[CI/Build][Doc] Update gte-Qwen2-1.5B-instruct usage (#18683)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Signed-off-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Isotr0py <2037008807@qq.com>

Author: DarkLight1337

docs/models/supported_models.md

@@ -404,10 +404,7 @@ Specified using `--task embed`.
     You should manually set mean pooling by passing `--override-pooler-config '{"pooling_type": "MEAN"}'`.
 
 !!! note
-    The HF implementation of `Alibaba-NLP/gte-Qwen2-1.5B-instruct` is hardcoded to use causal attention despite what is shown in `config.json`. To compare vLLM vs HF results,
-    you should set `--hf-overrides '{"is_causal": true}'` in vLLM so that the two implementations are consistent with each other.
-
-    For both the 1.5B and 7B variants, you also need to enable `--trust-remote-code` for the correct tokenizer to be loaded.
+    For `Alibaba-NLP/gte-Qwen2-*`, you need to enable `--trust-remote-code` for the correct tokenizer to be loaded.
     See [relevant issue on HF Transformers](https://github.com/huggingface/transformers/issues/34882).
 
 !!! note

tests/models/language/pooling/test_embedding.py

@@ -15,13 +15,12 @@
                      marks=[pytest.mark.core_model, pytest.mark.cpu_model]),
         pytest.param("sentence-transformers/all-MiniLM-L12-v2"),
         pytest.param("intfloat/multilingual-e5-small"),
-        pytest.param("Alibaba-NLP/gte-Qwen2-7B-instruct"),
+        pytest.param("Alibaba-NLP/gte-Qwen2-1.5B-instruct"),
         # [Decoder-only]
         pytest.param("BAAI/bge-multilingual-gemma2",
                      marks=[pytest.mark.core_model]),
         pytest.param("intfloat/e5-mistral-7b-instruct",
                      marks=[pytest.mark.core_model, pytest.mark.cpu_model]),
-        pytest.param("Alibaba-NLP/gte-Qwen2-1.5B-instruct"),
         pytest.param("ssmits/Qwen2-7B-Instruct-embed-base"),
         # [Cross-Encoder]
         pytest.param("sentence-transformers/stsb-roberta-base-v2"),
@@ -47,9 +46,6 @@ def test_models(
         vllm_extra_kwargs["override_pooler_config"] = \
             PoolerConfig(pooling_type="MEAN")
 
-    if model == "Alibaba-NLP/gte-Qwen2-1.5B-instruct":
-        vllm_extra_kwargs["hf_overrides"] = {"is_causal": True}
-
     # The example_prompts has ending "\n", for example:
     # "Write a short story about a robot that dreams for the first time.\n"
     # sentence_transformers will strip the input texts, see:

tests/models/language/pooling/test_gte.py

@@ -45,9 +45,6 @@
     EmbedModelInfo("Alibaba-NLP/gte-Qwen2-1.5B-instruct",
                    architecture="Qwen2ForCausalLM",
                    enable_test=True),
-    EmbedModelInfo("Alibaba-NLP/gte-Qwen2-7B-instruct",
-                   architecture="Qwen2ForCausalLM",
-                   enable_test=False),
     ########## ModernBertModel
     EmbedModelInfo("Alibaba-NLP/gte-modernbert-base",
                    architecture="ModernBertModel",
@@ -61,9 +58,6 @@ def test_models_mteb(hf_runner, vllm_runner,
     from .mteb_utils import mteb_test_embed_models
 
     vllm_extra_kwargs: dict[str, Any] = {}
-    if model_info.name == "Alibaba-NLP/gte-Qwen2-1.5B-instruct":
-        vllm_extra_kwargs["hf_overrides"] = {"is_causal": True}
-
     if model_info.architecture == "GteNewModel":
         vllm_extra_kwargs["hf_overrides"] = {"architectures": ["GteNewModel"]}
 
@@ -81,9 +75,6 @@ def test_models_correctness(hf_runner, vllm_runner, model_info: EmbedModelInfo,
     example_prompts = [str(s).strip() for s in example_prompts]
 
     vllm_extra_kwargs: dict[str, Any] = {}
-    if model_info.name == "Alibaba-NLP/gte-Qwen2-1.5B-instruct":
-        vllm_extra_kwargs["hf_overrides"] = {"is_causal": True}
-
     if model_info.architecture == "GteNewModel":
         vllm_extra_kwargs["hf_overrides"] = {"architectures": ["GteNewModel"]}