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)
huggingface · Jul 16, 2024 · 733257f · 733257f
1 parent 66e6aae
commit 733257f
Showing 1 changed file with 2 additions and 1 deletion.
diff --git 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.")