mirror of
https://github.com/huggingface/transformers.git
synced 2025-07-16 02:58:23 +06:00
94306352f4
11 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Alazar
|
94306352f4
|
Port IDEFICS to tensorflow (#26870)
* Initial commit * Just a copy of modeling_idefics.py that will be ported to TF * - Prepend TF to the name of all classes - Convert pytorch ops to TF (not all operations are converted yet) * Add TF imports * Add autotranslated files * Add TF classes to model_tf_auto.py * Add the TF classes in model_doc * include auto-translated code * Adopted from auto-translated version * Add a forgotten super().build * Add test code for TF version. * Fix indentation and load pytorch weights for now * Some fixes. Many tests are still failing but some are passing now. - I have added TODO's for some of the hacks I made to unblock me and I will address them soon - I have the processing_idefics.py hacked in my view to support TF temporarily * Add ALL_LAYERNORM_LAYERS to match pytorch * Revert "Add ALL_LAYERNORM_LAYERS to match pytorch" This reverts commit 7e0a35119b4d7a6284d04d8c543fba1b29e573c9 as it is not needed in the tf implementation. * Fix freeze_relevant_params() * Some more fixes * Fix test_attention_outputs * Add tf stuff to processing_idefics.py processing_idefics.py supports both pytorch and tf now. test_processor_idefics.py for pytorch is passing, so i didn't break anything but still some issues with tf. I also need to add tf tests in test_processor_idefics.py. * Pass return_tensors to image processing code and fix test * Pass return_tensors to the image processor __init__ * Fix several test cases - Make input to some of the forward pass of type `TFModelInputType` - Decorate main layer forward pass with `@unpack_inputs` - Decorate main layer with `@keras_serializable` - Pass `inputs` to TFIdeficsModel * Some more fixes forgotten in last commit * Fix processing code and vision_tf.py * Fix perceiver bug * Import from * Auto-add build() methods + style pass * Fix build() errors due to `None` being passed as shape to some layers * Change name in TFIdeficsForVisionText2Text to attribute in IdeficsForVisionText2Text * Fix pytorch weights load for tf2 There were a lot of `name=` missing in weight initialization code. * Attempt to fix CI * Add back accidently removed line * Remove torch-specific stuff from the TF test file * make fix-copies, make style, remove autotranslated files * Fixes to imports/docstrings * Let's try the from future import in desperation * Fix the core random_attention_mask fn to match the torch/flax behaviour * Clean random_attention_mask up correctly * Remove torch-only test * Fix loss shape, couple of nits * make style * Don't test for OOB embeddings because IDEFICS uses those deliberately * Fix loss computation to handle masking * Fix test failures when flattening * Fix some test failures - Add cross attention gate which was missing and wasn't being passed arround - Fix overwriting of image_attention_mask due to hack I had for dummy inputs * Add a proper stateless scaled_dot_product_attention * make style * Adding missing attribute from the PyTorch version * Small cleanups to decoupledlinearlayer in case that helps * Pass epsilon to LayerNormalization * Attemp to fix pytorch weight cross-loading for TFIdeficsEmbedding * Fix a bug in TFIdeficsGatedCrossAttentionLayer * Patching up build() methods * Constant self.inv_freq * Constant self.inv_freq * First working version The TF implementation works now, there was a bug in the TFIdeficsDecoupledLinear where the weights were mis-intialized (in_features,out_features) when it should be: (out_features, in_features) I have tested this so far with tiny-random and idefics-9b-instruct and gives correct output. I also dumped the final outputs for both pytorch and TF and they are identical. * Fix some test failures * remove print statement * Fix return_tensors * Fix CI test failure check_code_quality * Attempt to fix CI failures by running `make fixup` The hardcoded IDs in test_modeling_tf_idefics.py are for the integration test and makes that file unreadable and should probably be moved to a seperate file. * Attempt to fix tests_pr_documentation_tests * Fix a test failure in test_image_processing_idefics.py * Fix test test_pt_tf_model_equivalence * Fix a few failures * Tiny fix * Some minor fixes * Remove a duplicate test * Override a few test failures for IDEFICS - `test_keras_save_load` is passing now - `test_compile_tf_model` is still failing * Fix processing_idefics.py after rebase * Guard import keras with is_tf_available * fix check code quality * fix check code quality * Minor fixes * Skip