Bump transformers version (#3272)

### Changes

- Bumped `transformers` version to `>=4.48.0`
- Bumped `optimum-intel` version to `>=1.22.0` (as it requires
`transformers`)
- Bumped `optimum` version to `>=1.24.0` (as it requires `transformers`)
- Updated statistics caching test with `position_ids` to fix:

    >Exception from src/plugins/intel_cpu/src/node.cpp:769:
[CPU] Concat node with name
'__module.model.model.decoder.layers.0.self_attn/aten::cat/Concat' Check
'TRShape::merge_into(output_shape, in_copy)' failed at
src/core/shape_inference/include/concat_shape_inference.hpp:43:
While validating node 'opset1::Concat
__module.model.model.decoder.layers.0.self_attn/aten::cat/Concat
(opset1::Parameter Parameter_753753[0]:f32[?,4,?,4], opset1::Transpose
__module.model.model.decoder.layers.0.self_attn/aten::transpose/Transpose_1[0]:f32[?,4,?,4])
-> (f32[?,4,?,4])' with friendly_name
'__module.model.model.decoder.layers.0.self_attn/aten::cat/Concat':
Shape inference input shapes {{1,4,0,4},{0,4,0,4}}
Argument shapes are inconsistent; they must have the same rank, and must
have equal dimension everywhere except on the concatenation axis (axis
2).
- Bumped `accelerate` version to `>=1.1.0` to fix:

> NotImplementedError: data_seed requires Accelerate version
`accelerate` >= 1.1.0. This is not supported and we recommend you to
update your version.
- Added `num_items_in_batch` to `compute_loss` to fix:

> TypeError: CompressionTrainer.compute_loss() got an unexpected keyword
argument 'num_items_in_batch'
- Bumped `_ONNX_DEFAULT_OPSET` from 13 to 14 to fix:

> torch.onnx.errors.UnsupportedOperatorError: Exporting the operator
'aten::scaled_dot_product_attention' to ONNX opset version 13 is not
supported. Support for this operator was added in version 14, try
exporting with this version.

- Updated sparsity/pruning/nas references for tests.
### Reason for changes

- Security issues with `transformers<4.48`

### Related tickets

- N/A

### Tests

- install -
https://github.com/openvinotoolkit/nncf/actions/runs/13393410660 -
passed
- weight compression -
https://github.com/openvinotoolkit/nncf/actions/runs/13393412141 -
passed
- examples -
https://github.com/openvinotoolkit/nncf/actions/runs/13393412919 -
passed
- PTQ - manual/job/post_training_quantization/622/ - passed
- PTWC - manual/job/post_training_weight_compression/327/ - passed

---------

Co-authored-by: Nikolay Lyalyushkin <nikolay.lyalyushkin@intel.com>

Author: nikita-malininn

examples/llm_compression/openvino/smollm2_360m_fp8/requirements.txt

@@ -1,5 +1,5 @@
 datasets
 openvino==2024.6
-optimum-intel[openvino]
-transformers
+optimum-intel[openvino]>=1.22.0
+transformers>=4.48.0
 onnx==1.17.0

examples/llm_compression/openvino/tiny_llama/requirements.txt

@@ -1,5 +1,5 @@
-transformers
+transformers>=4.48.0
 datasets==2.14.7
 openvino==2025.0
-optimum-intel[openvino]
+optimum-intel[openvino]>=1.22.0
 onnx==1.17.0

examples/llm_compression/openvino/tiny_llama_find_hyperparams/requirements.txt

@@ -2,6 +2,6 @@ datasets
 whowhatbench @ git+https://github.com/openvinotoolkit/openvino.genai#subdirectory=tools/who_what_benchmark
 numpy>=1.23.5,<2
 openvino==2025.0
-optimum-intel>=1.13.0
-transformers>=4.35.2
+optimum-intel>=1.22.0
+transformers>=4.48.0
 onnx==1.17.0

examples/llm_compression/openvino/tiny_llama_synthetic_data/requirements.txt

@@ -2,6 +2,6 @@ torch==2.5.1
 datasets==3.0.1
 numpy>=1.23.5,<2
 openvino==2025.0
-optimum-intel>=1.13.0
-transformers>=4.35.2
+optimum-intel>=1.22.0
+transformers>=4.48.0
 onnx==1.17.0

nncf/torch/exporter.py

@@ -71,7 +71,7 @@ class PTExporter(Exporter):
     This class provides export of the compressed model to the ONNX format.
     """
 
-    _ONNX_DEFAULT_OPSET = 13
+    _ONNX_DEFAULT_OPSET = 14
 
     @staticmethod
     def parse_format(save_format: str) -> Tuple[str, dict]:

tests/openvino/native/quantization/test_weights_compression_statistics_caching.py

@@ -14,6 +14,7 @@
 from typing import Tuple
 
 import datasets
+import numpy as np
 import openvino as ov
 from optimum.intel.openvino import OVModelForCausalLM
 from transformers import AutoTokenizer
@@ -37,6 +38,11 @@ def transform_fn(data, model=model, tokenizer=tokenizer):
         input_ids = tokenized_text["input_ids"]
         inputs = {"input_ids": input_ids, "attention_mask": tokenized_text["attention_mask"]}
 
+        if "position_ids" in model.input_names:
+            position_ids = np.cumsum(inputs["attention_mask"], axis=1) - 1
+            position_ids[inputs["attention_mask"] == 0] = 1
+            inputs["position_ids"] = position_ids
+
         batch_size = input_ids.shape[0]
         if hasattr(model, "key_value_input_names"):
             for input_name in model.key_value_input_names:

tests/openvino/requirements.txt

@@ -13,6 +13,6 @@ addict>=2.4.0
 timm==0.9.2
 efficientnet_pytorch==0.7.1
 datasets==3.0.1
-transformers==4.45.2
-optimum-intel==1.20.0
-optimum==1.23.1
+transformers==4.48.3
+optimum-intel==1.22.0
+optimum==1.24.0

tests/post_training/requirements.txt

@@ -10,13 +10,13 @@ pytest-split
 
 librosa==0.10.0
 memory-profiler==0.61.0
-optimum-intel==1.21.0
-optimum==1.23.3
+optimum-intel==1.22.0
+optimum==1.24.0
 scikit-learn>=1.2.2,<=1.5.0
 soundfile==0.12.1
 tensorboard==2.13.0
 tensorflow-io==0.32.0
 timm==0.9.2
-transformers==4.46.3
+transformers==4.48.3
 whowhatbench @ git+https://github.com/openvinotoolkit/openvino.genai.git@2025.0.0.0#subdirectory=tools/who_what_benchmark
 datasets==3.1.0

tests/torch/data/experimental/sparsify_activations/dummy_llama_int8_sym_weights_sparse_activations.dot

