v0.7.3 support speculative decoding #252
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
Signed-off-by: mengwei805 <mengwei25@huawei.com>
Signed-off-by: mengwei805 <mengwei25@huawei.com>
Signed-off-by: mengwei805 <mengwei25@huawei.com>
|
|
wangxiyuan
reviewed
Mar 7, 2025
| logger = init_logger(__name__) | ||
|
|
||
|
|
||
| def create_worker( |
There was a problem hiding this comment.
There was a problem hiding this comment.
The adaptation work here has been completed in #236
wangxiyuan
approved these changes
Mar 7, 2025
wangxiyuan
reviewed
Mar 10, 2025
| # the use of TP1DraftModelRunner | ||
| if draft_tp == 1 and draft_model_config.hf_config.model_type !=\ | ||
| "deepseek_mtp": | ||
| draft_worker_kwargs["model_runner_cls"] = TP1DraftModelRunner |
There was a problem hiding this comment.
So the patch code here does two change mainly:
- remove
is_cuda_likehard code - change draft_worker to a new
TP1DraftModelRunnerin vllm-ascend
right?
There was a problem hiding this comment.
- remove
is_cuda_likewhen importing the package and here, because the mode of using EAGLE based draft models must useTP1DraftModelRunner; - change draft_worker to a new
TP1DraftModelRunnerin vllm-ascend; - When the draft_model type is
deepseek_mtp, forceTP1DraftModelRunnernot to be used. Refer to [Model][Speculative Decoding] Expand DeepSeek MTP code to support k > n_predict vllm#13626 and [Model][Speculative Decoding] DeepSeek MTP spec decode vllm#12755. I think this version can be done this way, because fixing this problem may require a lot of work(may need to adaptmulti_step_worker, modifynpu_worker, etc.), we can do better in the next version.
In summary, I have verified that the current modification can correctly run 4+1 speculative decoding modes: The 4 is speculating with a draft model、by matching n-grams in the prompt、using MLP speculators and using EAGLE based draft models, and The 1 is deepseek_mtp.
This was referenced Apr 9, 2025
wangxiyuan
pushed a commit
that referenced
this pull request
Apr 17, 2025
### What this PR does / why we need it? Backport: #252 This support speculative decoding in Ascend, including speculating with a draft model、by matching n-grams in the prompt、using MLP speculators and using EAGLE based draft models. Backport: #423 spec decode MultiStepWorker support TP1DraftModelRunner fully, support run the draft_model_runner with multi-step prepare on the NPU directly and support draft_model_runner use MLA. 1. before this pr, `MultiStepWorker` would not step into the branch using NPU prepare, but only into the branch using CPU prepare (`line 52` of `vllm_ascend/patch/patch_multi_step_worker.py`). Although this has `no effect` on the `correct operation` of speculative decoding and the performance of the two branches is basically the same as of the current version, I support entering this branch in this PR. In general, there are two main changes in `patch_multi_step_worker.py`: first, the `is_cuda_like()` check is removed and the `TP1DraftModelRunner` rewritten in vllm_ascend is used; second, the `supports_gpu_multi_step()` function is made to return true on NPU devices when outer Multi_step_worker could work correct. 3. before this pr, `TP1DraftModelRunner` only supports Attention on NPU, but not MLA. The relevant adaptation is in `vllm_ascend/worker/draft_model_runner.py`. Although I don’t know why the `input_positions` of `model_input.attn_metadata` in vllm-ascend needs to be added in `execute_model`, it is done in `model_runner.py`, so I also made corresponding changes. Otherwise, when atten_backend is MLA, it will prompt that input_positions cannot be found. 4. I commented out two lines in `draft_model_runner.py` in `line118` to support the scenario of K>1. ``` # lora_mapping=model_input.lora_mapping, # lora_requests=model_input.lora_requests, ``` I added comments. In the future, when vllm-ascend supports lora feature, the changes here can be restored. TODO： - [ ] revert the patch when the related issues are addressed in vllm ### How was this patch tested? CI passed with new added test. - e2e test for medusa proposer: tests/singlecard/spec_decode/e2e/test_medusa_correctness.py - e2e test for mlp proposer: tests/singlecard/spec_decode/e2e/test_mlp_correctness.py - e2e test for n-gram proposer: tests/singlecard/spec_decode/e2e/test_ngram_correctness.py Tests for patched files: - tests/singlecard/spec_decode/test_dynamic_spec_decode.py - tests/singlecard/spec_decode/test_multi_step_worker.py - tests/singlecard/spec_decode/test_ngram_worker.py - tests/singlecard/spec_decode/test_spec_decode_worker.py --------- Signed-off-by: MengqingCao <cmq0113@163.com> Co-authored-by: mengwei805 <mengwei25@huawei.com>
ttanzhiqiang
pushed a commit
to ttanzhiqiang/vllm-ascend
that referenced
this pull request
Apr 27, 2025
