Draft
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17222
Note: Links to docs will display an error until the docs builds have been completed. ❌ 17 New FailuresAs of commit 9634244 with merge base 6f780c7 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
This PR needs a
|
larryliu0820
reviewed
Feb 4, 2026
| # Module]`. | ||
| model.vocab_size, | ||
| llm_config.base.metadata, | ||
| use_ring_buffer=llm_config.model.local_global_attention is not None, |
Contributor
There was a problem hiding this comment.
I don't think this is right. local global attention uses a sliding window which may or may not relevant to ring buffer.
Contributor
There was a problem hiding this comment.
I think the high level seems right, we need to have this metadata serialized into .pte.
larryliu0820
reviewed
Feb 4, 2026
| } | ||
|
|
||
| auto error = runner->generate(prompt, config); | ||
| auto error2 = runner->generate(prompt, config); |
larryliu0820
reviewed
Feb 4, 2026
|
|
||
| // Resolve max_new_tokens based on config | ||
| // Check if ring buffer is enabled - if so, we can exceed context length | ||
| bool use_ring_buffer = metadata_.at(kUseRingBuffer); |
Contributor
There was a problem hiding this comment.
I think we need to set this in llm_runner_helper not here.
larryliu0820
reviewed
Feb 4, 2026
Comment on lines
188
to
+195
| int64_t max_context_len = | ||
| metadata_.at(kMaxContextLen) - 0; // No start_pos offset | ||
| int32_t max_new_tokens = config.resolve_max_new_tokens(max_context_len, pos_); | ||
| // When ring buffer is enabled, use a large context length to allow unlimited | ||
| // generation. | ||
| int64_t effective_context_len = | ||
| use_ring_buffer ? INT64_MAX : max_context_len; | ||
| int32_t max_new_tokens = | ||
| config.resolve_max_new_tokens(effective_context_len, pos_); |
Contributor
There was a problem hiding this comment.
This logic needs to be applied to text_llm_runner as well.
larryliu0820
reviewed
Feb 4, 2026
Comment on lines
+156
to
+159
| int64_t effective_context_len = | ||
| use_ring_buffer ? INT64_MAX : max_context_len; | ||
| int max_new_tokens = | ||
| config.resolve_max_new_tokens(max_context_len, num_prompt_tokens); | ||
| config.resolve_max_new_tokens(effective_context_len, num_prompt_tokens); |
Contributor
There was a problem hiding this comment.
Seems like duplicate code to me.
larryliu0820
reviewed
Feb 4, 2026
| {llm::kMaxContextLen, 128}, | ||
| {llm::kUseKVCache, true}, | ||
| {llm::kUseSDPAWithKVCache, false}, | ||
| {llm::kUseRingBuffer, true}, |
Contributor
There was a problem hiding this comment.
I think we need to set max_context_len to INT MAX here
Refactor position shift handling in attention sink to use torch buffers and dynamic shape conditions.
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
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.
Summary
Allow exceeding context window
Test plan
CI