Skip to content

support for vllm 0.19.1 #3613

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

Closed
winglian wants to merge 2 commits into from
Closed

support for vllm 0.19.1 #3613

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d248242
support for vllm 0.19.1
winglian Apr 19, 2026
cec99c4
fix test dims
winglian Apr 21, 2026
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
2 changes: 1 addition & 1 deletion setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,7 +89,7 @@ def parse_requirements(extras_require_map):
]
if not install_xformers:
_install_requires.pop(_install_requires.index(xformers_version))
extras_require_map["vllm"] = ["vllm>=0.19.0"]
extras_require_map["vllm"] = ["vllm>=0.19.1"]
elif (major, minor) >= (2, 9):
extras_require_map.pop("fbgemm-gpu")
extras_require_map["fbgemm-gpu"] = [
Expand Down
36 changes: 19 additions & 17 deletions tests/e2e/integrations/test_scattermoe_lora_olmoe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -422,22 +422,22 @@ def test_peft_creates_correct_shapes(self):
)

# gate_up_proj [E, 2*inter, hidden]
# peft: in_features=2*inter (dim 1), out_features=hidden (dim 2)
# peft: in_features=hidden (last dim), out_features=2*inter (middle dim)
assert trainable[
"base_model.model.moe.experts.base_layer.lora_A.default.weight"
].shape == (E * r, 2 * config.intermediate_size)
].shape == (E * r, config.hidden_size)
assert trainable[
"base_model.model.moe.experts.base_layer.lora_B.default.weight"
].shape == (config.hidden_size, E * r)
].shape == (2 * config.intermediate_size, E * r)

# down_proj [E, hidden, inter]
# peft: in_features=hidden (dim 1), out_features=inter (dim 2)
# peft: in_features=inter (last dim), out_features=hidden (middle dim)
assert trainable[
"base_model.model.moe.experts.lora_A.default.weight"
].shape == (E * r, config.hidden_size)
].shape == (E * r, config.intermediate_size)
assert trainable[
"base_model.model.moe.experts.lora_B.default.weight"
].shape == (config.intermediate_size, E * r)
].shape == (config.hidden_size, E * r)

@requires_cuda
def test_peft_forward_runs(self):
Expand Down Expand Up @@ -489,26 +489,28 @@ def test_unwrap_experts_lora(self):
assert down_lora is not None, "down_proj LoRA not detected"

# Check shapes (after peft->scattermoe conversion with A<->B swap)
# gate_up_proj W = param.T = [E, hidden, 2*inter], K=hidden, N=2*inter
# gate_up_proj: peft A [E*r, hidden] / B [2*inter, E*r]
# After swap: smoe_A [E*r, 2*inter], smoe_B [hidden, E*r]
E, r = config.num_experts, 4
gup_A, gup_B, gup_s = gup_lora
assert gup_A.shape == (E * r, config.hidden_size), (
f"gate_up_proj smoe_A: expected [r*E, K=hidden]={(E * r, config.hidden_size)}, "
assert gup_A.shape == (E * r, 2 * config.intermediate_size), (
f"gate_up_proj smoe_A: expected [r*E, 2*inter]={(E * r, 2 * config.intermediate_size)}, "
f"got {gup_A.shape}"
)
assert gup_B.shape == (2 * config.intermediate_size, E * r), (
f"gate_up_proj smoe_B: expected [N=2*inter, r*E]="
f"{(2 * config.intermediate_size, E * r)}, got {gup_B.shape}"
assert gup_B.shape == (config.hidden_size, E * r), (
f"gate_up_proj smoe_B: expected [hidden, r*E]="
f"{(config.hidden_size, E * r)}, got {gup_B.shape}"
)

# down_proj W = param.T = [E, inter, hidden], K=inter, N=hidden
# down_proj: peft A [E*r, inter] / B [hidden, E*r]
# After swap: smoe_A [E*r, hidden], smoe_B [inter, E*r]
down_A, down_B, down_s = down_lora
assert down_A.shape == (E * r, config.intermediate_size), (
f"down_proj smoe_A: expected [r*E, K=inter]={(E * r, config.intermediate_size)}, "
assert down_A.shape == (E * r, config.hidden_size), (
f"down_proj smoe_A: expected [r*E, hidden]={(E * r, config.hidden_size)}, "
f"got {down_A.shape}"
)
assert down_B.shape == (config.hidden_size, E * r), (
f"down_proj smoe_B: expected [N=hidden, r*E]={(config.hidden_size, E * r)}, "
assert down_B.shape == (config.intermediate_size, E * r), (
f"down_proj smoe_B: expected [inter, r*E]={(config.intermediate_size, E * r)}, "
f"got {down_B.shape}"
)

Expand Down
Loading