diff --git a/.env b/.env index 56f5fa2..6d75f13 100644 --- a/.env +++ b/.env @@ -15,7 +15,7 @@ OWL_DB_DIR=db OWL_LOG_DIR=logs DOCIO_WORKERS=1 DOCIO_DEVICE=cpu -EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2 -RERANKER_MODEL=cross-encoder/ms-marco-TinyBERT-L-2 +EMBEDDING_MODEL=BAAI/bge-small-en-v1.5 +RERANKER_MODEL=mixedbread-ai/mxbai-rerank-xsmall-v1 OWL_CONCURRENT_ROWS_BATCH_SIZE=3 -OWL_CONCURRENT_COLS_BATCH_SIZE=5 \ No newline at end of file +OWL_CONCURRENT_COLS_BATCH_SIZE=5 diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index 9e788d9..af40c9f 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -55,8 +55,8 @@ jobs: run: | set -e export API_DEVICE=cpu - export EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2 - export RERANKER_MODEL=cross-encoder/ms-marco-TinyBERT-L-2 + export EMBEDDING_MODEL=BAAI/bge-small-en-v1.5 + export RERANKER_MODEL=mixedbread-ai/mxbai-rerank-xsmall-v1 # Edit .env file ORGS=$(printenv | grep API_KEY | xargs -I {} echo {} | cut -d '=' -f 1) diff --git a/docker/compose.cpu.yml b/docker/compose.cpu.yml index 3650e8a..6c33e0a 100644 --- a/docker/compose.cpu.yml +++ b/docker/compose.cpu.yml @@ -1,6 +1,6 @@ services: infinity: - image: michaelf34/infinity:0.0.32 + image: michaelf34/infinity:0.0.63 container_name: jamai_infinity entrypoint: [ diff --git a/services/api/src/owl/configs/models.json b/services/api/src/owl/configs/models.json index 6989c0b..fc8bc3d 100644 --- a/services/api/src/owl/configs/models.json +++ b/services/api/src/owl/configs/models.json @@ -135,9 +135,9 @@ "owned_by": "cohere" }, { - "id": "ellm/sentence-transformers/all-MiniLM-L6-v2", - "litellm_id": "openai/sentence-transformers/all-MiniLM-L6-v2", - "context_length": 8192, + "id": "ellm/BAAI/bge-small-en-v1.5", + "litellm_id": "openai/BAAI/bge-small-en-v1.5", + "context_length": 512, "embedding_size": 1024, "languages": ["mul"], "api_base": "http://infinity:6909", @@ -149,10 +149,10 @@ { "id": "cohere/rerank-english-v3.0", "context_length": 512, "languages": ["en"], "owned_by": "cohere" }, { "id": "cohere/rerank-multilingual-v3.0", "context_length": 512, "languages": ["mul"], "owned_by": "cohere" }, { - "id": "ellm/cross-encoder/ms-marco-TinyBERT-L-2", - "context_length": 8192, + "id": "ellm/mixedbread-ai/mxbai-rerank-xsmall-v1", + "context_length": 512, "languages": ["mul"], - "api_base": "http://infinity:6919", + "api_base": "http://infinity:6909", "owned_by": "ellm" } ]