Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

add support for apple silicon #3

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

add support for apple silicon #3

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 9 additions & 3 deletions mistral_example.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,9 @@
from transformers.models.mistral.modeling_mistral import MistralAttention
from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch

device = torch.device("mps" if torch.backends.mps.is_available() else "auto")

original_mistral_forward = MistralAttention.forward
self_extend_forward = partial(MistralSE.self_extend_forward, group_size_1=4, group_size_2=1024)
Expand All @@ -17,18 +20,18 @@
model_path = 'mistralai/Mistral-7B-Instruct-v0.1'
config = transformers.AutoConfig.from_pretrained(model_path)
config.sliding_window = 200000000 # disable mistral's default SWA mechanism (4096), mistral's true window is 8192.
model = AutoModelForCausalLM.from_pretrained(model_path, config=config, device_map="auto")
model = AutoModelForCausalLM.from_pretrained(model_path, config=config, device_map=device)


tokenizer = AutoTokenizer.from_pretrained(model_path)
model.eval()
model.eval().to(device)
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

device_map value "auto" can't be used in tensor.to(device) or model.eval().to(device). Valid device values are cpu, cuda, ipu, xpu, mkldnn, opengl, opencl, ideep, hip, ve, fpga, ort, xla, lazy, vulkan, mps, meta, hpu, mtia, privateuseone.


# In the example task file, the passkey is placed within the last 4096 tokens, this means, if you use SWA, mistral will successfully find the passkey.
for line in open("passkey_examples_10k.jsonl", "r"):
example = json.loads(line)
prompt_postfix = "What is the pass key? The pass key is "
prompt = example["input"] + prompt_postfix
input_ids = tokenizer(prompt, return_tensors="pt").input_ids
input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

device_map value "auto" can't be used in tensor.to(device) or model.eval().to(device). Valid device values are cpu, cuda, ipu, xpu, mkldnn, opengl, opencl, ideep, hip, ve, fpga, ort, xla, lazy, vulkan, mps, meta, hpu, mtia, privateuseone.

print( "-----------------------------------" )
print( f"#Tokens of Prompt:", input_ids.shape[1], end=" " )
print( "Passkey target:", example["target"] )
Expand All @@ -47,6 +50,9 @@
answer = answer.replace("\n", "\\n")
print( answer )
print( "-----------------------------------\n" )

if torch.backends.mps.is_available():
torch.mps.empty_cache()
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seems like a pytorch issue in pytorch/pytorch#105839 ?




Expand Down