Commit Graph

19564 Commits

Author SHA1 Message Date
Arthur
9968c85e4b fixes 2025-07-03 15:36:52 +02:00
Arthur
5af5bccd56 current updates 2025-07-03 15:28:32 +02:00
Arthur
3cba8ac3f3 fix stupid kosmos2 2025-07-03 13:17:18 +02:00
Arthur
0f3c368384 nits 2025-07-03 11:44:07 +02:00
Arthur
d462a8ea38 fix csm! 2025-07-03 11:41:30 +02:00
Arthur
a9690f43fd fix cross attention outputs! 2025-07-03 11:32:48 +02:00
Arthur
6eb5e53e75 more fixes to moonshine! 2025-07-03 11:12:29 +02:00
Arthur
cfe62b6b95 generic needs to support more 2025-07-03 11:10:11 +02:00
Arthur
b81df9bd56 nits 2025-07-03 10:54:31 +02:00
Arthur
b3c8641f24 more moonshine fixes, 3 failures left! 2025-07-03 10:51:45 +02:00
Arthur
499ae87ef7 fix moonshine 2025-07-03 10:48:50 +02:00
Arthur
4fc83fa3a2 fix samhq? 2025-07-03 10:20:02 +02:00
Arthur
c4d43c5324 updates 2025-07-03 10:14:28 +02:00
Arthur
17cf5424b0 protect torch
Some checks are pending
Secret Leaks / trufflehog (push) Waiting to run
2025-07-02 13:35:02 +02:00
Arthur
a267d8d472 holy shit it was just graph breaks 2025-07-02 12:17:30 +02:00
Arthur
253307a305 update
Some checks are pending
Secret Leaks / trufflehog (push) Waiting to run
2025-07-01 17:47:17 +02:00
Arthur
501aead20b dose this fix it? 2025-07-01 17:28:48 +02:00
Arthur
0c9f6de0fd more fixes? 2025-07-01 16:29:15 +02:00
Arthur
e2973440d1 phix phi3 2025-07-01 16:23:22 +02:00
Arthur
d8ee27e495 fixup 2025-07-01 16:20:27 +02:00
Arthur
4834aeca61 only for some models 2025-07-01 16:12:54 +02:00
Arthur
6a5f410d26 fix janusss 2025-07-01 16:09:58 +02:00
Arthur
5065b9a285 small fixes 2025-07-01 16:05:29 +02:00
Arthur
d04c2b1ab6 fix mistral now
Some checks are pending
Secret Leaks / trufflehog (push) Waiting to run
2025-07-01 15:56:37 +02:00
Arthur
075bd0c2f3 fux csm and mistral 2025-07-01 15:53:44 +02:00
Arthur
5e5ae84a05 fix csm now 2025-07-01 15:41:52 +02:00
Arthur
aaae861fc8 fix another one 2025-07-01 15:37:33 +02:00
Arthur
9fa5f266a1 fix small lm3 2025-07-01 15:32:17 +02:00
Arthur
6a132a0799 finish fixing gemma3n 2025-07-01 15:22:52 +02:00
Arthur
f7a1f0da3d some fixes, loss_kwargs should never had been 2025-07-01 15:19:32 +02:00
Arthur
0b119ffb1f quel enfer 2025-07-01 15:06:54 +02:00
Arthur
3ac6c52f34 move the fix a bit 2025-07-01 15:00:38 +02:00
Arthur
00afce9837 fix emu3 2025-07-01 14:58:12 +02:00
Arthur
10fb88ae84 fix emu3 2025-07-01 14:53:05 +02:00
Arthur
209d5022ac update 2025-07-01 14:47:38 +02:00
Arthur
da50ccc549 fix conflicts 2025-07-01 14:42:33 +02:00
Arthur
2748b99388 update 2025-07-01 14:39:58 +02:00
Arthur
22423738c4 update 2025-07-01 14:27:21 +02:00
Arthur
15a8ff4fe9 update 2025-07-01 14:20:56 +02:00
Arthur
a13a98c6da more fix 2025-07-01 14:19:38 +02:00
Arthur
7a0512a1f5 fixes 2025-07-01 14:16:22 +02:00
StevenBucaille
1283877571
[superglue] fix wrong concatenation which made batching results wrong (#38850)
Some checks are pending
Self-hosted runner (benchmark) / Benchmark (aws-g5-4xlarge-cache) (push) Waiting to run
Build documentation / build (push) Waiting to run
New model PR merged notification / Notify new model (push) Waiting to run
Slow tests on important models (on Push - A10) / Get all modified files (push) Waiting to run
Slow tests on important models (on Push - A10) / Slow & FA2 tests (push) Blocked by required conditions
Self-hosted runner (push-caller) / Check if setup was changed (push) Waiting to run
Self-hosted runner (push-caller) / build-docker-containers (push) Blocked by required conditions
Self-hosted runner (push-caller) / Trigger Push CI (push) Blocked by required conditions
Secret Leaks / trufflehog (push) Waiting to run
Update Transformers metadata / build_and_package (push) Waiting to run
2025-07-01 12:14:44 +00:00
Raushan Turganbay
f8b88866f5
[VLMs] support passing embeds along with pixels (#38467)
* VLMs can work with embeds now

