Currently, different parts of the E2E tests use different models. Some use real HuggingFace models, e.g. Qwen/Qwen2-0.5B orQwen/Qwen3-32B, and some places use model="food-review".
The E2E test uses a simulator as the backend, and its latest version (v0.9.0) utilizes a vLLM render that expects real model names.
Since the specific model used doesn't impact the test logic, standardizing on a single model across all tests will make it much easier to manage the render sidecar.
Update all tests to use Qwen/Qwen2.5-1.5B-Instruct. This model should be used because it supports both simple text requests and multi-modal messages.
Please see draft PR for reference - #1373
Currently, different parts of the E2E tests use different models. Some use real HuggingFace models, e.g. Qwen/Qwen2-0.5B orQwen/Qwen3-32B, and some places use model="food-review".
The E2E test uses a simulator as the backend, and its latest version (v0.9.0) utilizes a vLLM render that expects real model names.
Since the specific model used doesn't impact the test logic, standardizing on a single model across all tests will make it much easier to manage the render sidecar.
Update all tests to use Qwen/Qwen2.5-1.5B-Instruct. This model should be used because it supports both simple text requests and multi-modal messages.
Please see draft PR for reference - #1373