From 1c5fa02d1d31ca00274a361a5cb60be46f3526a8 Mon Sep 17 00:00:00 2001
From: Christopher Laffey <47035210+laffeychris@users.noreply.github.com>
Date: Thu, 6 Jun 2024 11:22:43 +0100
Subject: [PATCH] fix: specify embedding retriever version (#7811)

---
 haystack/nodes/retriever/_embedding_encoder.py               | 5 ++++-
 pyproject.toml                                               | 2 +-
 ...-sentence-transformer-model-version-55b633e8b294189c.yaml | 5 +++++
 3 files changed, 10 insertions(+), 2 deletions(-)
 create mode 100644 releasenotes/notes/specify-sentence-transformer-model-version-55b633e8b294189c.yaml

diff --git a/haystack/nodes/retriever/_embedding_encoder.py b/haystack/nodes/retriever/_embedding_encoder.py
index f75b997123..d05af6ee45 100644
--- a/haystack/nodes/retriever/_embedding_encoder.py
+++ b/haystack/nodes/retriever/_embedding_encoder.py
@@ -135,7 +135,10 @@ def __init__(self, retriever: "EmbeddingRetriever"):
         # e.g. 'roberta-base-nli-stsb-mean-tokens'
         torch_and_transformers_import.check()
         self.embedding_model = SentenceTransformer(
-            retriever.embedding_model, device=str(retriever.devices[0]), use_auth_token=retriever.use_auth_token
+            retriever.embedding_model,
+            device=str(retriever.devices[0]),
+            use_auth_token=retriever.use_auth_token,
+            revision=retriever.model_version,
         )
         self.batch_size = retriever.batch_size
         self.embedding_model.max_seq_length = retriever.max_seq_len
diff --git a/pyproject.toml b/pyproject.toml
index 5f576ac52a..42b8acda5a 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -87,7 +87,7 @@ dependencies = [
 [project.optional-dependencies]
 inference = [
   "transformers[torch,sentencepiece]==4.39.3",
-  "sentence-transformers>=2.2.0",  # See haystack/nodes/retriever/_embedding_encoder.py, _SentenceTransformersEmbeddingEncoder
+  "sentence-transformers>=2.3.1",  # See haystack/nodes/retriever/_embedding_encoder.py, _SentenceTransformersEmbeddingEncoder
   "huggingface-hub>=0.5.0",
 ]
 elasticsearch = [
diff --git a/releasenotes/notes/specify-sentence-transformer-model-version-55b633e8b294189c.yaml b/releasenotes/notes/specify-sentence-transformer-model-version-55b633e8b294189c.yaml
new file mode 100644
index 0000000000..2c43594726
--- /dev/null
+++ b/releasenotes/notes/specify-sentence-transformer-model-version-55b633e8b294189c.yaml
@@ -0,0 +1,5 @@
+---
+fixes:
+  - |
+    Fix bug causing latest version of sentence transformer model always being downloaded, even if specific version
+    is given.