Skip to content

chore: upgrade vLLM to 0.18.0 and torch/unsloth to equivalent versions #342

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
mckornfield wants to merge 1 commit into
base: main
Choose a base branch
from
+510 −388
Open

chore: upgrade vLLM to 0.18.0 and torch/unsloth to equivalent versions #342

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
32 changes: 16 additions & 16 deletions pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -127,17 +127,17 @@ cpu = [
"peft",
"opacus",
"sentence-transformers",
"torch==2.9.1; sys_platform == 'darwin'",
"torch==2.9.1+cpu; sys_platform == 'linux'",
"torchvision==0.24.1; sys_platform == 'darwin'",
"torchvision==0.24.1+cpu; sys_platform == 'linux'",
"torchao==0.15.0",
"torch==2.10.0; sys_platform == 'darwin'",
"torch==2.10.0+cpu; sys_platform == 'linux'",
"torchvision==0.25.0; sys_platform == 'darwin'",
"torchvision==0.25.0+cpu; sys_platform == 'linux'",
"torchao==0.16.0",
"transformers==4.57.3",
"triton>=2.0.0; sys_platform=='linux'",
"trl>=0.23.0",
"unsloth[cu128-torch291]==2025.12.4; sys_platform=='linux'",
"unsloth_zoo==2025.12.4; sys_platform=='linux'",
"vllm==0.15.0; sys_platform=='linux'",
"unsloth[cu128-torch2100]==2026.3.18; sys_platform=='linux'",
"unsloth_zoo==2026.3.7; sys_platform=='linux'",
"vllm==0.18.0; sys_platform=='linux'",
]

cu128 = [
Expand All @@ -153,17 +153,17 @@ cu128 = [
"opacus",
"peft",
"sentence-transformers",
"torch==2.9.1+cu128; sys_platform == 'linux'",
"torch==2.10.0+cu128; sys_platform == 'linux'",
"torch-c-dlpack-ext",
"torchvision==0.24.1+cu128; sys_platform == 'linux'",
"torchao==0.15.0; sys_platform == 'linux'",
"torchvision==0.25.0+cu128; sys_platform == 'linux'",
"torchao==0.16.0; sys_platform == 'linux'",
"transformers==4.57.3",
"triton>=2.0.0; sys_platform == 'linux'",
"trl>=0.23.0",
"unsloth[cu128-torch291]==2025.12.4; sys_platform == 'linux'",
"unsloth_zoo==2025.12.4; sys_platform == 'linux'",
"vllm==0.15.0; sys_platform == 'linux'",
"xformers==v0.0.33.post2; sys_platform == 'linux'",
"unsloth[cu128-torch2100]==2026.3.18; sys_platform == 'linux'",
"unsloth_zoo==2026.3.7; sys_platform == 'linux'",
"vllm==0.18.0; sys_platform == 'linux'",
"xformers==v0.0.34; sys_platform == 'linux'",
]

# at some point, do per-subpackage dependencies
Expand Down Expand Up @@ -204,7 +204,7 @@ environments = [
"sys_platform == 'linux' and platform_machine == 'aarch64'",
]

constraint-dependencies = ["torch==2.9.1", "regex==2025.07.34", "pandas<3"]
constraint-dependencies = ["torch==2.10.0", "regex==2025.07.34", "pandas<3"]



Expand Down
5 changes: 1 addition & 4 deletions src/nemo_safe_synthesizer/generation/vllm_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -137,7 +137,6 @@ def __init__(
prompt_template=self.model_metadata.prompt_config.template,
)
self.llm: vLLM | None = None
self.logits_processors = []

# Do not generate detailed error messages in production to avoid leaking sensitive data.
self.use_detailed_logs = kwargs.pop("use_detailed_logs", False)
Expand Down Expand Up @@ -191,7 +190,6 @@ def initialize(self, **kwargs) -> None:
# vllm requires this "config" to set the backend ahead of time.
structured_outputs_config = StructuredOutputsConfig(
backend=self.config.generation.structured_generation_backend,
disable_fallback=True,
)
# Unsloth patches model attention forward functions with torch.compiler.disable().
# vLLM compiles TransformersForCausalLM with fullgraph=True via @support_torch_compile.
Expand Down Expand Up @@ -222,7 +220,7 @@ def _build_structured_output_params(self) -> StructuredOutputsParams | None:
if not self.config.generation.use_structured_generation:
return None

params: dict[str, Any] = {"disable_fallback": True}
params: dict[str, Any] = {}

if self.config.generation.structured_generation_schema_method == "regex":
logger.info("Structured generation is enabled, using a regex to enforce the schema")
Expand Down Expand Up @@ -511,7 +509,6 @@ def generate(
top_p=self.config.generation.top_p,
top_k=FIXED_RUNTIME_GENERATE_ARGS["top_k"],
min_p=FIXED_RUNTIME_GENERATE_ARGS["min_p"],
logits_processors=self.logits_processors,
max_tokens=self.model_metadata.max_seq_length,
skip_special_tokens=not need_special_token_outputs,
include_stop_str_in_output=need_special_token_outputs,
Expand Down
2 changes: 1 addition & 1 deletion tests/e2e/test_safe_synthesizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,7 +85,7 @@ def test_train_and_generate_dp(fixture_financial_transactions_dataset, fixture_s

@pytest.mark.e2e
@pytest.mark.requires_gpu
@pytest.mark.timeout(500)
@pytest.mark.timeout(900)
@pytest.mark.skipif(sys.platform == "darwin", reason="Not applicable on macOS")
def test_train_and_generate_defaults(fixture_financial_transactions_dataset, fixture_save_path):
df = fixture_financial_transactions_dataset
Expand Down
Loading
Loading