Conversation
- Pin transformers to 4.57.3 - Keep QEff cache internals self-owned (CacheLayerMixin/Cache adapter path), with legacy interop. - Update model kv_seq_len calls to use cross-version cache-length resolution. - Add small quantizer compatibility guards (AWQ/update_dtype paths). Signed-off-by: vbaddi <vbaddi@qti.qualcomm.com>
Signed-off-by: vbaddi <vbaddi@qti.qualcomm.com>
Signed-off-by: vbaddi <vbaddi@qti.qualcomm.com>
Signed-off-by: vbaddi <vbaddi@qti.qualcomm.com>
…ve unused kv helper
- Add a causal-LM unit quickcheck that exports with use_onnx_subfunctions=True and
asserts decoder-block subfunction cardinality (single vs multi) per model
expectations.
- Count only decoder block functions derived from get_submodules_for_export(), not
all ONNX helper functions.
- Remove unused resolve_kv_seq_len from QEfficient/utils/_utils.py after migrating
wrappers away from it.
Signed-off-by: vbaddi <vbaddi@qti.qualcomm.com>
vbaddi
requested review from
ochougul,
quic-hemagnih and
quic-rishinr
as code owners
Signed-off-by: Abhishek Kumar Singh <sabhis@qti.qualcomm.com>
Signed-off-by: Abhishek Kumar Singh <sabhis@qti.qualcomm.com>
Signed-off-by: Abhishek Kumar Singh <sabhis@qti.qualcomm.com>
Signed-off-by: Abhishek Kumar Singh <sabhis@qti.qualcomm.com>
Signed-off-by: Abhishek Kumar Singh <sabhis@qti.qualcomm.com>
vbaddi
commented
Mar 25, 2026
|
|
||
| query_states, key_states = qeff_apply_rotary_pos_emb( | ||
| query_states, key_states, cos_cached, sin_cached, position_ids[1:], self.rope_scaling["mrope_section"] | ||
| query_states, key_states, cos, sin, position_ids[1:], self.rope_scaling["mrope_section"] |
Contributor
Author
There was a problem hiding this comment.
nit: Let's keep this consistent as *_cached
| def __qeff_init__(self): | ||
| self.rotary_emb = QEffQwen2_5_VLRotaryEmbedding(config=self.config) | ||
| QEffQwen2_5_VLRotaryEmbedding._max_seq_len_cached = self.config.max_position_embeddings | ||
| self.rotary_emb._set_cos_sin_cache( |
Contributor
Author
There was a problem hiding this comment.
nit: is this done for all models too?
Signed-off-by: abhishek-singh591 <sabhis@qti.qualcomm.com>
Signed-off-by: Dipankar Sarkar <dipankar@qti.qualcomm.com>
Signed-off-by: abhishek-singh591 <sabhis@qti.qualcomm.com>
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
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.
This PR rebases dev/rebase_transformers_v4_57_3 onto main and consolidates our transformer rebase changes with the PR #880 subfunction/KV alignment so we keep the branch simpler and unify the subfunction approach.
What changed
Decoder-block subfunction status (causal model list)
Tests verified