Add Mistral-7B-Instruct-v0.1 from huggingface. (#2010)

Summary:
Add Mistral-7B-Instruct-v0.1 from huggingface. See https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1

Pull Request resolved: #2010

Reviewed By: aaronenyeshi

Differential Revision: D50692789

Pulled By: xuzhao9

fbshipit-source-id: 7aa3c455882726ac3b1a50c2ca349eee9c34c938

requirements.txt

@@ -10,7 +10,8 @@ pytest-benchmark
 requests
 tabulate
 git+https://github.com/huggingface/pytorch-image-models.git@730b907
-git+https://github.com/huggingface/transformers.git@6c26faa
+# this version of transformers is required as per this page https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1
+transformers==4.34.1
 MonkeyType
 psutil
 pyyaml

torchbenchmark/canary_models/mistral_7b_instruct/__init__.py

@@ -0,0 +1,19 @@
+from torchbenchmark.tasks import NLP
+from torchbenchmark.util.framework.huggingface.model_factory import HuggingFaceModel
+
+class Model(HuggingFaceModel):
+    task = NLP.LANGUAGE_MODELING
+    # DEFAULT_TRAIN_BSIZE not specified since we're not implementing a train test
+    # DEFAULT_TRAIN_BSIZE = 1
+    DEFAULT_EVAL_BSIZE = 1
+
+    def __init__(self, test, device, batch_size=None, extra_args=[]):
+        super().__init__(name="mistral_7b_instruct", test=test, device=device, batch_size=batch_size, extra_args=extra_args)
+
+    def train(self):
+        return NotImplementedError("Not implemented")
+
+    def eval(self):
+        if (self.device == "cpu"):
+            raise NotImplementedError("mistral_7b_instruct model is too slow on CPU - skip CPU test.")
+        super().eval()

torchbenchmark/canary_models/mistral_7b_instruct/install.py

@@ -0,0 +1,9 @@
+import subprocess
+import sys
+import os
+from torchbenchmark.util.framework.huggingface.patch_hf import patch_transformers, cache_model
+
+if __name__ == '__main__':
+    patch_transformers()
+    model_name = os.path.basename(os.path.dirname(os.path.abspath(__file__)))
+    cache_model(model_name)

torchbenchmark/canary_models/mistral_7b_instruct/metadata.yaml

@@ -0,0 +1,11 @@
+devices:
+  - device: NVIDIA A10G
+  - device: NVIDIA A100-SXM4-40GB
+eval_batch_size: 1
+eval_benchmark: false
+eval_deterministic: false
+eval_nograd: true
+train_benchmark: false
+train_deterministic: false
+not_implemented:
+  - device: cpu

torchbenchmark/models/phi_1_5/requirements.txt

@@ -1 +1,2 @@
-einops
+einops
+flash_attn

torchbenchmark/util/framework/huggingface/model_factory.py

@@ -36,6 +36,8 @@
     'llama_v2_13b' : (512,512, 'AutoConfig.from_pretrained("meta-llama/Llama-2-13b-hf")', 'AutoModelForCausalLM'),
     'llama_v2_70b' : (512, 512, 'AutoConfig.from_pretrained("meta-llama/Llama-2-70b-hf")', 'AutoModelForMaskedLM'),
     'phi_1_5' : (512, 512, 'AutoConfig.from_pretrained("microsoft/phi-1_5", trust_remote_code=True)', 'AutoModelForCausalLM'),
+    # as per this page https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1 trust_remote_code=True is not required
+    'mistral_7b_instruct' : (128, 128, 'AutoConfig.from_pretrained("mistralai/Mistral-7B-Instruct-v0.1")', 'AutoModelForCausalLM'),
     'hf_Yi' : (512, 512, 'AutoConfig.from_pretrained("01-ai/Yi-6B", trust_remote_code=True)', 'AutoModelForCausalLM'),
 
 }
@@ -89,8 +91,8 @@ def __init__(self, name, test, device, batch_size=None, extra_args=[]):
             config.num_buckets = 128
         class_ctor = getattr(transformers, class_models[name][3])
         kwargs = {}
-        remote_code_required = ['hf_Falcon_7b', 'hf_MPT_7b_instruct', 'phi_1_5', 'hf_Yi']
-        if name in remote_code_required:
+        hugging_face_models_requiring_trust_remote_code = ["hf_Falcon_7b", "hf_MPT_7b_instruct", "phi_1_5", "hf_Yi"]
+        if name in hugging_face_models_requiring_trust_remote_code:
             kwargs["trust_remote_code"] = True
         self.model = class_ctor.from_config(config, **kwargs).to(device)
         self.optimizer = optim.Adam(