Skip to content

Reduce eagle example test memory usage from 28 to 1 GB #299

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 3 commits into from
Sep 8, 2025
Merged

Reduce eagle example test memory usage from 28 to 1 GB #299

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
254d6b0
Reduce eagle example test memory usage from 28 to 1 GB
kevalmorabia97 Sep 8, 2025
bc531e8
Merge branch 'main' into kmorabia/eagle-test-mem
kevalmorabia97 Sep 8, 2025
984e30f
Merge branch 'main' into kmorabia/eagle-test-mem
kevalmorabia97 Sep 8, 2025
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
1 change: 1 addition & 0 deletions examples/speculative_decoding/launch.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -150,6 +150,7 @@ CMD="accelerate launch $MULTI_GPU --mixed_precision bf16 main.py \
--logging_steps 100 \
--tf32 True \
--data_path $DATA \
--report_to tensorboard \
$SPECULATIVE_ARGS
"

Expand Down
19 changes: 18 additions & 1 deletion tests/examples/speculative_decoding/test_eagle.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,12 +13,27 @@
# See the License for the specific language governing permissions and
# limitations under the License.

import json

from _test_utils.examples.run_command import run_example_command


# fmt: off
def test_llama_eagle(tiny_llama_path, num_gpus, tiny_daring_anteater_path, tmp_path):
def test_llama_eagle3(tiny_llama_path, num_gpus, tiny_daring_anteater_path, tmp_path):
# Create an ultra-tiny EAGLE config for testing to reduce memory usage
tiny_eagle_config = {
"max_position_embeddings": 128,
"num_hidden_layers": 1,
"intermediate_size": 64,
"num_attention_heads": 2,
"num_key_value_heads": 2,
}

# Write the tiny config to a temporary file
config_file = tmp_path / "tiny_eagle_config.json"
with open(config_file, "w") as f:
json.dump(tiny_eagle_config, f)

run_example_command(
[
"./launch.sh",
Expand All @@ -29,7 +44,9 @@ def test_llama_eagle(tiny_llama_path, num_gpus, tiny_daring_anteater_path, tmp_p
"--do_eval", "False",
"--num_gpu", str(num_gpus),
"--mode", "eagle3",
"--eagle_config", str(config_file),
"--output_dir", tmp_path / "eagle-tinyllama",
"--training_seq_len", "128", # Match max_position_embeddings
],
"speculative_decoding",
)
6 changes: 0 additions & 6 deletions tests/gpu/torch/export/test_unified_export_megatron.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -110,12 +110,6 @@ def _test_unified_export_megatron(tmp_path, model_type, arch, algo, rank, size):
],
)
def test_unified_export_megatron(tmp_path, model_type, arch, algo):
if algo == "eagle":
try:
import megatron.core.post_training # noqa: F401
except ImportError:
pytest.skip("megatron.core.post_training not found")

# TODO: Fix TP>1 failures
spawn_multiprocess_job(
size=1, # torch.cuda.device_count(),
Expand Down
Loading