new link

docs/source/en/kv_cache.md

@@ -403,13 +403,13 @@ Sometimes you would want to first fill-in cache object with key/values for certa
 ```
 
 
-## Converting to/from the legacy cache format
+## Legacy cache format
 
 Prior to the introduction of the `Cache` object, the cache of LLMs used to be a tuple of tuples of tensors. The legacy
 format has a dynamic size, growing as we generate text -- very similar to `DynamicCache`. If your project depend on
 this legacy format, you can seamlessly convert it to a `DynamicCache` and back.
 
-```py
+```python
 >>> import torch
 >>> from transformers import AutoTokenizer, AutoModelForCausalLM, DynamicCache
 

src/transformers/models/bloom/modeling_bloom.py

@@ -696,8 +696,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         batch_size, seq_length, _ = inputs_embeds.shape

src/transformers/models/codegen/modeling_codegen.py

@@ -536,8 +536,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         seq_length = inputs_embeds.shape[1]

src/transformers/models/cohere/modeling_cohere.py

@@ -920,8 +920,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/dbrx/modeling_dbrx.py

@@ -1069,8 +1069,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/falcon/modeling_falcon.py

@@ -1041,8 +1041,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         # Compute alibi tensor: check build_alibi_tensor documentation

src/transformers/models/gemma/diff_gemma.py

@@ -486,8 +486,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/gemma/modeling_gemma.py

@@ -838,8 +838,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/git/modeling_git.py

@@ -427,8 +427,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         all_hidden_states = () if output_hidden_states else None

src/transformers/models/gpt_neo/modeling_gpt_neo.py

@@ -751,8 +751,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         seq_length = inputs_embeds.shape[1]

src/transformers/models/gpt_neox/modeling_gpt_neox.py

@@ -953,8 +953,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         seq_length = inputs_embeds.shape[1]

src/transformers/models/gpt_neox_japanese/modeling_gpt_neox_japanese.py

@@ -673,8 +673,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         seq_length = inputs_embeds.shape[1]

src/transformers/models/gptj/modeling_gptj.py

@@ -823,8 +823,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         seq_length = inputs_embeds.shape[1]

src/transformers/models/granite/modeling_granite.py

@@ -844,8 +844,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/idefics/modeling_idefics.py

@@ -1249,8 +1249,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         batch_size, seq_length, _ = inputs_embeds.shape

src/transformers/models/idefics2/modeling_idefics2.py

@@ -1355,8 +1355,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
             past_seen_tokens = past_key_values.get_seq_length()
 

src/transformers/models/jetmoe/modeling_jetmoe.py

@@ -1043,8 +1043,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/llama/modeling_llama.py

@@ -954,8 +954,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/mistral/modeling_mistral.py

@@ -772,8 +772,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/mixtral/modeling_mixtral.py

@@ -1028,8 +1028,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if inputs_embeds is None:

src/transformers/models/olmo/modeling_olmo.py

@@ -876,8 +876,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/olmoe/modeling_olmoe.py

@@ -1017,8 +1017,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if cache_position is None:

src/transformers/models/persimmon/modeling_persimmon.py

@@ -695,8 +695,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if inputs_embeds is None:

src/transformers/models/phi/modeling_phi.py

@@ -986,8 +986,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if inputs_embeds is None:

src/transformers/models/phi3/modeling_phi3.py

@@ -1013,8 +1013,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if inputs_embeds is None:

src/transformers/models/qwen2/modeling_qwen2.py

@@ -925,8 +925,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if inputs_embeds is None:

src/transformers/models/qwen2_moe/modeling_qwen2_moe.py

@@ -1089,8 +1089,8 @@ def forward(
                 past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 logger.warning_once(
                     "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
-                    "will be removed in v4.47. Please use an appropriate `Cache` class "
-                    "(https://huggingface.co/docs/transformers/internal/generation_utils#transformers.Cache)"
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
                 )
 
         if inputs_embeds is None: