Tests (llm): update LLM tests for new transformers version (#1088)

---------

Co-authored-by: Giuseppe Franco <[email protected]>

requirements/requirements-llm.txt

@@ -1,3 +1,3 @@
 # optimum-amd[brevitas] @ git+https://github.com/huggingface/optimum-amd.git@main
 tqdm
-transformers[sentencepiece]==4.45.2
+transformers[sentencepiece]

src/brevitas_examples/llm/llm_quant/mha_layers.py

@@ -152,6 +152,7 @@ def forward(
         attention_mask: Optional[torch.Tensor] = None,
         layer_head_mask: Optional[torch.Tensor] = None,
         output_attentions: bool = False,
+        position_ids: Optional[torch.Tensor] = None,
     ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
         if key_value_states is None:
             key_value_states = hidden_states
@@ -164,14 +165,17 @@ def forward(
             query_seq_length, batch_size = hidden_states.shape[:2]
             key_value_seq_length = key_value_states.shape[0]
         num_heads = self.num_heads
-        attention_mask = attention_mask_handler(
-            attention_mask, batch_size, num_heads, query_seq_length, key_value_seq_length)
+        attention_mask = (
+            attention_mask_handler(
+                attention_mask, batch_size, num_heads, query_seq_length, key_value_seq_length)
+            if attention_mask is not None else None)
         attn_output, attn_output_weights = self.mha(
             hidden_states,
             key_value_states,
             key_value_states,
             attn_mask=attention_mask,
             need_weights=output_attentions,
-            average_attn_weights=False)
+            average_attn_weights=False,
+        )
         past_key_value = None
         return attn_output, attn_output_weights, past_key_value

src/brevitas_examples/llm/llm_quant/prepare_for_quantize.py

@@ -2,11 +2,16 @@
 
 import torch
 from transformers.models.opt.modeling_opt import OPTAttention
+from transformers.models.opt.modeling_opt import OPTSdpaAttention
 
 from brevitas.graph import ModuleToModuleByClass
 from brevitas_examples.llm.llm_quant.mha_layers import QuantizableOPTAttention
 
-QUANTIZABLE_MHA_MAP = {OPTAttention: (QuantizableOPTAttention, {'batch_first': True})}
+QUANTIZABLE_MHA_MAP = {
+    OPTAttention: (QuantizableOPTAttention, {
+        'batch_first': True}),
+    OPTSdpaAttention: (QuantizableOPTAttention, {
+        'batch_first': True}),}
 
 
 def replace_mha_with_quantizable_layers(model, dtype):

tests/brevitas_examples/test_llm.py

@@ -14,6 +14,7 @@
 import pytest
 import pytest_cases
 import torch
+import transformers
 
 from brevitas import config
 from brevitas import torch_version
@@ -40,6 +41,10 @@ def allexact(x, y):
     return np.allclose(x, y, rtol=0.0, atol=0.0, equal_nan=False)
 
 
+def transformers_version_ge(required_version: str):
+    return version.parse(required_version) >= version.parse(transformers.__version__)
+
+
 # Check that all args in args are used
 def validate_args(args):
     a = vars(args)
@@ -126,6 +131,7 @@ def default_run_args(request):
     args.weight_quant_granularity = "per_channel"  # "per_tensor", "per_channel", "per_group".
     args.input_bit_width = 8
     args.act_calibration = True
+    args.no_float16 = True
     return args
 
 
@@ -203,18 +209,18 @@ def test_small_models_toggle_run_args_pt_ge_2_4(
         "llama",
         "mistral",],
     params=[
-        {
-            "model": "hf-internal-testing/tiny-random-MistralForCausalLM",
-            "act_equalization": "layerwise",
-            "gptq": True,
-            "float_ppl": 31274.05078125,
-            "quant_ppl": 33139.23046875},
         {
             "model": "hf-internal-testing/tiny-random-LlamaForCausalLM",
             "act_equalization": "fx",
             "bias_corr": True,
-            "float_ppl": 33239.5,
-            "quant_ppl": 33283.75390625},])
+            "float_ppl": 33312.0 if transformers_version_ge('4.46.0') else 33239.5,
+            "quant_ppl": 33056.0 if transformers_version_ge('4.46.0') else 33283.75390625},
+        {
+            "model": "hf-internal-testing/tiny-random-MistralForCausalLM",
+            "act_equalization": "layerwise",
+            "gptq": True,
+            "float_ppl": 31056.0 if transformers_version_ge('4.46.0') else 31274.05078125,
+            "quant_ppl": 33056.0 if transformers_version_ge('4.46.0') else 33139.23046875},])
 def acc_args_and_acc(default_run_args, request):
     args = default_run_args
     run_dict = request.param