[XPU] support XPU VL model inference #4030
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thanks for your contribution! |
mayang002
reviewed
Sep 9, 2025
mayang002
reviewed
Sep 9, 2025
| int64_t hidden_size, | ||
| bool is_scatter); | ||
| } // namespace plugin | ||
| } // namespace xpu2 |
fastdeploy/worker/xpu_worker.py
Outdated
| used_memory = xpu_get_used_global_memory(int(self.device_ids[self.local_rank])) | ||
| available_kv_cache_memory = total_available_memory - used_memory | ||
| # 3. Statistical memory information | ||
| paddle_reserved_mem_after_run = paddle.device.xpu.max_memory_reserved(local_rank) |
Collaborator
There was a problem hiding this comment.
local_rank 不是 device_id 吧？是不是应该通过 self.device_ids[self.local_rank] 获取？
Contributor
Author
There was a problem hiding this comment.
对于 4567卡场景确实会有问题，我改一下
cqulilujia
force-pushed
the
xpu-vl
branch
2 times, most recently
from
2675a6d to
d0e3f1f
Compare
ming1753
reviewed
Sep 12, 2025
| # TODO: set condition to new _get_num_new_tokens | ||
| num_new_tokens = request.need_prefill_tokens - request.num_computed_tokens | ||
| num_new_tokens = min(num_new_tokens, token_budget) | ||
| request.with_image = False |
Collaborator
There was a problem hiding this comment.
166行已经在多模下默认为False了，这里重复了
Contributor
Author
There was a problem hiding this comment.
这里如果是文本请求，with_image不是request的成员变量，在模型里访问会挂掉，之前遗留的一个bug
ming1753
reviewed
Sep 12, 2025
| from fastdeploy.model_executor.ops.xpu import ( | ||
| text_image_gather_scatter, | ||
| text_image_index_out, | ||
| ) |
Collaborator
There was a problem hiding this comment.
这里最好参考其他算子做下封装，定义一个op类，在类内部区分不同硬件的实现
| image_features: Optional[paddle.Tensor], | ||
| forward_meta: ForwardMeta, | ||
| ): | ||
| vl_moe_meta = self.ernie.prepare_vl_moe_meta(ids_remove_padding=ids_remove_padding) |
Collaborator
There was a problem hiding this comment.
这里的改动建议让何泓域review下，可能会对cuda_graph的支持有影响
| ) | ||
|
|
||
| assert self.device_ids[self.local_rank] is not None, f"device_id is none for rank {self.local_rank}" | ||
| assert ( |
Collaborator
There was a problem hiding this comment.
这个部分的 assert 是不是应该加上
cqulilujia
force-pushed
the
xpu-vl
branch
5 times, most recently
from
4a5e580 to
0471e4f
Compare
aquagull
reviewed
Sep 17, 2025
| ids_remove_padding: paddle.Tensor, | ||
| image_token_num: int, | ||
| image_features: Optional[paddle.Tensor] = None, | ||
| vl_moe_meta: Optional[VLMoEMeta] = None, |
Contributor
|
LGTM for vl model cudagraph |
Contributor
Author
|
/re-run all-failed |
cqulilujia
force-pushed
the
xpu-vl
branch
4 times, most recently
from
eac75f3 to
4d71517
Compare
Collaborator
|
@cqulilujia 请确保 paddle 相关PR 合入到 3.2.1，否则会影响后续 FD 的发版。 |
heavengate
approved these changes
Sep 25, 2025
DDDivano
approved these changes
Sep 25, 2025
phlrain
approved these changes
Sep 25, 2025
Jiang-Jia-Jun
approved these changes
Sep 25, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
10 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
XPU平台支持Ernie4.5T-VL模型