update examples

Signed-off-by: Kyle Sayers <[email protected]>

Author: kylesayrs

examples/multimodal_vision/idefics3_example.py

@@ -31,7 +31,6 @@ def data_collator(batch):
     GPTQModifier(
         targets="Linear",
         scheme="W4A16",
-        sequential_targets=["LlamaDecoderLayer"],
         ignore=["re:.*lm_head", "re:model.vision_model.*", "re:model.connector.*"],
     ),
 ]
@@ -91,6 +90,7 @@ def tokenize(sample):
     num_calibration_samples=NUM_CALIBRATION_SAMPLES,
     trust_remote_code_model=True,
     data_collator=data_collator,
+    sequential_targets=["LlamaDecoderLayer"],
 )
 
 # Confirm generations of the quantized model look sane.

examples/multimodal_vision/llava_example.py

@@ -30,7 +30,6 @@ def data_collator(batch):
     GPTQModifier(
         targets="Linear",
         scheme="W4A16",
-        sequential_targets=["LlamaDecoderLayer"],
         ignore=["re:.*lm_head", "re:vision_tower.*", "re:multi_modal_projector.*"],
     ),
 ]
@@ -46,6 +45,7 @@ def data_collator(batch):
     num_calibration_samples=NUM_CALIBRATION_SAMPLES,
     trust_remote_code_model=True,
     data_collator=data_collator,
+    sequential_targets=["LlamaDecoderLayer"],
 )
 
 # Confirm generations of the quantized model look sane.

examples/multimodal_vision/mistral3_example.py

@@ -43,7 +43,6 @@ def data_collator(batch):
     GPTQModifier(
         targets="Linear",
         scheme="W4A16",
-        sequential_targets=["MistralDecoderLayer"],
         ignore=["re:.*lm_head", "re:vision_tower.*", "re:multi_modal_projector.*"],
     ),
 ]
@@ -59,6 +58,7 @@ def data_collator(batch):
     num_calibration_samples=NUM_CALIBRATION_SAMPLES,
     trust_remote_code_model=True,
     data_collator=data_collator,
+    sequential_targets=["MistralDecoderLayer"],
 )
 
 # Confirm generations of the quantized model look sane.

examples/multimodal_vision/mllama_example.py

@@ -30,7 +30,6 @@ def data_collator(batch):
     GPTQModifier(
         targets="Linear",
         scheme="W4A16",
-        sequential_targets=["MllamaSelfAttentionDecoderLayer"],
         ignore=["re:.*lm_head", "re:multi_modal_projector.*", "re:vision_model.*"],
     ),
 ]
@@ -46,6 +45,7 @@ def data_collator(batch):
     num_calibration_samples=NUM_CALIBRATION_SAMPLES,
     trust_remote_code_model=True,
     data_collator=data_collator,
+    sequential_targets=["MllamaSelfAttentionDecoderLayer"],
 )
 
 # Confirm generations of the quantized model look sane.

examples/multimodal_vision/phi3_vision_example.py

@@ -75,7 +75,6 @@ def data_collator(batch):
 recipe = GPTQModifier(
     targets="Linear",
     scheme="W4A16",
-    sequential_targets=["Phi3DecoderLayer"],
     ignore=["lm_head", "re:model.vision_embed_tokens.*"],
 )
 
@@ -88,6 +87,7 @@ def data_collator(batch):
     num_calibration_samples=NUM_CALIBRATION_SAMPLES,
     trust_remote_code_model=True,
     data_collator=data_collator,
+    sequential_targets=["Phi3DecoderLayer"],
 )
 
 # Confirm generations of the quantized model look sane.

examples/multimodal_vision/pixtral_example.py

@@ -36,7 +36,6 @@ def data_collator(batch):
     GPTQModifier(
         targets="Linear",
         scheme="W4A16",
-        sequential_targets=["MistralDecoderLayer"],
         ignore=["re:.*lm_head", "re:vision_tower.*", "re:multi_modal_projector.*"],
     ),
 ]
@@ -52,6 +51,7 @@ def data_collator(batch):
     num_calibration_samples=NUM_CALIBRATION_SAMPLES,
     trust_remote_code_model=True,
     data_collator=data_collator,
+    sequential_targets=["MistralDecoderLayer"],
 )
 
 # Confirm generations of the quantized model look sane.

examples/multimodal_vision/qwen2_vl_example.py

@@ -79,7 +79,6 @@ def data_collator(batch):
     GPTQModifier(
         targets="Linear",
         scheme="W4A16",
-        sequential_targets=["Qwen2VLDecoderLayer"],
         ignore=["lm_head", "re:visual.*", "re:model.visual.*"],
     ),
 ]
@@ -94,6 +93,7 @@ def data_collator(batch):
     num_calibration_samples=NUM_CALIBRATION_SAMPLES,
     trust_remote_code_model=True,
     data_collator=data_collator,
+    sequential_targets=["Qwen2VLDecoderLayer"],
 )
 
 # Confirm generations of the quantized model look sane.

examples/multimodal_vision/qwen_2_5_vl_example.py

@@ -73,7 +73,6 @@ def data_collator(batch):
     GPTQModifier(
         targets="Linear",
         scheme="W4A16",
-        sequential_targets=["Qwen2_5_VLDecoderLayer"],
         ignore=["lm_head", "re:visual.*", "re:model.visual.*"],
     ),
 ]
@@ -88,6 +87,7 @@ def data_collator(batch):
     num_calibration_samples=NUM_CALIBRATION_SAMPLES,
     trust_remote_code_model=True,
     data_collator=data_collator,
+    sequential_targets=["Qwen2_5_VLDecoderLayer"],
 )
 
 # Confirm generations of the quantized model look sane.