tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

Refactor separate attention flags with attn_implementation and capability/concerns feature flags (#3602)

Browse Source 
* upgrade to torchao 0.17.0

* chore: lint

* refactor attention handling

* replace legacy attention boolean flags with capability properties

Replace checks with capability-based properties derived from attn_implementation

This separates three concerns that were conflated under flash_attention:
1. Backend selection -> attn_implementation enum
2. Packing capability -> attn_supports_packing property
3. Flash-attn library dependency -> attn_uses_flash_lib property

* compute attn capability flags in normalizer instead of properties

* make attn_implementation the single source of truth

* move attention-dependent validators to mode=after

* migrate remaining consumers to canonical attn_implementation

* expand attention tests + rewrite docs

* migrate example configs to canonical attn_implementation

* update doc snippets + reject gemma4-hybrid with non-FA2 backend

* remove dead gemma4 branch in _set_attention_config

* fix duplicate attn_implementation in gpt-oss yamls and flaky caplog tests

* drop "Phase 2" naming from attn-implementation tests

* regroup attn_implementation tests by feature concern

* clean up verbose comments and remove MD

Signed-off-by: Wing Lian <wing@axolotl.ai>
Co-authored-by: Axolotl Swarm <no-reply@axolotl.ai>

* fix(collator): pass return_dict=True at apply_chat_template top level for transformers 5.x

In transformers 5.x, ProcessorMixin.apply_chat_template gained its own
`return_dict` parameter (defaulting to False).  When return_dict=False
and tokenize=True the method returns out["input_ids"] directly — a 2-D
tensor — rather than the full BatchFeature dict.

The old code placed `return_dict=True` inside processor_kwargs.  In
transformers 5.x those kwargs are forwarded to the underlying processor
call self(...) where _merge_kwargs silently ignores any key not present
in MllamaProcessorKwargs (emitting a warning).  The outer return_dict
therefore stayed False, apply_chat_template returned the raw input_ids
tensor, and the subsequent `batch["input_ids"]` attempted to index a
2-D tensor with the 9-character string "input_ids", producing:

  IndexError: too many indices for tensor of dimension 2

The fix is to pass return_dict=True as a top-level keyword argument to
apply_chat_template (where it is actually consumed) and remove it from
processor_kwargs (where it was silently dropped).  No version guard is
needed: transformers is pinned to ==5.5.4 in pyproject.toml.

Adds a unit-level regression test (tests/test_mm_chat_collator.py) that
mocks the processor to return a raw tensor when apply_chat_template is
called without top-level return_dict=True, verifying the four invariants:
process_rows returns a dict, input_ids is 2-D, labels is 2-D, and
apply_chat_template receives return_dict=True as a top-level kwarg.

Fixes: tests/e2e/test_llama_vision.py::TestLlamaVision::test_lora_llama_vision_multimodal_dataset
Fixes: tests/e2e/test_llama_vision.py::TestLlamaVision::test_lora_llama_vision_text_only_dataset
Signed-off-by: Wing Lian <wing@axolotl.ai>
Co-authored-by: Axolotl Swarm <no-reply@axolotl.ai>

* fix(collator): process_rows returns dict (BatchFeature) shape

Two related changes for the multimodal chat collator under transformers 5.x:

1. Wrap apply_chat_template result in dict(...) so process_rows returns
   a plain dict rather than a BatchFeature instance. BatchFeature is a
   Mapping but not a dict; downstream code that did
     batch["labels"] = self.processing_strategy.process_labels(batch["input_ids"])
   would index on a tensor when the result wasn't dict-shaped, raising
     IndexError: too many indices for tensor of dimension 2

2. Soften the regression test's contract from `dict` to `Mapping` so it
   exercises the actual semantic guarantee (key/value access) rather
   than the implementation detail (dict vs BatchFeature). Test guards
   against the original transformers 5.x breakage where apply_chat_template's
   return_dict default went from True to False.

Includes regression test under tests/test_mm_chat_collator.py.

Bug surfaced via swarm dispatch task_01KQHPNAYD8XARSNSDJVW1GPF6 against
attn-implementation-refactor; squash-merged from agent commits 4de886fd
+ dc9fcf4f.

Signed-off-by: Wing Lian <wing@axolotl.ai>

---------

Signed-off-by: Wing Lian <wing@axolotl.ai>
Co-authored-by: Axolotl Swarm <no-reply@axolotl.ai>
This commit is contained in:
Wing Lian
2026-05-05 10:15:18 -04:00
committed by GitHub
parent 6136ae627b
commit e4032fc90f
252 changed files with 1502 additions and 572 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
docs/optimizations.qmd
View File

@@ -22,12 +22,12 @@ Improves GPU utilization by combining multiple short sequences into a single pac
Using an optimized attention implementation is critical for training speed.
- **[Flash Attention 2](https://github.com/Dao-AILab/flash-attention)**: `flash_attention: true`. **(Recommended)** The industry standard for fast attention on modern GPUs. Requires Ampere or higher. For AMD, check [AMD Support](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#amd-rocm-support).
- **[Flex Attention](https://pytorch.org/blog/flexattention/)**: `flex_attention: true`.
- **[SDP Attention](https://docs.pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html)**: `sdp_attention: true`. PyTorch's native implementation.
- **[Xformers](https://github.com/facebookresearch/xformers)**: `xformers_attention: true`. Works with FP16.
- **[Flash Attention 2](https://github.com/Dao-AILab/flash-attention)**: `attn_implementation: flash_attention_2`. **(Recommended)** The industry standard for fast attention on modern GPUs. Requires Ampere or higher. For AMD, check [AMD Support](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#amd-rocm-support).
- **[Flex Attention](https://pytorch.org/blog/flexattention/)**: `attn_implementation: flex_attention`.
- **[SDP Attention](https://docs.pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html)**: `attn_implementation: sdpa`. PyTorch's native implementation.
- **[Xformers](https://github.com/facebookresearch/xformers)**: `attn_implementation: xformers`. Works with FP16.
*Note: You should only enable one attention backend.*
See [Attention](attention.qmd) for the full list of backends and the canonical values.
### LoRA Optimizations