add qwen3

vllm-v0.6.2/tests/prompt_adapter/test_bloom.py

@@ -0,0 +1,45 @@
+import pytest
+
+import vllm
+from vllm.prompt_adapter.request import PromptAdapterRequest
+
+MODEL_PATH = "bigscience/bloomz-560m"
+PA_PATH = 'stevhliu/bloomz-560m_PROMPT_TUNING_CAUSAL_LM'
+
+
+def do_sample(llm, pa_name: str, pa_id: int):
+
+    prompts = [
+        "Tweet text : @nationalgridus I have no water and the bill is \
+        current and paid. Can you do something about this? Label : ",
+        "Tweet text : @nationalgridus Looks good thanks! Label : "
+    ]
+    sampling_params = vllm.SamplingParams(temperature=0.0,
+                                          max_tokens=3,
+                                          stop_token_ids=[3])
+
+    outputs = llm.generate(prompts,
+                           sampling_params,
+                           prompt_adapter_request=PromptAdapterRequest(
+                               pa_name, pa_id, PA_PATH, 8) if pa_id else None)
+
+    # Print the outputs.
+    generated_texts = []
+    for output in outputs:
+        prompt = output.prompt
+        generated_text = output.outputs[0].text.strip()
+        generated_texts.append(generated_text)
+        print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
+    return generated_texts
+
+
+@pytest.mark.parametrize("enforce_eager", [True, False])
+def test_twitter_prompt_adapter(enforce_eager: bool):
+    llm = vllm.LLM(MODEL_PATH,
+                   enforce_eager=enforce_eager,
+                   enable_prompt_adapter=True,
+                   max_prompt_adapter_token=8)
+
+    expected_output = ['complaint', 'no complaint']
+
+    assert do_sample(llm, "twitter_pa", pa_id=1) == expected_output

vllm-v0.6.2/tests/prompt_adapter/test_multi_adapter_inference.py

@@ -0,0 +1,53 @@
+from vllm import EngineArgs, LLMEngine, SamplingParams
+from vllm.prompt_adapter.request import PromptAdapterRequest
+
+MODEL_PATH = "bigscience/bloomz-560m"
+pa_path = 'stevhliu/bloomz-560m_PROMPT_TUNING_CAUSAL_LM'
+pa_path2 = 'swapnilbp/angry_tweet_ptune'
+
+
+def do_sample(engine):
+
+    prompts = [
+        ("Tweet text: I have complaints! Label: ",
+         SamplingParams(temperature=0.0, max_tokens=3, stop_token_ids=[3]),
+         PromptAdapterRequest("hate_speech", 1, pa_path2, 8)),
+        ("Tweet text: I have no problems Label: ",
+         SamplingParams(temperature=0.0, max_tokens=3, stop_token_ids=[3]),
+         PromptAdapterRequest("hate_speech2", 2, pa_path2, 8)),
+        ("Tweet text: I have complaints! Label: ",
+         SamplingParams(temperature=0.0, max_tokens=3), None),
+        ("Tweet text: I have no problems Label: ",
+         SamplingParams(temperature=0.0, max_tokens=3, stop_token_ids=[3]),
+         PromptAdapterRequest("complain", 3, pa_path, 8)),
+    ]
+
+    request_id = 0
+    results = set()
+    while prompts or engine.has_unfinished_requests():
+        if prompts:
+            prompt, sampling_params, pa_request = prompts.pop(0)
+            engine.add_request(str(request_id),
+                               prompt,
+                               sampling_params,
+                               prompt_adapter_request=pa_request)
+            request_id += 1
+
+        request_outputs = engine.step()
+
+        for request_output in request_outputs:
+            if request_output.finished:
+                results.add(request_output.outputs[0].text)
+    return results
+
+
+def test_multi_prompt_adapters():
+    engine_args = EngineArgs(model=MODEL_PATH,
+                             max_prompt_adapters=3,
+                             enable_prompt_adapter=True,
+                             max_prompt_adapter_token=8)
+    engine = LLMEngine.from_engine_args(engine_args)
+    expected_output = {
+        ' quot;I', 'hate speech', 'no complaint', 'not hate speech'
+    }
+    assert do_sample(engine) == expected_output

vllm-v0.6.2/tests/prompt_adapter/test_pa_lora.py

@@ -0,0 +1,61 @@
+from huggingface_hub import snapshot_download
+
+from vllm import EngineArgs, LLMEngine, SamplingParams
+from vllm.lora.request import LoRARequest
+from vllm.prompt_adapter.request import PromptAdapterRequest
+
+MODEL_PATH = "meta-llama/Llama-2-7b-hf"
+pa_path = snapshot_download(repo_id="swapnilbp/llama_tweet_ptune")
+lora_path = snapshot_download(repo_id="yard1/llama-2-7b-sql-lora-test")
+
+
+def do_sample(engine):
+
+    prompt_text = "[user] Write a SQL query to answer the question based on the table schema.\n\n context: CREATE TABLE table_name_74 (icao VARCHAR, airport VARCHAR)\n\n question: Name the ICAO for lilongwe international airport [/user] [assistant]"  # noqa: E501
+
+    # first prompt with a prompt adapter and second without adapter
+    prompts = [
+        (prompt_text,
+         SamplingParams(temperature=0.0, max_tokens=100,
+                        stop=["[/assistant]"]),
+         PromptAdapterRequest("hate_speech", 1, pa_path,
+                              8), LoRARequest("sql_test", 1, lora_path)),
+        (prompt_text,
+         SamplingParams(temperature=0.0, max_tokens=100,
+                        stop=["[/assistant]"]), None,
+         LoRARequest("sql_test", 1, lora_path)),
+    ]
+
+    request_id = 0
+    results = set()
+    while prompts or engine.has_unfinished_requests():
+        if prompts:
+            prompt, sampling_params, pa_request, lora_request = prompts.pop(0)
+            engine.add_request(str(request_id),
+                               prompt,
+                               sampling_params,
+                               prompt_adapter_request=pa_request,
+                               lora_request=lora_request)
+            request_id += 1
+
+        request_outputs = engine.step()
+
+        for request_output in request_outputs:
+            if request_output.finished:
+                results.add(request_output.outputs[0].text)
+    return results
+
+
+def test_lora_prompt_adapter():
+    engine_args = EngineArgs(model=MODEL_PATH,
+                             enable_prompt_adapter=True,
+                             enable_lora=True,
+                             max_num_seqs=60,
+                             max_prompt_adapter_token=8)
+    engine = LLMEngine.from_engine_args(engine_args)
+    result = do_sample(engine)
+
+    expected_output = {
+        "  SELECT icao FROM table_name_74 WHERE airport = 'lilongwe international airport' "  # noqa: E501
+    }
+    assert result == expected_output