* update more models

* fix tests

* fix copies

* fixup

* fix

* style

* unskip tests

* fix copies

* fix tests

* style

* omni modality models

* qwen models had extra indentation

* fix some other tests

* fix copies

* fix test last time

* unrelated changes revert

* we can't rely only on embeds

* delete file

* de-flake mistral3

* fix qwen models

* fix style

* fix tests

* fix copies

* deflake the test

* modular reverted by fixes, fix again

* flaky test, overwritten

* fix copies

* style
2025-07-01 11:33:20 +00:00
Ayush Singh
20901f1d68
[typing] LlamaAttention return typehint (#38998)
* helo llama

* helo llama

* helo llama

* apply modular

* fix dia

---------

Co-authored-by: qubvel <qubvel@gmail.com>
2025-07-01 11:29:52 +01:00
Raushan Turganbay
7a25f8dfdb
[qwen2-vl] fix FA2 inference (#39121)
* fix FA2

* update is causal flag and remove mask for FA2

* update for FA2 with varlen path

* how the tests were passing with different devices?

* add comment and ref to the PR

* move mask preparation to base pretrained model

* seq len is the first dim, not second

* fix copies to fix GLM4V
2025-07-01 10:18:37 +00:00
Mehant Kammakomati
def9663239
feat: support indivisible shards for TP model loading and TPlizing. (#37220)
* feat: support uneven loading and sharding
resolve merge conflicts
Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

* fix: allow for empty tensor computations

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

* test: add llama1b test case

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

* due to q_proj colwise it has to be multi of 2

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

* refactor: use slice API

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

* refactor: use slice API

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

* refactor: use slice API

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

* refactor: use slice API

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>

---------

Signed-off-by: Mehant Kammakomati <mehant.kammakomati2@ibm.com>
2025-07-01 10:03:22 +00:00
jiqing-feng
06c4a4d499
fix caching_allocator_warmup with tie weights (#39070)
* fix caching_allocator_warmup with tie weights

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

* fix comment

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

---------

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
2025-07-01 11:32:20 +02:00
Raushan Turganbay
e435574721
🚨 Don't use cache in non-generative models (#38751)
* deprecate for 1 version

* style

* fix some tests

* fix esm

* skip for now, GC requires positional args but we have keyword args

* remove transpose for scores in modified models only

* skip fx trace tests
2025-07-01 09:08:21 +00:00
Arthur
3c0c56b84d test this 2025-07-01 10:58:16 +02:00
Arthur
780141ca52 same 2025-07-01 10:56:29 +02:00