From 0c02dc845cb733d5fa73e6719e98dd7bc72602e5 Mon Sep 17 00:00:00 2001 From: David Gardner Date: Mon, 24 Jun 2024 11:22:27 -0700 Subject: [PATCH 1/4] Don't use shared mem, triton cpp impl doesn't support it --- examples/llm/vdb_upload/vdb_utils.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/examples/llm/vdb_upload/vdb_utils.py b/examples/llm/vdb_upload/vdb_utils.py index d9e39b2553..3aee5584ec 100644 --- a/examples/llm/vdb_upload/vdb_utils.py +++ b/examples/llm/vdb_upload/vdb_utils.py @@ -284,7 +284,7 @@ def build_cli_configs(source_type, "force_convert_inputs": True, "model_name": embedding_model_name, "server_url": triton_server_url, - "use_shared_memory": True, + "use_shared_memory": False, }, "num_threads": num_threads, } From 846921420f7c9050c5753a777b9130dda1bbdfaa Mon Sep 17 00:00:00 2001 From: David Gardner Date: Mon, 24 Jun 2024 11:22:41 -0700 Subject: [PATCH 2/4] Don't use shared mem, triton cpp impl doesn't support it --- examples/llm/vdb_upload/vdb_config.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/examples/llm/vdb_upload/vdb_config.yaml b/examples/llm/vdb_upload/vdb_config.yaml index 5698cc2e83..0931665637 100644 --- a/examples/llm/vdb_upload/vdb_config.yaml +++ b/examples/llm/vdb_upload/vdb_config.yaml @@ -20,7 +20,7 @@ vdb_pipeline: force_convert_inputs: true model_name: "all-MiniLM-L6-v2" server_url: "http://localhost:8001" - use_shared_memory: true + use_shared_memory: false pipeline: edge_buffer_size: 128 From 6b9054a1ed05499c1705e4d05796bb58b465a172 Mon Sep 17 00:00:00 2001 From: David Gardner Date: Mon, 24 Jun 2024 11:23:10 -0700 Subject: [PATCH 3/4] Restore the default source type to rss --- examples/llm/vdb_upload/run.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/examples/llm/vdb_upload/run.py b/examples/llm/vdb_upload/run.py index 19ea993605..b889f6c17a 100644 --- a/examples/llm/vdb_upload/run.py +++ b/examples/llm/vdb_upload/run.py @@ -104,7 +104,7 @@ def run(): @click.option("--source_type", multiple=True, type=click.Choice(['rss', 'filesystem', 'doca'], case_sensitive=False), - default=[], + default=['rss'], show_default=True, help="The type of source to use. Can specify multiple times for different source types.") @click.option( From 5350104c88239dd5a53168d50a82283b1fc803bf Mon Sep 17 00:00:00 2001 From: David Gardner Date: Mon, 24 Jun 2024 12:01:11 -0700 Subject: [PATCH 4/4] Give the --question flag a default values since the pipeline fails without one --- examples/llm/rag/run.py | 1 + 1 file changed, 1 insertion(+) diff --git a/examples/llm/rag/run.py b/examples/llm/rag/run.py index 9798f875bf..c060f34127 100644 --- a/examples/llm/rag/run.py +++ b/examples/llm/rag/run.py @@ -80,6 +80,7 @@ def run(): "--question", type=str, multiple=True, + default=["What are some new attacks discovered in the cyber security industry?"] * 5, help="The question to answer with the RAG pipeline. Specify multiple times to answer multiple questions at once.", ) def pipeline(**kwargs):