EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[ModelRunner] Add hunyuan-vl basic support (#5151)

Browse Source 
### What this PR does / why we need it?
This patch add handling of `XDRotaryEmbedding` in modelrunner to support
for `hunyuan-vl`
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?
CI passed with added/exist tests

Closes: https://github.com/vllm-project/vllm-ascend/issues/4992

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: wangli <wangli858794774@gmail.com>
This commit is contained in:
Li Wang
2025-12-23 10:46:54 +08:00
committed by GitHub
parent c9b5881bcd
commit 9a79cbaecb
3 changed files with 63 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
vllm_ascend/worker/model_runner_v1.py
View File

@@ -654,15 +654,23 @@ class NPUModelRunner(GPUModelRunner):
else:
self.positions.np[:total_num_scheduled_tokens] = positions_np
# Calculate M-RoPE positions.
# Only relevant for models using M-RoPE (e.g, Qwen2-VL)
if self.uses_mrope:
self._calc_mrope_positions(scheduler_output)
# Only relevant for models using M-RoPE (e.g, Qwen2-VL)
self._calc_mrope_positions(scheduler_output)
self.mrope_positions.gpu[:, :total_num_scheduled_tokens].copy_(
self.mrope_positions.cpu[:, :total_num_scheduled_tokens],
non_blocking=True)
non_blocking=True,
)
elif self.uses_xdrope_dim > 0:
self._calc_xdrope_positions(scheduler_output)
# Only relevant for models using XD-RoPE (e.g, HunYuan-VL)
self.xdrope_positions.gpu[:, :total_num_scheduled_tokens].copy_(
self.xdrope_positions.cpu[:, :total_num_scheduled_tokens],
non_blocking=True,
)
else:
# Common case (1D positions)
self.positions.copy_to_gpu(total_num_scheduled_tokens)
# Get token indices.
# E.g., [0, 1, 0, 1, 2, 3, 4, 0, 1, 2]
@@ -845,9 +853,12 @@ class NPUModelRunner(GPUModelRunner):
# then the embedding layer is not included in the ACL graph.
input_ids = self.input_ids.gpu[:num_input_tokens]
inputs_embeds = None
positions = self.positions.gpu[:num_input_tokens]
if self.uses_mrope:
positions = self.mrope_positions.gpu[:, :num_input_tokens]
elif self.uses_xdrope_dim > 0:
positions = self.xdrope_positions.gpu[:, :num_input_tokens]
else:
positions = self.positions.gpu[:num_input_tokens]
# type: ignore
if get_pp_group().is_first_rank:
@@ -2070,6 +2081,8 @@ class NPUModelRunner(GPUModelRunner):
if self.uses_mrope:
positions = self.mrope_positions.gpu[:, :num_tokens_padded]
elif self.uses_xdrope_dim > 0:
positions = self.xdrope_positions.gpu[:, :num_tokens_padded]
else:
positions = self.positions.gpu[:num_tokens_padded]