let's not warn when someone is running a forward

src/transformers/models/cohere/modeling_cohere.py

@@ -769,7 +769,9 @@ def forward(
 
         past_seen_tokens = 0
         return_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)
+        if (
+            use_cache and not isinstance(past_key_values, Cache) and not self.training
+        ):  # kept for BC (non `Cache` `past_key_values` inputs)
             return_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
 

src/transformers/models/dbrx/modeling_dbrx.py

@@ -1005,7 +1005,9 @@ def forward(
         inputs_embeds = nn.functional.dropout(inputs_embeds, p=self.emb_pdrop, training=self.training)
 
         return_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)
+        if (
+            use_cache and not isinstance(past_key_values, Cache) and not self.training
+        ):  # kept for BC (non `Cache` `past_key_values` inputs)
             return_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/gemma/diff_gemma.py

@@ -474,7 +474,9 @@ def forward(
             inputs_embeds = self.embed_tokens(input_ids)
 
         return_legacy_cache = False  # noqa: F841
-        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)
+        if (
+            use_cache and not isinstance(past_key_values, Cache) and not self.training
+        ):  # kept for BC (non `Cache` `past_key_values` inputs)
             return_legacy_cache = True  # noqa: F841
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
 

src/transformers/models/gemma/modeling_gemma.py

@@ -770,7 +770,9 @@ def forward(
             inputs_embeds = self.embed_tokens(input_ids)
 
         return_legacy_cache = False  # noqa: F841
-        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)
+        if (
+            use_cache and not isinstance(past_key_values, Cache) and not self.training
+        ):  # kept for BC (non `Cache` `past_key_values` inputs)
             return_legacy_cache = True  # noqa: F841
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
 
@@ -795,7 +797,9 @@ def forward(
         # See https://github.com/huggingface/transformers/pull/29402
         normalizer = torch.tensor(self.config.hidden_size**0.5, dtype=hidden_states.dtype)
         hidden_states = hidden_states * normalizer
-        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)
+        if (
+            use_cache and not isinstance(past_key_values, Cache) and not self.training
+        ):  # kept for BC (non `Cache` `past_key_values` inputs)
             return_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/jetmoe/modeling_jetmoe.py

@@ -978,7 +978,9 @@ def forward(
             inputs_embeds = self.embed_tokens(input_ids)
 
         return_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)
+        if (
+            use_cache and not isinstance(past_key_values, Cache) and not self.training
+        ):  # kept for BC (non `Cache` `past_key_values` inputs)
             return_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
 

src/transformers/models/llama/modeling_llama.py

@@ -894,7 +894,9 @@ def forward(
             inputs_embeds = self.embed_tokens(input_ids)
 
         return_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)
+        if (
+            use_cache and not isinstance(past_key_values, Cache) and not self.training
+        ):  # kept for BC (non `Cache` `past_key_values` inputs)
             return_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/mistral/modeling_mistral.py

@@ -758,7 +758,7 @@ def forward(
             inputs_embeds = self.embed_tokens(input_ids)
 
         return_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             return_legacy_cache = True
             logger.warning_once(

src/transformers/models/mixtral/modeling_mixtral.py

@@ -960,7 +960,7 @@ def forward(
                 use_cache = False
 
         use_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             use_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/olmo/modeling_olmo.py

@@ -811,7 +811,9 @@ def forward(
             inputs_embeds = self.embed_tokens(input_ids)
 
         return_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):  # kept for BC (non `Cache` `past_key_values` inputs)
+        if (
+            use_cache and not isinstance(past_key_values, Cache) and not self.training
+        ):  # kept for BC (non `Cache` `past_key_values` inputs)
             return_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/persimmon/modeling_persimmon.py

@@ -626,7 +626,7 @@ def forward(
                 use_cache = False
 
         use_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             use_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/phi/modeling_phi.py

@@ -909,7 +909,7 @@ def forward(
                 use_cache = False
 
         use_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             use_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/phi3/modeling_phi3.py

@@ -950,7 +950,7 @@ def forward(
                 use_cache = False
 
         use_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             use_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/qwen2/modeling_qwen2.py

@@ -808,7 +808,7 @@ def forward(
                 use_cache = False
 
         use_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             use_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/qwen2_moe/modeling_qwen2_moe.py

@@ -970,7 +970,7 @@ def forward(
                 use_cache = False
 
         use_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             use_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/stablelm/modeling_stablelm.py

@@ -902,7 +902,7 @@ def forward(
                 use_cache = False
 
         use_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             use_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(

src/transformers/models/starcoder2/modeling_starcoder2.py

@@ -784,7 +784,7 @@ def forward(
                 use_cache = False
 
         use_legacy_cache = False
-        if use_cache and not isinstance(past_key_values, Cache):
+        if use_cache and not isinstance(past_key_values, Cache) and not self.training:
             use_legacy_cache = True
             past_key_values = DynamicCache.from_legacy_cache(past_key_values)
             logger.warning_once(