RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.
Go agent-harnessagent-native-databaseagentic-aiagentic-retrievalagentic-searchaiai-agentscontext-enginecontext-engineeringcontext-managementharness-engineeringknowledge-compilationragretrieval-augmented-generationretrieval-harnesssearch-harness
fix(llm): raise a clear error for unsupported Bedrock embedding models (#16044)
Lucas Kim86edfc45a8bc
diff --git a/rag/llm/embedding_model.py b/rag/llm/embedding_model.pyindex 3ffc61513e..9d0d6596a0 100644--- a/rag/llm/embedding_model.py+++ b/rag/llm/embedding_model.py@@ -758,6 +758,8 @@ class BedrockEmbed(Base):body = {"inputText": text}elif self.is_cohere:body = {"texts": [text], "input_type": "search_document"}+ else:+ raise EmbeddingError(f"BedrockEmbed: unsupported embedding model '{self.model_name}' (expected an 'amazon.' or 'cohere.' model)")response = self.client.invoke_model(modelId=self.model_name, body=json.dumps(body))model_response = json.loads(response["body"].read())# Bedrock does not report token usage; count locally.@@ -772,6 +774,8 @@ class BedrockEmbed(Base):body = {"inputText": text}elif self.is_cohere:body = {"texts": [text], "input_type": "search_query"}+ else:+ raise EmbeddingError(f"BedrockEmbed: unsupported embedding model '{self.model_name}' (expected an 'amazon.' or 'cohere.' model)")try:response = self.client.invoke_model(modelId=self.model_name, body=json.dumps(body))model_response = json.loads(response["body"].read())diff --git a/test/unit_test/rag/llm/test_embedding_model.py b/test/unit_test/rag/llm/test_embedding_model.pyindex cb879684ca..98fec1393b 100644--- a/test/unit_test/rag/llm/test_embedding_model.py+++ b/test/unit_test/rag/llm/test_embedding_model.py@@ -408,6 +408,22 @@ class TestBedrockResponseParsing:vector, _ = embed.encode_queries("q")np.testing.assert_array_equal(vector, np.array([5.0, 6.0]))+ # The message must name the offending model and the expected prefixes so the+ # error stays actionable; assert on it to catch future wording regressions.+ _UNSUPPORTED_MSG = r"unsupported embedding model 'meta\.embed-model'.*amazon\..*cohere\."++ def test_unknown_provider_raises_embedding_error(self):+ # A model that is neither amazon.* nor cohere.* must surface a clear+ # EmbeddingError instead of an UnboundLocalError on the unset `body`.+ embed = self._make("meta")+ with pytest.raises(EmbeddingError, match=self._UNSUPPORTED_MSG):+ embed.encode(["hello"])++ def test_unknown_provider_query_raises_embedding_error(self):+ embed = self._make("meta")+ with pytest.raises(EmbeddingError, match=self._UNSUPPORTED_MSG):+ embed.encode_queries("q")+@pytest.mark.p2class TestBaiduYiyanResponseParsing: