From 11440395f4f1055f52f483d0adcec9aff13e635b Mon Sep 17 00:00:00 2001
From: bogdankostic <bogdankostic@web.de>
Date: Fri, 1 Sep 2023 11:48:41 +0200
Subject: [PATCH] fix: Set model_max_length in the Tokenizer of
 `DefaultPromptHandler` (#5596)

* Set model_max_length in tokenizer in prompt handler

* Add release note
---
 haystack/nodes/prompt/invocation_layer/handlers.py     |  1 +
 ...del_max_length-prompt_handler-7f34c40c62a8c55b.yaml |  4 ++++
 test/prompt/test_handlers.py                           | 10 ++++++++++
 3 files changed, 15 insertions(+)
 create mode 100644 releasenotes/notes/fix-model_max_length-prompt_handler-7f34c40c62a8c55b.yaml

diff --git a/haystack/nodes/prompt/invocation_layer/handlers.py b/haystack/nodes/prompt/invocation_layer/handlers.py
index 446ddedf0..073561b3f 100644
--- a/haystack/nodes/prompt/invocation_layer/handlers.py
+++ b/haystack/nodes/prompt/invocation_layer/handlers.py
@@ -63,6 +63,7 @@ class DefaultPromptHandler:
 
     def __init__(self, model_name_or_path: str, model_max_length: int, max_length: int = 100):
         self.tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
+        self.tokenizer.model_max_length = model_max_length
         self.model_max_length = model_max_length
         self.max_length = max_length
 
diff --git a/releasenotes/notes/fix-model_max_length-prompt_handler-7f34c40c62a8c55b.yaml b/releasenotes/notes/fix-model_max_length-prompt_handler-7f34c40c62a8c55b.yaml
new file mode 100644
index 000000000..d2fe46866
--- /dev/null
+++ b/releasenotes/notes/fix-model_max_length-prompt_handler-7f34c40c62a8c55b.yaml
@@ -0,0 +1,4 @@
+---
+fixes:
+  - |
+    Fix model_max_length not being set in the Tokenizer in DefaultPromptHandler.
diff --git a/test/prompt/test_handlers.py b/test/prompt/test_handlers.py
index 2d6da1318..f7b8a3e9b 100644
--- a/test/prompt/test_handlers.py
+++ b/test/prompt/test_handlers.py
@@ -57,6 +57,13 @@ def test_prompt_handler_negative():
     }
 
 
+@pytest.mark.unit
+@patch("haystack.nodes.prompt.invocation_layer.handlers.AutoTokenizer.from_pretrained")
+def test_prompt_handler_model_max_length_set_in_tokenizer(mock_tokenizer):
+    prompt_handler = DefaultPromptHandler(model_name_or_path="model_path", model_max_length=10, max_length=3)
+    assert prompt_handler.tokenizer.model_max_length == 10
+
+
 @pytest.mark.integration
 def test_prompt_handler_basics():
     handler = DefaultPromptHandler(model_name_or_path="gpt2", model_max_length=20, max_length=10)
@@ -65,6 +72,9 @@ def test_prompt_handler_basics():
     handler = DefaultPromptHandler(model_name_or_path="gpt2", model_max_length=20)
     assert handler.max_length == 100
 
+    # test model_max_length is set in tokenizer
+    assert handler.tokenizer.model_max_length == 20
+
 
 @pytest.mark.integration
 def test_gpt2_prompt_handler():