[Cherry-Pick][Text Generation] Terminate the inference when kv cache …

…is full (#1447)

* [Fix] Remove erronous LIB.kv_cache input when using external kv cache management (#1337)

* initial commit

* initial commit

* cleanup

* cleanup2

* initial commit

* initial commit

* Needs to be >=

src/deepsparse/transformers/pipelines/text_generation.py

@@ -829,6 +829,11 @@ def engine_forward(
                     generated_tokens.append(token)
                     generated_logits.append(logits)
 
+                    if session.total_num_processed_tokens >= session.capacity:
+                        # if the kv cache is full, stop generation
+                        finished_reason.append(FinishReason.CAPACITY)
+                        break
+
                     if (
                         token == self.tokenizer.eos_token_id
                         and not self.force_max_tokens