test_save_load temporarily This test passed on my local box but fails on the CI, skipping for now to see if there are other remaining failures on the CI. * Run `ruff format tests src utils` * Fix last failing test, `test_compile_tf_model` * Add fixes for vision_tf.py I forgot to add this file in last commit. * Minor fixes * Replace "<<<" with "<<" for doc tests IDEFICS-9B is too big for doctest runner, so don't run it there * Make code more readable * Fix bug after code review I added a layer_norm_eps to IdeficsConfig but I don't even need it since the vision config has a layer_norm_eps. * Fix after code review Use original code tokenizer.convert_tokens_to_ids * Keep PyTorch as the default return_tensors * Fixes to modeling_tf after code review * Fixes from code review - Remove all references of `TF_IDEFICS_PRETRAINED_MODEL_ARCHIVE_LIST` - Pass 1e-5 to LayerNormalization in perceiver * Run ruff * Undo a change * Refactor processing code after Matt's suggestion * Remove TODO's that aren't needed anymore * For pytorch, Use original pytorch processing code from main Since this PR is a TF port it shouldn't make any modifications to pytorch IDEFICS code. This changes undo's the pytorch processing modifications I made and uses original code from main. * Update tests/models/idefics/test_modeling_idefics.py * Update tests/models/idefics/test_modeling_tf_idefics.py * Add missing imports for is_pt_tf_cross_test * [DO NOT MERGE]: This is a commit for debugging and will be reverted The cross test `test_pt_tf_model_equivalence` passes locally but fails when running on the CI. This commit is to help debug that and will be reverted. * Revert "[DO NOT MERGE]: This is a commit for debugging and will be reverted" This reverts commit 8f0d709ec5bd46685fb0b4259d914ffee794875b. * [DO NOT MERGE]: This commit is for debugging a CI failure and will be reverted * [DO NOT MERGE]: This commit is for debugging a CI failure and will be reverted * Revert "[DO NOT MERGE]: This commit is for debugging a CI failure and will be reverted" This reverts commit 998cc38b8c3d313bf5e5eb55a7f5b7b881897b89. * Revert "[DO NOT MERGE]: This commit is for debugging a CI failure and will be reverted" This reverts commit 1c695ac4219c4ae4d39b330b01744dc27deb7dd4. * Don't skip test_save_load IIRC test_save_load was also failing on the CI but not on my local box, it might be easier to debug that on the CI first than the cross tests * Debugging commit, will be reverted * Revert "Debugging commit, will be reverted" This reverts commit 8eafc8e41e20c4e95a3a90834f06a6e9f445e2d5. * Override `test_save_load` and push model to save Maybe this will help me repro this weird bug * pass my repo_id * add endpoint * Pass a temp (write) token just for this CI * Undo last few commits, still pushing to hub for model debugging The issue seems to be with save_pretrained(), when I looked at the model saved from the CI test failure it is basically empty and has no weights. `self.save_weights(..)` seems to be failing in save_pretrained but needs more debugging * Add logging to modeling tf utils, will be reverted just for debugging * Debugging, will revert * Revert "Debugging, will revert" This reverts commit 9d0d3075fb7c82d8cde3a5c76bc8f3876c5c55d3. * Revert "Add logging to modeling tf utils, will be reverted just for debugging" This reverts commit 774b6b7b1c17b3ce5d7634ade768f2f686cee617. * Remove `test_save_load` The CI failures are gone after my latest rebase, no idea why but I was still saving the model to my hub on HF and the tf_model.h5 file now has everything. * Run make fix-copies * Run ruff format tests src utils * Debugging commit, will be reverted * Run ruff, also trigger CI run * Run ruff again * Undo debugging commit --------- Co-authored-by: Matt <rocketknight1@gmail.com> Co-authored-by: Matt <Rocketknight1@users.noreply.github.com> |
||
|
byi8220
|
75b76a5ea4
|
[ProcessingIdefics] Attention mask bug with padding (#29449)
* Defaulted IdeficsProcessor padding to 'longest', removed manual padding * make fixup * Defaulted processor call to padding=False * Add padding to processor call in IdeficsModelIntegrationTest as well * Defaulted IdeficsProcessor padding to 'longest', removed manual padding * make fixup * Defaulted processor call to padding=False * Add padding to processor call in IdeficsModelIntegrationTest as well * redefaulted padding=longest again * fixup/doc |
||
|
Lysandre Debut
|
39114c0383
|
Remove static pretrained maps from the library's internals (#29112)
