From 733257ff2db9cea0c5b2dc0971c50088b2404d29 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Cl=C3=A9mentine=20Fourrier?= <22726840+clefourrier@users.noreply.github.com> Date: Tue, 16 Jul 2024 15:46:46 +0200 Subject: [PATCH] should fix most inference endpoints issues of version config (#226) Updates version + allow trust remote code to launch Qwen 2 models with TGI (see slack for a bug that @meg-huggingface encountered) --- src/lighteval/models/endpoint_model.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/lighteval/models/endpoint_model.py b/src/lighteval/models/endpoint_model.py index 83f42916c..87959ef61 100644 --- a/src/lighteval/models/endpoint_model.py +++ b/src/lighteval/models/endpoint_model.py @@ -91,10 +91,11 @@ def __init__( "MAX_INPUT_LENGTH": "2047", "MAX_TOTAL_TOKENS": "2048", "MODEL_ID": "/repository", + "HF_MODEL_TRUST_REMOTE_CODE": "true", **config.get_dtype_args(), **config.get_custom_env_vars(), }, - "url": (config.image_url or "ghcr.io/huggingface/text-generation-inference:1.1.0"), + "url": (config.image_url or "ghcr.io/huggingface/text-generation-inference:latest"), }, ) hlog("Deploying your endpoint. Please wait.")