Skip to content

[serge] Fix 8 integration tests for model mamba2 failing with OOM (other (8)) #47058

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

Open
sergereview wants to merge 1 commit into
base: main
Choose a base branch
from
+6 −4
Open

[serge] Fix 8 integration tests for model mamba2 failing with OOM (other (8)) #47058

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
10 changes: 6 additions & 4 deletions tests/models/mamba2/test_modeling_mamba2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -404,6 +404,8 @@ class Mamba2IntegrationTest(unittest.TestCase):
def setUp(self):
self.model_id = "mistralai/Mamba-Codestral-7B-v0.1"
self.tokenizer = AutoTokenizer.from_pretrained(self.model_id, from_slow=True, legacy=False)
if torch.cuda.is_available():
torch.cuda.empty_cache()
self.prompt = ("[INST]Write a hello world program in C++.",)

@slow
Expand All @@ -418,7 +420,7 @@ def test_simple_generate(self):
tokenizer = self.tokenizer
tokenizer.pad_token_id = tokenizer.eos_token_id

model = Mamba2ForCausalLM.from_pretrained(self.model_id, dtype=torch.bfloat16)
model = Mamba2ForCausalLM.from_pretrained(self.model_id, torch_dtype=torch.bfloat16)
model.to(torch_device)
input_ids = tokenizer("[INST]Write a hello world program in C++.[/INST]", return_tensors="pt")["input_ids"].to(
torch_device
Expand Down Expand Up @@ -450,7 +452,7 @@ def test_batched_equivalence_with_cache(self):
"[INST] Write a simple Fibonacci number computation function in Rust that does memoization, with comments, in safe Rust.[/INST]",
]

model = Mamba2ForCausalLM.from_pretrained(self.model_id, dtype=torch.bfloat16).to(torch_device)
model = Mamba2ForCausalLM.from_pretrained(self.model_id, torch_dtype=torch.bfloat16).to(torch_device)
tokenizer.pad_token_id = tokenizer.eos_token_id
# batched generation
tokenized_prompts = tokenizer(prompt, return_tensors="pt", padding="longest").to(torch_device)
Expand Down Expand Up @@ -480,7 +482,7 @@ def test_batched_equivalence_without_cache(self):
"[INST] Write a simple Fibonacci number computation function in Rust that does memoization, with comments, in safe Rust.[/INST]",
]

model = Mamba2ForCausalLM.from_pretrained(self.model_id, dtype=torch.bfloat16).to(torch_device)
model = Mamba2ForCausalLM.from_pretrained(self.model_id, torch_dtype=torch.bfloat16).to(torch_device)
tokenizer.pad_token_id = tokenizer.eos_token_id
# batched generation
tokenized_prompts = tokenizer(prompt, return_tensors="pt", padding="longest").to(torch_device)
Expand All @@ -501,7 +503,7 @@ def test_mamba2_mixer_train_vs_eval_equivalence(self):
# Based on https://github.com/sustcsonglin/flash-linear-attention/issues/63
# Credit to zhixuan-lin

B, T, D = 4, 512, 768
B, T, D = 4, 128, 768
dtype = torch.bfloat16
config = Mamba2Config(num_heads=24, head_dim=64, hidden_size=768, expand=2, n_groups=1)

Expand Down
Loading