[Misc] Split up pooling tasks (#10820)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

examples/offline_inference_embedding.py

@@ -9,7 +9,12 @@ prompts = [
 ]
 
 # Create an LLM.
-model = LLM(model="intfloat/e5-mistral-7b-instruct", enforce_eager=True)
+model = LLM(
+    model="intfloat/e5-mistral-7b-instruct",
+    task="embed",  # You should pass task="embed" for embedding models
+    enforce_eager=True,
+)
+
 # Generate embedding. The output is a list of PoolingRequestOutputs.
 outputs = model.encode(prompts)
 # Print the outputs.

examples/offline_inference_vision_language_embedding.py

@@ -59,7 +59,7 @@ def run_e5_v(query: Query):
 
     llm = LLM(
         model="royokong/e5-v",
-        task="embedding",
+        task="embed",
         max_model_len=4096,
     )
 
@@ -88,7 +88,7 @@ def run_vlm2vec(query: Query):
 
     llm = LLM(
         model="TIGER-Lab/VLM2Vec-Full",
-        task="embedding",
+        task="embed",
         trust_remote_code=True,
         mm_processor_kwargs={"num_crops": 4},
     )