[BUGFIX] main-sd-bugfix && [UT] add mtp UT #593
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
mengwei805
force-pushed
the
main-sd-bugfix
branch
3 times, most recently
from
46b8b8f to
f9c61ea
Compare
mengwei805
changed the title
mengwei805
force-pushed
the
main-sd-bugfix
branch
from
f9c61ea to
1256fd9
Compare
MengqingCao
reviewed
Apr 21, 2025
| run: | | ||
| if [[ "${{ matrix.os }}" == "linux-arm64-npu-1" ]]; then | ||
| pytest -sv tests/singlecard/spec_decode | ||
| pytest -sv tests/singlecard/spec_decode/e2e/test_mtp_correctness.py # it needs a clean process |
There was a problem hiding this comment.
let's make more comments on why it needs a clean process
There was a problem hiding this comment.
i will add more comments
- i see a lot of comments
it needs a clean processin vLLM test_pipeline, so i add it here, i think vLLM UT arch maybe have some problem? - in this case, it needs a clean process becase it use bf16 and other UTs use fp16, if they are in a common process, it will failed.
There was a problem hiding this comment.
Yeah, I think 2 is the main reason in our case.
| expert_tensor_parallel_size) | ||
|
|
||
| global _EP | ||
| assert _EP is None, ("expert parallel group is already initialized") |
There was a problem hiding this comment.
why do we delete this assertion here?
There was a problem hiding this comment.
As spec decode worker would use init_worker twice, Second init _EP is not None
| group_ranks = [] | ||
| global _ETP | ||
| assert _ETP is None, ( | ||
| "expert tensor parallel group is already initialized") |
There was a problem hiding this comment.
As spec decode worker would use init_worker twice, Second init _EP is not None
MengqingCao
reviewed
Apr 21, 2025
| from vllm.model_executor.parameter import PerTensorScaleParameter | ||
| from vllm.model_executor.utils import set_weight_attrs | ||
|
|
||
| from ..ops.fused_moe import AscendUnquantizedFusedMoEMethod |
There was a problem hiding this comment.
I prefer to use absolute path references here
Suggested change
| from ..ops.fused_moe import AscendUnquantizedFusedMoEMethod | |
| from vllm_ascend.ops.fused_moe import AscendUnquantizedFusedMoEMethod |
There was a problem hiding this comment.
ok. i have modified it
mengwei805
force-pushed
the
main-sd-bugfix
branch
from
1256fd9 to
3b35c24
Compare
Signed-off-by: mengwei805 <mengwei25@huawei.com>
mengwei805
force-pushed
the
main-sd-bugfix
branch
from
3b35c24 to
523db6b
Compare
wangxiyuan
approved these changes
Apr 21, 2025
| if self.is_layer_skipped_ascend(prefix, | ||
| self.packed_modules_mapping): | ||
| return UnquantizedFusedMoEMethod() | ||
| return AscendUnquantizedFusedMoEMethod() |
There was a problem hiding this comment.
I wonder why it's not raised error before. @ganyi1996ppo
There was a problem hiding this comment.
if weights only has main model, all fusedmoe weight is w8a8,
but current mtp weights are float in deepseek w8a8 weights,
so when enable mtp, it will raise error, and it will run success if not enable mtp
83 tasks
ttanzhiqiang
pushed a commit
to ttanzhiqiang/vllm-ascend
that referenced
this pull request
Apr 27, 2025
### What this PR does / why we need it? The pr will fix some bug about spec decode / MTP The pr add a mtp e2e UT `test_mtp_correctness.py` **vllm_ascend/attention/attention.py** 1. add support `self.attn_mask_cache` only has 1 element to cover scene in which both spec docode and chunked prefill are enabled. **vllm_ascend/distributed/parallel_state.py** 1. remove 2 assert because spec decode worker would use init_worker twice **vllm_ascend/models/deepseek_mtp.py** 1. remove unused params; 2. add support w8a8 in `CustomDeepSeekMTP` **vllm_ascend/quantization/quant_config.py** 1. use `AscendUnquantizedFusedMoEMethod` instead of `UnquantizedFusedMoEMethod` **other** 1. replace `from vllm.logger import init_logger` to `from vllm.logger import logger` all of the vllm-ascend project ### Does this PR introduce _any_ user-facing change? ### How was this patch tested? Signed-off-by: mengwei805 <mengwei25@huawei.com>
Angazenn
pushed a commit
to Angazenn/vllm-ascend
that referenced
this pull request
Oct 21, 2025
### What this PR does / why we need it? The pr will fix some bug about spec decode / MTP The pr add a mtp e2e UT `test_mtp_correctness.py` **vllm_ascend/attention/attention.py** 1. add support `self.attn_mask_cache` only has 1 element to cover scene in which both spec docode and chunked prefill are enabled. **vllm_ascend/distributed/parallel_state.py** 1. remove 2 assert because spec decode worker would use init_worker twice **vllm_ascend/models/deepseek_mtp.py** 1. remove unused params; 2. add support w8a8 in `CustomDeepSeekMTP` **vllm_ascend/quantization/quant_config.py** 1. use `AscendUnquantizedFusedMoEMethod` instead of `UnquantizedFusedMoEMethod` **other** 1. replace `from vllm.logger import init_logger` to `from vllm.logger import logger` all of the vllm-ascend project ### Does this PR introduce _any_ user-facing change? ### How was this patch tested? Signed-off-by: mengwei805 <mengwei25@huawei.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
What this PR does / why we need it?
The pr will fix some bug about spec decode / MTP
The pr add a mtp e2e UT
test_mtp_correctness.pyvllm_ascend/attention/attention.py
self.attn_mask_cacheonly has 1 element to cover scene in which both spec docode and chunked prefill are enabled.vllm_ascend/distributed/parallel_state.py
vllm_ascend/models/deepseek_mtp.py
CustomDeepSeekMTPvllm_ascend/quantization/quant_config.py
AscendUnquantizedFusedMoEMethodinstead ofUnquantizedFusedMoEMethodother
from vllm.logger import init_loggertofrom vllm.logger import loggerall of the vllm-ascend projectDoes this PR introduce any user-facing change?
How was this patch tested?