### What this PR does / why we need it? Backport: vllm-project#252 This support speculative decoding in Ascend, including speculating with a draft model、by matching n-grams in the prompt、using MLP speculators and using EAGLE based draft models. Backport: vllm-project#423 spec decode MultiStepWorker support TP1DraftModelRunner fully, support run the draft_model_runner with multi-step prepare on the NPU directly and support draft_model_runner use MLA. 1. before this pr, `MultiStepWorker` would not step into the branch using NPU prepare, but only into the branch using CPU prepare (`line 52` of `vllm_ascend/patch/patch_multi_step_worker.py`). Although this has `no effect` on the `correct operation` of speculative decoding and the performance of the two branches is basically the same as of the current version, I support entering this branch in this PR. In general, there are two main changes in `patch_multi_step_worker.py`: first, the `is_cuda_like()` check is removed and the `TP1DraftModelRunner` rewritten in vllm_ascend is used; second, the `supports_gpu_multi_step()` function is made to return true on NPU devices when outer Multi_step_worker could work correct. 3. before this pr, `TP1DraftModelRunner` only supports Attention on NPU, but not MLA. The relevant adaptation is in `vllm_ascend/worker/draft_model_runner.py`. Although I don’t know why the `input_positions` of `model_input.attn_metadata` in vllm-ascend needs to be added in `execute_model`, it is done in `model_runner.py`, so I also made corresponding changes. Otherwise, when atten_backend is MLA, it will prompt that input_positions cannot be found. 4. I commented out two lines in `draft_model_runner.py` in `line118` to support the scenario of K>1. ``` # lora_mapping=model_input.lora_mapping, # lora_requests=model_input.lora_requests, ``` I added comments. In the future, when vllm-ascend supports lora feature, the changes here can be restored. TODO： - [ ] revert the patch when the related issues are addressed in vllm ### How was this patch tested? CI passed with new added test. - e2e test for medusa proposer: tests/singlecard/spec_decode/e2e/test_medusa_correctness.py - e2e test for mlp proposer: tests/singlecard/spec_decode/e2e/test_mlp_correctness.py - e2e test for n-gram proposer: tests/singlecard/spec_decode/e2e/test_ngram_correctness.py Tests for patched files: - tests/singlecard/spec_decode/test_dynamic_spec_decode.py - tests/singlecard/spec_decode/test_multi_step_worker.py - tests/singlecard/spec_decode/test_ngram_worker.py - tests/singlecard/spec_decode/test_spec_decode_worker.py --------- Signed-off-by: MengqingCao <cmq0113@163.com> Co-authored-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? Backport: vllm-project#252 This support speculative decoding in Ascend, including speculating with a draft model、by matching n-grams in the prompt、using MLP speculators and using EAGLE based draft models. Backport: vllm-project#423 spec decode MultiStepWorker support TP1DraftModelRunner fully, support run the draft_model_runner with multi-step prepare on the NPU directly and support draft_model_runner use MLA. 1. before this pr, `MultiStepWorker` would not step into the branch using NPU prepare, but only into the branch using CPU prepare (`line 52` of `vllm_ascend/patch/patch_multi_step_worker.py`). Although this has `no effect` on the `correct operation` of speculative decoding and the performance of the two branches is basically the same as of the current version, I support entering this branch in this PR. In general, there are two main changes in `patch_multi_step_worker.py`: first, the `is_cuda_like()` check is removed and the `TP1DraftModelRunner` rewritten in vllm_ascend is used; second, the `supports_gpu_multi_step()` function is made to return true on NPU devices when outer Multi_step_worker could work correct. 3. before this pr, `TP1DraftModelRunner` only supports Attention on NPU, but not MLA. The relevant adaptation is in `vllm_ascend/worker/draft_model_runner.py`. Although I don’t know why the `input_positions` of `model_input.attn_metadata` in vllm-ascend needs to be added in `execute_model`, it is done in `model_runner.py`, so I also made corresponding changes. Otherwise, when atten_backend is MLA, it will prompt that input_positions cannot be found. 4. I commented out two lines in `draft_model_runner.py` in `line118` to support the scenario of K>1. ``` # lora_mapping=model_input.lora_mapping, # lora_requests=model_input.lora_requests, ``` I added comments. In the future, when vllm-ascend supports lora feature, the changes here can be restored. TODO： - [ ] revert the patch when the related issues are addressed in vllm ### How was this patch tested? CI passed with new added test. - e2e test for medusa proposer: tests/singlecard/spec_decode/e2e/test_medusa_correctness.py - e2e test for mlp proposer: tests/singlecard/spec_decode/e2e/test_mlp_correctness.py - e2e test for n-gram proposer: tests/singlecard/spec_decode/e2e/test_ngram_correctness.py Tests for patched files: - tests/singlecard/spec_decode/test_dynamic_spec_decode.py - tests/singlecard/spec_decode/test_multi_step_worker.py - tests/singlecard/spec_decode/test_ngram_worker.py - tests/singlecard/spec_decode/test_spec_decode_worker.py --------- Signed-off-by: MengqingCao <cmq0113@163.com> Co-authored-by: mengwei805 <mengwei25@huawei.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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?
support speculative decoding in Ascend, including speculating with a draft model、by matching n-grams in the prompt、using MLP speculators and using EAGLE based draft models.
Does this PR introduce any user-facing change?
u can refer to https://docs.vllm.ai/en/latest/features/spec_decode.html#
How was this patch tested?
Four modes of speculative decoding have been tested, consistent with GPU devices