* [test_all] Remove static pretrained maps from the library's internals * Deprecate archive maps instead of removing them * Revert init changes * [test_all] Deprecate instead of removing * [test_all] PVT v2 support * [test_all] Tests should all pass * [test_all] Style * Address review comments * Update src/transformers/models/deprecated/_archive_maps.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/models/deprecated/_archive_maps.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * [test_all] trigger tests * [test_all] LLAVA * [test_all] Bad rebase --------- Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> |
||
|
fxmarty
|
80377eb018
|
F.scaled_dot_product_attention support (#26572)
* add sdpa * wip * cleaning * add ref * yet more cleaning * and more :) * wip llama * working llama * add output_attentions=True support * bigcode sdpa support * fixes * gpt-bigcode support, require torch>=2.1.1 * add falcon support * fix conflicts falcon * style * fix attention_mask definition * remove output_attentions from attnmaskconverter * support whisper without removing any Copied from statement * fix mbart default to eager renaming * fix typo in falcon * fix is_causal in SDPA * check is_flash_attn_2_available in the models init as well in case the model is not initialized through from_pretrained * add warnings when falling back on the manual implementation * precise doc * wip replace _flash_attn_enabled by config.attn_implementation * fix typo * add tests * style * add a copy.deepcopy on the config in from_pretrained, as we do not want to modify it inplace * obey to config.attn_implementation if a config is passed in from_pretrained * fix is_torch_sdpa_available when torch is not installed * remove dead code * Update src/transformers/modeling_attn_mask_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/modeling_attn_mask_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/modeling_attn_mask_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/modeling_attn_mask_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/modeling_attn_mask_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/models/bart/modeling_bart.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * remove duplicate pretraining_tp code * add dropout in llama * precise comment on attn_mask * add fmt: off for _unmask_unattended docstring * precise num_masks comment * nuke pretraining_tp in LlamaSDPAAttention following Arthur's suggestion * cleanup modeling_utils * backward compatibility * fix style as requested * style * improve documentation * test pass * style * add _unmask_unattended tests * skip meaningless tests for idefics * hard_check SDPA requirements when specifically requested * standardize the use if XXX_ATTENTION_CLASSES * fix SDPA bug with mem-efficient backend on CUDA when using fp32 * fix test * rely on SDPA is_causal parameter to handle the causal mask in some cases * fix FALCON_ATTENTION_CLASSES * remove _flash_attn_2_enabled occurences * fix test * add OPT to the list of supported flash models * improve test * properly test on different SDPA backends, on different dtypes & properly handle separately the pad tokens in the test * remove remaining _flash_attn_2_enabled occurence * Update src/transformers/modeling_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/modeling_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/modeling_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/modeling_attn_mask_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update docs/source/en/perf_infer_gpu_one.md Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * remove use_attn_implementation * fix docstring & slight bug * make attn_implementation internal (_attn_implementation) * typos * fix tests * deprecate use_flash_attention_2=True * fix test * add back llama that was removed by mistake * fix tests * remove _flash_attn_2_enabled occurences bis * add check & test that passed attn_implementation is valid * fix falcon torchscript export * fix device of mask in tests * add tip about torch.jit.trace and move bt doc below sdpa * fix parameterized.expand order * move tests from test_modeling_attn_mask_utils to test_modeling_utils as a relevant test class is already there * update sdpaattention class with the new cache * Update src/transformers/configuration_utils.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/models/bark/modeling_bark.py * address review comments * WIP torch.jit.trace fix. left: test both eager & sdpa * add test for torch.jit.trace for both eager/sdpa * fix falcon with torch==2.0 that needs to use sdpa * fix doc * hopefully last fix * fix key_value_length that has no default now in mask converter * is it flacky? * fix speculative decoding bug * tests do pass * fix following #27907 --------- Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> |
