From 3e555aa5ed7b9d14aca37e0a2f84c6b14fe155d4 Mon Sep 17 00:00:00 2001 From: technillogue Date: Fri, 17 May 2024 18:44:34 -0400 Subject: [PATCH] upgrade cog and use new format to specify target concurrency --- cog.yaml | 4 +++- default.nix | 4 ++-- 2 files changed, 5 insertions(+), 3 deletions(-) diff --git a/cog.yaml b/cog.yaml index 2d2dcd0..6a48f34 100644 --- a/cog.yaml +++ b/cog.yaml @@ -2,4 +2,6 @@ build: gpu: true python_version: "3.10" predict: "predict.py:Predictor" -concurrency: 64 +concurrency: + max: 64 + default_target: 16 diff --git a/default.nix b/default.nix index 176e5bc..d9f7473 100644 --- a/default.nix +++ b/default.nix @@ -12,7 +12,7 @@ in imports = [ ./interface.nix ]; cog.build = { python_version = "3.10"; - cog_version = "0.10.0-alpha5"; + cog_version = "0.10.0-alpha7"; cuda = "12.1"; # todo: 12.2 gpu = true; # inspiration: echo tensorrt_llm==0.8.0 | uv pip compile - --extra-index-url https://pypi.nvidia.com -p 3.10 --prerelease=allow --annotation-style=line @@ -45,7 +45,7 @@ in overrides = [ "tokenizers==0.19.0" "transformers==4.40.0" - "cog @ https://r2.drysys.workers.dev/tmp/cog-0.10.0a6-py3-none-any.whl" + "cog==0.10.0a7" ]; }; cognix.includeNix = true;