@@ -2,61 +2,46 @@ strict digraph  {
 0 [color=grey, label="Block: S:1__O:0
 Producers: 10
 Consumers: ", style=filled];
-1 [color=grey, label="Block: S:64__O:0
+1 [color=red, label="Block: S:64__O:0
 Producers: 10
 Consumers: ", style=filled];
-2 [color=grey, label="Block: S:64__O:0
-Producers: 10,11
-Consumers: ", style=filled];
-3 [color=green, label="Block: S:64__O:0
-Producers: 10,11,14
-Consumers: 29", style=filled];
-4 [color=grey, label="Block: S:1__O:64
+2 [color=red, label="Block: S:1__O:64
 Producers: 10
 Consumers: ", style=filled];
-5 [color=green, label="Block: S:1__O:64
-Producers: 10,11
-Consumers: 20", style=filled];
-6 [color=grey, label="Block: S:1__O:0
-Producers: 11
+3 [color=grey, label="Block: S:1__O:0
+Producers: 13
+Consumers: ", style=filled];
+4 [color=red, label="Block: S:64__O:0
+Producers: 13
 Consumers: ", style=filled];
-7 [color=grey, label="Block: S:64__O:0
-Producers: 11
+5 [color=red, label="Block: S:1__O:64
+Producers: 13
 Consumers: ", style=filled];
-8 [color=grey, label="Block: S:1__O:64
-Producers: 11
+6 [color=grey, label="Block: S:1__O:0
+Producers: 16
 Consumers: ", style=filled];
-9 [color=grey, label="Block: S:1__O:0
-Producers: 14
+7 [color=red, label="Block: S:64__O:0
+Producers: 16
 Consumers: ", style=filled];
-10 [color=grey, label="Block: S:64__O:0
-Producers: 14
+8 [color=red, label="Block: S:1__O:64
+Producers: 16
 Consumers: ", style=filled];
-11 [color=green, label="Block: S:1__O:64
-Producers: 14
-Consumers: 29", style=filled];
-12 [color=red, label="Block: S:1__O:0
-Producers: 29
+9 [color=red, label="Block: S:1__O:0
+Producers: 22
 Consumers: ", style=filled];
-13 [color=green, label="Block: S:1__O:0
-Producers: 33
-Consumers: 35", style=filled];
-14 [color=red, label="Block: S:1__O:0
-Producers: 35
+10 [color=green, label="Block: S:1__O:0
+Producers: 26
+Consumers: 28", style=filled];
+11 [color=red, label="Block: S:1__O:0
+Producers: 28
 Consumers: ", style=filled];
-15 [color=red, label="Block: S:1__O:0
-Producers: 39
+12 [color=red, label="Block: S:1__O:0
+Producers: 32
 Consumers: ", style=filled];
 0 -> 1;
-0 -> 4;
-1 -> 2;
-2 -> 3;
-4 -> 5;
+0 -> 2;
+3 -> 4;
+3 -> 5;
 6 -> 7;
 6 -> 8;
-7 -> 2;
-8 -> 5;
-9 -> 10;
-9 -> 11;
-10 -> 3;
 }

tests/torch/data/experimental/sparsify_activations/dummy_llama_sparse_activations.dot

@@ -2,62 +2,47 @@ strict digraph  {
 0 [color=grey, label="Block: S:1__O:0
 Producers: 10
 Consumers: ", style=filled];
-1 [color=grey, label="Block: S:2__O:0
+1 [color=red, label="Block: S:2__O:0
 Producers: 10
 Consumers: ", style=filled];
-2 [color=grey, label="Block: S:2__O:0
-Producers: 10,12
-Consumers: ", style=filled];
-3 [color=green, label="Block: S:2__O:0
-Producers: 10,12,16
-Consumers: 34", style=filled];
-4 [color=grey, label="Block: S:1__O:2
+2 [color=red, label="Block: S:1__O:2
 Producers: 10
 Consumers: ", style=filled];
-5 [color=green, label="Block: S:1__O:2
-Producers: 10,12
-Consumers: 27", style=filled];
+3 [color=grey, label="Block: S:1__O:0
+Producers: 11
+Consumers: ", style=filled];
+4 [color=red, label="Block: S:2__O:0
+Producers: 11
+Consumers: ", style=filled];
+5 [color=red, label="Block: S:1__O:2
+Producers: 11
+Consumers: ", style=filled];
 6 [color=grey, label="Block: S:1__O:0
 Producers: 12
 Consumers: ", style=filled];
-7 [color=grey, label="Block: S:2__O:0
+7 [color=red, label="Block: S:2__O:0
 Producers: 12
 Consumers: ", style=filled];
-8 [color=grey, label="Block: S:1__O:2
+8 [color=red, label="Block: S:1__O:2
 Producers: 12
 Consumers: ", style=filled];
-9 [color=grey, label="Block: S:1__O:0
-Producers: 16
+9 [color=red, label="Block: S:1__O:0
+Producers: 22
 Consumers: ", style=filled];
-10 [color=grey, label="Block: S:2__O:0
-Producers: 16
+10 [color=grey, label="Block: S:1__O:0
+Producers: 25
 Consumers: ", style=filled];
-11 [color=green, label="Block: S:1__O:2
-Producers: 16
-Consumers: 34", style=filled];
+11 [color=green, label="Block: S:1__O:0
+Producers: 25
+Consumers: 29", style=filled];
 12 [color=red, label="Block: S:1__O:0
-Producers: 34
-Consumers: ", style=filled];
-13 [color=grey, label="Block: S:1__O:0
-Producers: 37
-Consumers: ", style=filled];
-14 [color=green, label="Block: S:1__O:0
-Producers: 37
-Consumers: 41", style=filled];
-15 [color=red, label="Block: S:1__O:0
-Producers: 41
+Producers: 29
 Consumers: ", style=filled];
 0 -> 1;
-0 -> 4;
-1 -> 2;
-2 -> 3;
-4 -> 5;
+0 -> 2;
+3 -> 4;
+3 -> 5;
 6 -> 7;
 6 -> 8;
-7 -> 2;
-8 -> 5;
-9 -> 10;
-9 -> 11;
-10 -> 3;
-13 -> 14;
+10 -> 11;
 }