||
|
Leo Tronchon
|
851a4f7088
|
Idefics: Fix information leak with cross attention gate in modeling (#26839)
* fix image_attention gate in idefics modeling
* update comment
* cleaner gating
* fix gate condition
* create attention gate once
* update comment
* update doc of cross-attention forward
* improve comment
* bring back no_images
* pass cross_attention_gate similarly to no_images gate
* add information on gate shape
* fix no_images placement
* make tests for gate
* take off no_images logic
* update test based on comments
* raise value error if cross_attention_gate is None
* send cross_attention_gate to device
* Revert "send cross_attention_gate to device"
This reverts commit
|
||
|
Younes Belkada
|
f7ea959b96
|
[core/ GC / tests] Stronger GC tests (#27124)
* stronger GC tests * better tests and skip failing tests * break down into 3 sub-tests * break down into 3 sub-tests * refactor a bit * more refactor * fix * last nit * credits contrib and suggestions * credits contrib and suggestions --------- Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com> Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> |
||
|
Arthur
|
63864e057f
|
Fix model integration ci (#26322)
* fix wav2vec2 * nit * stash * one more file to update * fix byt5 * vocab size is 256, don't change that! * use other revision * test persimon in smaller size * style * tests * nits * update add tokens from pretrained * test tokenization * nits * potential fnet fix? * more nits * nits * correct test * assert close * udpate * ouch * fix it * some more nits * FINALLU * use `adept` checkpoints * more adept checkpoints * that was invlved! |
||
|
Yih-Dar
|
d9e4bc2895
|
Update tiny model information and pipeline tests (#26285)
* Update tiny model summary file * add to pipeline tests * revert * fix import * fix import * fix * fix * update * update * update * fix * remove BarkModelTest * fix --------- Co-authored-by: ydshieh <ydshieh@users.noreply.github.com> |
||
|
Leo Tronchon
|
869733ab62
|
IDEFICS: allow interpolation of vision's pos embeddings (#26029)
* add pos embed interpolation for vision encoder * style * update config with interpolate_pos_encoding arg * fix imports formatting * take off copied from on vision embeddings * add test for image embeddings interpolation * add credit for interpolation code * Update src/transformers/models/idefics/configuration_idefics.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/idefics/vision.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * fix condition to check nbr image patches match shape of pos embeddings * use kwargs in the forward methods for interpolation * fix tests * have interpolate_pos_encoding default to False instead of None * Update tests/models/idefics/test_modeling_idefics.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update tests/models/idefics/test_modeling_idefics.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update tests/models/idefics/test_modeling_idefics.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/idefics/configuration_idefics.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * take off for loop meant to print k,v * add interpolate_pos_encoding arg in prepare_inputs_for_generation * add test for interpolated generation * fix edge case num_patches == num_positions and height == width * add test for edge case * fix pos_embed in interpolate * allow interpolation in bf16 with upcasting * Update src/transformers/models/idefics/vision.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/models/idefics/vision.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * add multiple images tests for interpolation and generation --------- Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> |
||
|
Stas Bekman
|
7a6efe1e9f
|
[idefics] idefics-9b test use 4bit quant (#25734) | ||
|
Stas Bekman
|
6c811a322f
|
new model: IDEFICS via HuggingFaceM4 (#24796)
* rename * restore * mappings * unedited tests+docs * docs * fixes * fix auto-sync breakage * cleanup * wip * wip * add fetch_images * remove einops dependency * update * fix * fix * fix * fix * fix * re-add * add batching * rework * fix * improve * add Leo as I am extending his work * cleanup * fix * cleanup * slow-test * fix * fix * fixes * deal with warning * rename modified llama classes * rework fetch_images * alternative implementation * cleanup * strict version * cleanup * [`IDEFICS`] Fix idefics ci (#25056) * Fix IDEFICS CI * fix test file * fixup * some changes to make tests pass * fix * fixup * Update src/transformers/models/idefics/configuration_idefics.py Co-authored-by: Stas Bekman <stas00@users.noreply.github.com> --------- Co-authored-by: Stas Bekman <stas00@users.noreply.github.com> * remove compat checks * style * explain that Idefics is not for training from scratch * require pt>=2.0 * fix idefics vision config (#25092) * fix idefics vision config * fixup * clean * Update src/transformers/models/idefics/configuration_idefics.py --------- Co-authored-by: Stas Bekman <stas00@users.noreply.github.com> * cleanup * style * cleanup * Apply suggestions from code review Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * upcase * sequence of images * handle the case with no images * Update src/transformers/image_processing_utils.py Co-authored-by: Victor SANH <victorsanh@gmail.com> * support pure lm take 2 * support tokenizer options * parameterize num_channels * fix upcase * s|IdeficsForCausalLM|IdeficsForVisionText2Text|g * manual to one line * addressing review * unbreak * remove clip dependency * fix test * consistency * PIL import * Idefics prefix * Idefics prefix * hack to make tests work * style * fix * fix * revert * try/finally * cleanup * clean up * move * [`IDEFICS`] Fix idefics config refactor (#25149) * refactor config * nuke init weights * more refactor * oops * remove visual question answering pipeline support * Update src/transformers/models/idefics/clip.py Co-authored-by: Stas Bekman <stas00@users.noreply.github.com> * Update src/transformers/models/idefics/modeling_idefics.py * cleanup * mv clip.py vision.py * tidyup --------- Co-authored-by: Stas Bekman <stas00@users.noreply.github.com> Co-authored-by: Stas Bekman <stas@stason.org> * fix * license * condition on pt * fix * style * fix * rm torchvision dependency, allow custom transforms * address review * rework device arg * add_eos_token * s/transforms/transform/ * fix top level imports * fix return value * cleanup * cleanup * fix * style * license * license * Update src/transformers/models/idefics/image_processing_idefics.py Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * add a wrapper to freeze vision layears * tidyup * use the correct std/mean settings * parameterize values from config * add tests/models/idefics/test_image_processing_idefics.py * add test_processor_idefics.py * cleanup * cleanups * fix * fix * move to the right group * style * Apply suggestions from code review Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * add perceiver config * reset * missing arg docs * Apply suggestions from code review Co-authored-by: Leo Tronchon <leo.tronchon@gmail.com> * address review comments * inject automatic end of utterance tokens (#25218) * inject automatic end of utterance tokens * fix * fix * fix * rework to not use the config * not end_of_utterance_token at the end * Update src/transformers/models/idefics/processing_idefics.py Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * address review * Apply suggestions from code review Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com> * Update src/transformers/image_processing_utils.py Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com> * [`Idefics`] add image_embeddings option in generate-related methods (#25442) * add image_embeddings option in generate-related methods * style * rename image_embeddings and allow perceiver embeddings precomputation * compute embeddings within generate * make is_encoder_decoder= True the default in config * nested if else fix * better triple check * switch if elif order for pixel values / img embeds * update model_kwargs perceiver only at the end * use _prepare_model_inputs instead of encoder_decoder logic * fix comment typo * fix config default for is_encoder_decoder * style * add typehints * precompute in forward * doc builder * style * pop instead of get image hidden states * Trigger CI * Update src/transformers/models/idefics/modeling_idefics.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * Update src/transformers/models/idefics/modeling_idefics.py Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * fix * + indentation + style * simplify a bit the use_resampler logic using comments * update diocstrings * Trigger CI --------- Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> * fix rebase changes * unbreak #25237 - to be fixed in follow up PRs * is_composition = False * no longer needed --------- Co-authored-by: leot13 <leo.tronchon@gmail.com> Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com> Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> Co-authored-by: Victor SANH <victorsanh@gmail.com> Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com> Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com> Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com> |