[XPU] Fix Eagle3 initialization on XPU (#43957)
Signed-off-by: Chaojun Zhang <chaojun.zhang@intel.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
This commit is contained in:
co-authored by
Andreas Karatzas
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Kunshang Ji
parent
78a04c208d
commit
d2ec433e37
@@ -76,6 +76,26 @@ steps:
|
|||||||
pytest -v -s v1/sample/test_logprobs.py &&
|
pytest -v -s v1/sample/test_logprobs.py &&
|
||||||
pytest -v -s v1/sample/test_logprobs_e2e.py'
|
pytest -v -s v1/sample/test_logprobs_e2e.py'
|
||||||
|
|
||||||
|
- label: Basic Models Tests (Initialization)
|
||||||
|
timeout_in_minutes: 60
|
||||||
|
device: intel_gpu
|
||||||
|
no_plugin: true
|
||||||
|
working_dir: "."
|
||||||
|
env:
|
||||||
|
REGISTRY: "public.ecr.aws/q9t5s3a7"
|
||||||
|
REPO: "vllm-ci-test-repo"
|
||||||
|
VLLM_TEST_DEVICE: "xpu"
|
||||||
|
source_file_dependencies:
|
||||||
|
- vllm/
|
||||||
|
- tests/models/test_initialization.py
|
||||||
|
- tests/models/registry.py
|
||||||
|
commands:
|
||||||
|
- >-
|
||||||
|
bash .buildkite/scripts/hardware_ci/run-intel-test.sh
|
||||||
|
'export VLLM_XPU_FUSED_MOE_USE_REF=1 &&
|
||||||
|
cd tests &&
|
||||||
|
pytest -v -s models/test_initialization.py::test_can_initialize_large_subset[Eagle3MiniMaxM2ForCausalLM]'
|
||||||
|
|
||||||
- label: XPU CPU Offload
|
- label: XPU CPU Offload
|
||||||
timeout_in_minutes: 60
|
timeout_in_minutes: 60
|
||||||
device: intel_gpu
|
device: intel_gpu
|
||||||
|
|||||||
@@ -233,7 +233,10 @@ class LlamaModel(nn.Module):
|
|||||||
) -> tuple[torch.Tensor, torch.Tensor]:
|
) -> tuple[torch.Tensor, torch.Tensor]:
|
||||||
if input_embeds is None:
|
if input_embeds is None:
|
||||||
input_embeds = self.embed_input_ids(input_ids)
|
input_embeds = self.embed_input_ids(input_ids)
|
||||||
assert hidden_states.shape[-1] == input_embeds.shape[-1]
|
torch._assert(
|
||||||
|
hidden_states.shape[-1] == input_embeds.shape[-1],
|
||||||
|
"hidden_states and input_embeds must have the same last dimension",
|
||||||
|
)
|
||||||
|
|
||||||
residual = None
|
residual = None
|
||||||
for layer in self.layers:
|
for layer in self.layers:
|
||||||
|
|||||||
@@ -1468,6 +1468,26 @@ class SpecDecodeBaseProposer:
|
|||||||
"Sharing target model embedding weights with the draft model."
|
"Sharing target model embedding weights with the draft model."
|
||||||
)
|
)
|
||||||
|
|
||||||
|
if share_embeddings:
|
||||||
|
draft_embed = self.model.model.embed_tokens
|
||||||
|
# Only share when both models use the same embedding width.
|
||||||
|
# Guard with isinstance so non-Tensor weights (e.g. in tests)
|
||||||
|
# are not affected — mirrors the weight-equality check above.
|
||||||
|
if isinstance(target_embed_tokens.weight, torch.Tensor) and isinstance(
|
||||||
|
draft_embed.weight, torch.Tensor
|
||||||
|
):
|
||||||
|
target_dim = target_embed_tokens.weight.shape[-1]
|
||||||
|
draft_dim = draft_embed.weight.shape[-1]
|
||||||
|
if target_dim != draft_dim:
|
||||||
|
share_embeddings = False
|
||||||
|
logger.info(
|
||||||
|
"Target embedding dim (%d) differs from draft "
|
||||||
|
"embedding dim (%d). Keeping separate embedding "
|
||||||
|
"weights.",
|
||||||
|
target_dim,
|
||||||
|
draft_dim,
|
||||||
|
)
|
||||||
|
|
||||||
if share_embeddings:
|
if share_embeddings:
|
||||||
if hasattr(self.model.model, "embed_tokens"):
|
if hasattr(self.model.model, "embed_tokens"):
|
||||||
del self.model.model.embed_tokens
|
del self.model.model.embed_tokens
|
||||||
|
|||||||
Reference in New Issue
Block a user