tests/torch/data/reference_graphs/pruning_groups/1_layer_BERT.dot

@@ -1,62 +1,47 @@
 strict digraph  {
 0 [color=grey, label="Block: S:1__O:0
-Producers: 7
+Producers: 8
 Consumers: ", style=filled];
-1 [color=grey, label="Block: S:2__O:0
-Producers: 7
+1 [color=red, label="Block: S:2__O:0
+Producers: 8
 Consumers: ", style=filled];
-2 [color=grey, label="Block: S:2__O:0
-Producers: 7,10
+2 [color=red, label="Block: S:1__O:2
+Producers: 8
 Consumers: ", style=filled];
-3 [color=green, label="Block: S:2__O:0
-Producers: 7,10,13
-Consumers: 29", style=filled];
-4 [color=grey, label="Block: S:1__O:2
-Producers: 7
+3 [color=grey, label="Block: S:1__O:0
+Producers: 11
 Consumers: ", style=filled];
-5 [color=green, label="Block: S:1__O:2
-Producers: 7,10
-Consumers: 18", style=filled];
-6 [color=grey, label="Block: S:1__O:0
-Producers: 10
+4 [color=red, label="Block: S:2__O:0
+Producers: 11
 Consumers: ", style=filled];
-7 [color=grey, label="Block: S:2__O:0
-Producers: 10
+5 [color=red, label="Block: S:1__O:2
+Producers: 11
 Consumers: ", style=filled];
-8 [color=grey, label="Block: S:1__O:2
-Producers: 10
+6 [color=grey, label="Block: S:1__O:0
+Producers: 14
 Consumers: ", style=filled];
-9 [color=grey, label="Block: S:1__O:0
-Producers: 13
+7 [color=red, label="Block: S:2__O:0
+Producers: 14
 Consumers: ", style=filled];
-10 [color=grey, label="Block: S:2__O:0
-Producers: 13
+8 [color=red, label="Block: S:1__O:2
+Producers: 14
 Consumers: ", style=filled];
-11 [color=green, label="Block: S:1__O:2
-Producers: 13
-Consumers: 29", style=filled];
-12 [color=red, label="Block: S:1__O:0
-Producers: 29
+9 [color=red, label="Block: S:1__O:0
+Producers: 21
 Consumers: ", style=filled];
-13 [color=green, label="Block: S:1__O:0
-Producers: 32
-Consumers: 34", style=filled];
-14 [color=red, label="Block: S:1__O:0
-Producers: 34
+10 [color=green, label="Block: S:1__O:0
+Producers: 24
+Consumers: 26", style=filled];
+11 [color=red, label="Block: S:1__O:0
+Producers: 26
 Consumers: ", style=filled];
-15 [color=red, label="Block: S:1__O:0
-Producers: 39
+12 [color=red, label="Block: S:1__O:0
+Producers: 31
 Consumers: ", style=filled];
 0 -> 1;
-0 -> 4;
-1 -> 2;
-2 -> 3;
-4 -> 5;
+0 -> 2;
+3 -> 4;
+3 -> 5;
 6 -> 7;
 6 -> 8;
-7 -> 2;
-8 -> 5;
-9 -> 10;
-9 -> 11;
-10 -> 3;
 }