Skip to content

small bamba test for parallel execution #142

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

Merged
merged 1 commit into from
Sep 25, 2025
Merged

small bamba test for parallel execution #142

Show file tree
Hide file tree
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
19 changes: 19 additions & 0 deletions tests/unittest/_torch/auto_deploy/_utils_test/_model_test_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -446,6 +446,25 @@ def apply_rotary_pos_emb_ds(q, k, cos, sin, position_ids, unsqueeze_dim=1):
"vision_config": {"num_hidden_layers": 2},
},
},
"ibm-ai-platform/Bamba-9B-v2": {
"llm_models_subdir": "NO_SUBDIR", # no synced to llm_models_root at the moment
"model_kwargs": {
"torch_dtype": "bfloat16",
"hidden_size": 64,
"intermediate_size": 128,
"mamba_chunk_size": 64,
"mamba_d_conv": 2,
"mamba_d_head": 16,
"mamba_d_state": 64,
"mamba_expand": 1,
"mamba_n_groups": 1,
"mamba_n_heads": 4,
"model_type": "bamba",
"num_hidden_layers": 10,
"num_attention_heads": 4,
"num_key_value_heads": 2,
},
},
}


Expand Down
108 changes: 38 additions & 70 deletions tests/unittest/_torch/auto_deploy/unit/singlegpu/models/test_bamba.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,10 +2,10 @@
import torch.export as te
from torch.export import Dim # noqa

# import pytest
from tensorrt_llm._torch.auto_deploy.export import apply_export_patches, torch_export_to_gm
from tensorrt_llm._torch.auto_deploy.llm_args import AutoDeployConfig
from tensorrt_llm._torch.auto_deploy.transformations._graph import move_to_device # noqa
from _model_test_utils import get_small_model_config

MODEL_DIR = "ibm-ai-platform/Bamba-9B-v2"
# NOTE: find example inputs with the same tokenization length to avoid seq concat.
Expand All @@ -14,43 +14,36 @@
EXAMPLE_INPUT2 = "Tiger is a cat with the following properties:"


# @pytest.mark.parametrize(
# "model_on_meta_during_export",
# [
# True,
# False,
# ],
# )
# @pytest.mark.parametrize(
# "export_func",
# [
# "torch_export_to_gm",
# "torch_export",
# ],
# )
def test_bamba_patches(
model_on_meta_during_export: bool = True,
export_func: str = "torch_export_to_gm",
use_cache: bool = True,
):
llm_args = AutoDeployConfig(
**{
def test_bamba_patches():
model_on_meta_during_export = True
export_func: str = "torch_export_to_gm"
use_cache: bool = True

# NOTE: set to False if you want to locally test the full model
use_small_config: bool = True

common_kwargs = {
"world_size": 0,
"runtime": "demollm",
"compile_backend": "torch-simple",
"attn_backend": "flashinfer",
"model_factory": "AutoModelForCausalLM",
"max_seq_len": 512,
}

if use_small_config:
llm_args = get_small_model_config(MODEL_DIR, **common_kwargs)["args"]
llm_args["model_kwargs"]["use_cache"] = use_cache
else:
llm_args = {
"model": MODEL_DIR,
"world_size": 0,
"runtime": "demollm",
"compile_backend": "torch-simple",
"attn_backend": "flashinfer",
"model_factory": "AutoModelForCausalLM",
**common_kwargs,
"model_kwargs": {
# "use_cache": True,
"use_cache": use_cache,
"torch_dtype": "bfloat16",
# "num_hidden_layers": 10,
},
"max_seq_len": 512,
"skip_loading_weights": False,
},
)
}
llm_args = AutoDeployConfig(**llm_args)

torch.manual_seed(0)
if torch.cuda.is_available():
Expand Down Expand Up @@ -85,8 +78,7 @@ def _run_torch_export_to_gm():
dynamic_shapes=dynamic_shapes,
patch_list=[
"bamba",
# For "unsupported scalarType".
"autocast_noop",
"autocast_noop", # For "unsupported scalarType".
],
)

Expand All @@ -112,18 +104,14 @@ def _run_export():
gm = _run_export()
factory.load_or_random_init(gm, device="cuda")
move_to_device(gm, "cuda")

factory.load_or_random_init(model, device="cuda")

_verify_generation(factory, model, tokenizer)
# return

print("====== EXPORTING GRAPH MODULE ======")
if not model_on_meta_during_export:
factory._to_maybe_random(model, "cuda")
model.load_state_dict(gm.state_dict())
else:
factory.load_or_random_init(model, device="cuda")
gm = _run_export()
move_to_device(gm, "cuda")

gm.model.A_log = model.model.A_log
_verify_generation(factory, model, tokenizer)

# let's do a comparison of every state dict item between the model and the gm
torch.testing.assert_close(model.state_dict(), gm.state_dict(), rtol=0.0, atol=0.0)
Expand All @@ -142,20 +130,12 @@ def _run_export():
atol, rtol = 1e-3, 1e-3
for comp, outs in outputs_for_comparison.items():
print(f"====== COMPARISON ({comp}) ======")
try:
torch.testing.assert_close(
outs,
out_original,
rtol=rtol,
atol=atol,
)
print("Passed!")
except AssertionError as e:
print(e)
diff = torch.abs(outs.logits - out_original.logits)
print(f"abs diff: {diff}")
print(f"average diff: {diff.mean()}")
print(f"{comp=}")
torch.testing.assert_close(
outs,
out_original,
rtol=rtol,
atol=atol,
)


def _verify_generation(factory, model, tokenizer):
Expand All @@ -181,17 +161,5 @@ def _generate(tokenizer, model):
print("\n".join(tokenizer.batch_decode(response, skip_special_tokens=True)))


# def _get_example_inputs(llm_args, factory, device):
# batch_size = min(2, llm_args.max_batch_size)
# seq_len = min(4, llm_args.max_seq_len)
# inputs = {"input_ids": torch.ones(batch_size, seq_len, dtype=torch.int)}
# for key, value in inputs.items():
# if isinstance(value, torch.Tensor):
# dtype = torch.bfloat16 if isinstance(value, torch.FloatTensor) else None
# inputs[key] = value.to(device=device, dtype=dtype)
#
# return inputs


if __name__ == "__main__":
test_bamba_patches()
Loading