[NPUW] Enabled model with multiple outputs in LLMInferRequest #31520
+59
−41
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
github-actions
bot
added
category: NPU
AsyaPronina
changed the title
AsyaPronina
changed the title
AsyaPronina
force-pushed
the
fix_multi_outputs_issue
branch
from
28d3e37 to
b5a0806
Compare
AsyaPronina
force-pushed
the
fix_multi_outputs_issue
branch
from
b5a0806 to
11ab2cb
Compare
| // second and third and combine them using XOR | ||
| // and bit shifting: | ||
|
|
||
| return ((hash<std::size_t>()(port.get_index()) ^ (hash<const ov::Node*>()(port.get_node()) << 1)) >> 1) ^ |
There was a problem hiding this comment.
Wouldn't it be enough just to have hash from port.get_node() ?
There was a problem hiding this comment.
I am not sure, as ov::Node can have multiple outputs (with different indices)
| LOG_DEBUG("Input name " << input_name << " doesn't contain kv cache. Skipping."); | ||
| continue; | ||
| } | ||
| NPUW_ASSERT(m_kvcache_in_ports.find(input_name) == m_kvcache_in_ports.end()); |
There was a problem hiding this comment.
This assert looks very suspicious
| // model's I/O are appended to original model's I/O at the end, | ||
| // thus it is safe to loop over KVCache I/O blocks just using some | ||
| // start offsets. | ||
| std::size_t start_idx_in_outputs = 0u; |
There was a problem hiding this comment.
Where do we set this one to something different than 0?
There was a problem hiding this comment.
In constructor of LLMCompiledModel: https://github.com/openvinotoolkit/openvino/pull/31520/files#diff-7c0906d2d24827a3f37868e79470c1a41ea4bf66eae148a462c0442f8e9cc191R1087
AsyaPronina
commented
Jul 30, 2025
| @@ -531,9 +530,9 @@ void ov::npuw::LLMInferRequest::infer_prefill(ov::SoPtr<ov::ITensor> input_ids, | |||
| if (m_lm_head_request) { | |||
| LOG_DEBUG("Calling inference for LM head model."); | |||
| m_lm_head_request->infer(); | |||
| m_logits = m_lm_head_request->get_tensor(m_lm_head_logits_port); | |||
| update_out_tensors_from(m_lm_head_request); | |||
There was a problem hiding this comment.
Wrong. Only logits will be in LM head, other outputs should be gathered from prefill.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Details:
Tickets: