weights must be compatible with the specified model
+lora_id_or_path: edbeeching/gpt-neo-125M-imdb-lora # Example lora weights for gpt-neo-125m
+
 # otherwise, write a block for each model you want to test in the `models` section
 
 models:
@@ -14,7 +16,7 @@ models:
   model:
     name: hf_causal_lm
     pretrained_model_name_or_path: ${model_name_or_path}
-    init_device: cpu
+    init_device: mixed
     pretrained: true
     pretrained_lora_id_or_path: ${lora_id_or_path}
   tokenizer:

diff --git a/scripts/eval/yamls/mpt_eval.yaml b/scripts/eval/yamls/mpt_eval.yaml
@@ -3,7 +3,6 @@ tokenizer_name: EleutherAI/gpt-neox-20b
 seed: 1
 precision: amp_fp16
 
-
 models:
 -
   model_name: mpt_test
@@ -14,7 +13,8 @@ models:
       model_max_length: ${max_seq_len}
   model:
     name: mpt_causal_lm
-    init_device: meta
+    init_device: mixed
+    # Set the below model parameters to match the checkpoint specified with load_path
     d_model: 768
     n_heads: 12
     n_layers: 12

diff --git a/setup.py b/setup.py
@@ -47,7 +47,7 @@
 ]
 
 install_requires = [
-    'mosaicml[libcloud,wandb,mlflow]>=0.16.1,<0.17',
+    'mosaicml[libcloud,wandb,mlflow,oci,gcs]>=0.16.1,<0.17',
     'accelerate>=0.20,<0.21',  # for HF inference `device_map`
     'transformers>=4.33,<4.34',
     'mosaicml-streaming>=0.6,<0.7',
@@ -64,6 +64,8 @@
     'cmake>=3.25.0,<=3.26.3',  # required for triton-pre-mlir below
     # PyPI does not support direct dependencies, so we remove this line before uploading from PyPI
     'triton-pre-mlir@git+https://github.com/vchiley/triton.git@triton_pre_mlir_sm90#subdirectory=python',
+    'boto3>=1.21.45,<2',
+    'huggingface-hub>=0.17.0,<1.0',
 ]
 
 extra_deps = {}
@@ -103,6 +105,8 @@
     'openai==0.27.8',
     'tiktoken==0.4.0',
 ]
+extra_deps['all-cpu'] = set(
+    dep for key, deps in extra_deps.items() for dep in deps if 'gpu' not in key)
 extra_deps['all'] = set(dep for deps in extra_deps.values() for dep in deps)
 
 setup(