[CI/UT][Refactor] move e2e spec decode and deepseek acc test to per pr (#1136)

### What this PR does / why we need it?
1. run deepseek acc ut per pr --- multicard CI time increased by 9 min
2. run spec decode e2e test on v1 per pr --- singlecard CI time
increased by 3 min (partly is disabled due to not work now)
~~3. align the output of whether dbo is enabled or not~~
    The generated results with and without dbo cannot be aligned.

https://github.com/vllm-project/vllm-ascend/actions/runs/15822900528/job/44600029405?pr=1136
4. skip V0 mtp test due to failure in
https://github.com/vllm-project/vllm-ascend/actions/runs/16012172833/job/45171988816
5. fix some version conflicts
### How was this patch tested?
CI passed with new added test.

---------

Signed-off-by: MengqingCao <cmq0113@163.com>

tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py

@@ -0,0 +1,94 @@
+from __future__ import annotations
+
+import random
+from typing import Any
+
+import pytest
+from vllm import LLM, SamplingParams
+
+
+@pytest.fixture
+def test_prompts():
+    prompt_types = ["repeat", "sentence"]
+    num_prompts = 10
+    prompts = []
+
+    random.seed(0)
+    random_prompt_type_choices = random.choices(prompt_types, k=num_prompts)
+
+    # Generate a mixed batch of prompts, some of which can be easily
+    # predicted by n-gram matching and some which likely cannot.
+    for kind in random_prompt_type_choices:
+        word_choices = ["test", "temp", "hello", "where"]
+        word = random.choice(word_choices)
+        if kind == "repeat":
+            prompt = f"""
+            please repeat the word '{word}' 10 times.
+            give no other output than the word at least ten times in a row,
+            in lowercase with spaces between each word and without quotes.
+            """
+        elif kind == "sentence":
+            prompt = f"""
+            please give a ten-word sentence that
+            uses the word {word} at least once.
+            give no other output than that simple sentence without quotes.
+            """
+        else:
+            raise ValueError(f"Unknown prompt type: {kind}")
+        prompts.append([{"role": "user", "content": prompt}])
+
+    return prompts
+
+
+@pytest.fixture
+def sampling_config():
+    return SamplingParams(temperature=0, max_tokens=256, ignore_eos=False)
+
+
+@pytest.fixture
+def model_name():
+    return "wemaster/deepseek_mtp_main_random_bf16"
+
+
+@pytest.mark.skipif(
+    True, reason="TODO: Enable me after test_mtp_correctness is fixed")
+def test_mtp_correctness(
+    monkeypatch: pytest.MonkeyPatch,
+    test_prompts: list[list[dict[str, Any]]],
+    sampling_config: SamplingParams,
+    model_name: str,
+):
+    '''
+    Compare the outputs of a original LLM and a speculative LLM
+    should be the same when using mtp speculative decoding.
+    '''
+    with monkeypatch.context() as m:
+        m.setenv("VLLM_USE_V1", "1")
+
+        ref_llm = LLM(model=model_name, max_model_len=256, enforce_eager=True)
+        ref_outputs = ref_llm.chat(test_prompts, sampling_config)
+        del ref_llm
+
+        spec_llm = LLM(model=model_name,
+                       trust_remote_code=True,
+                       speculative_config={
+                           "method": "deepseek_mtp",
+                           "num_speculative_tokens": 1,
+                       },
+                       max_model_len=256,
+                       enforce_eager=True)
+        spec_outputs = spec_llm.chat(test_prompts, sampling_config)
+        matches = 0
+        misses = 0
+        for ref_output, spec_output in zip(ref_outputs, spec_outputs):
+            if ref_output.outputs[0].text == spec_output.outputs[0].text:
+                matches += 1
+            else:
+                misses += 1
+                print(f"ref_output: {ref_output.outputs[0].text}")
+                print(f"spec_output: {spec_output.outputs[0].text}")
+
+        # Heuristic: expect at least 66% of the prompts to match exactly
+        # Upon failure, inspect the outputs to check for inaccuracy.
+        assert matches > int(0.66 * len(ref_outputs))
+        del spec_llm

tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py

@@ -0,0 +1,161 @@
+# SPDX-License-Identifier: Apache-2.0
+from __future__ import annotations
+
+import random
+from typing import Any
+
+import pytest
+from vllm import LLM, SamplingParams
+
+
+@pytest.fixture
+def test_prompts():
+    prompt_types = ["repeat", "sentence"]
+    num_prompts = 10
+    prompts = []
+
+    random.seed(0)
+    random_prompt_type_choices = random.choices(prompt_types, k=num_prompts)
+
+    # Generate a mixed batch of prompts, some of which can be easily
+    # predicted by n-gram matching and some which likely cannot.
+    for kind in random_prompt_type_choices:
+        word_choices = ["test", "temp", "hello", "where"]
+        word = random.choice(word_choices)
+        if kind == "repeat":
+            prompt = f"""
+            please repeat the word '{word}' 10 times.
+            give no other output than the word at least ten times in a row,
+            in lowercase with spaces between each word and without quotes.
+            """
+        elif kind == "sentence":
+            prompt = f"""
+            please give a ten-word sentence that
+            uses the word {word} at least once.
+            give no other output than that simple sentence without quotes.
+            """
+        else:
+            raise ValueError(f"Unknown prompt type: {kind}")
+        prompts.append([{"role": "user", "content": prompt}])
+
+    return prompts
+
+
+@pytest.fixture
+def sampling_config():
+    return SamplingParams(temperature=0, max_tokens=10, ignore_eos=False)
+
+
+@pytest.fixture
+def model_name():
+    return "LLM-Research/Meta-Llama-3.1-8B-Instruct"
+
+
+def eagle_model_name():
+    return "vllm-ascend/EAGLE-LLaMA3.1-Instruct-8B"
+
+
+def eagle3_model_name():
+    return "vllm-ascend/EAGLE3-LLaMA3.1-Instruct-8B"
+
+
+def test_ngram_correctness(
+    monkeypatch: pytest.MonkeyPatch,
+    test_prompts: list[list[dict[str, Any]]],
+    sampling_config: SamplingParams,
+    model_name: str,
+):
+    '''
+    Compare the outputs of a original LLM and a speculative LLM
+    should be the same when using ngram speculative decoding.
+    '''
+    pytest.skip("Not current support for the test.")
+    with monkeypatch.context() as m:
+        m.setenv("VLLM_USE_V1", "1")
+
+        ref_llm = LLM(model=model_name, max_model_len=1024, enforce_eager=True)
+        ref_outputs = ref_llm.chat(test_prompts, sampling_config)
+        del ref_llm
+
+        spec_llm = LLM(
+            model=model_name,
+            speculative_config={
+                "method": "ngram",
+                "prompt_lookup_max": 5,
+                "prompt_lookup_min": 3,
+                "num_speculative_tokens": 3,
+            },
+            max_model_len=1024,
+            enforce_eager=True,
+        )
+        spec_outputs = spec_llm.chat(test_prompts, sampling_config)
+        matches = 0
+        misses = 0
+        for ref_output, spec_output in zip(ref_outputs, spec_outputs):
+            if ref_output.outputs[0].text == spec_output.outputs[0].text:
+                matches += 1
+            else:
+                misses += 1
+                print(f"ref_output: {ref_output.outputs[0].text}")
+                print(f"spec_output: {spec_output.outputs[0].text}")
+
+        # Heuristic: expect at least 70% of the prompts to match exactly
+        # Upon failure, inspect the outputs to check for inaccuracy.
+        assert matches > int(0.7 * len(ref_outputs))
+        del spec_llm
+
+
+@pytest.mark.parametrize("use_eagle3", [False, True], ids=["eagle", "eagle3"])
+def test_eagle_correctness(
+    monkeypatch: pytest.MonkeyPatch,
+    test_prompts: list[list[dict[str, Any]]],
+    sampling_config: SamplingParams,
+    model_name: str,
+    use_eagle3: bool,
+):
+    '''
+    Compare the outputs of a original LLM and a speculative LLM
+    should be the same when using eagle speculative decoding.
+    '''
+    if not use_eagle3:
+        pytest.skip("Not current support for the test.")
+    with monkeypatch.context() as m:
+        m.setenv("VLLM_USE_V1", "1")
+
+        ref_llm = LLM(model=model_name, max_model_len=2048, enforce_eager=True)
+        ref_outputs = ref_llm.chat(test_prompts, sampling_config)
+        del ref_llm
+
+        spec_model_name = eagle3_model_name(
+        ) if use_eagle3 else eagle_model_name()
+        spec_llm = LLM(
+            model=model_name,
+            trust_remote_code=True,
+            enable_chunked_prefill=True,
+            max_num_seqs=1,
+            max_num_batched_tokens=2048,
+            gpu_memory_utilization=0.6,
+            speculative_config={
+                "method": "eagle3" if use_eagle3 else "eagle",
+                "model": spec_model_name,
+                "num_speculative_tokens": 2,
+                "max_model_len": 128,
+            },
+            max_model_len=128,
+            enforce_eager=True,
+        )
+        spec_outputs = spec_llm.chat(test_prompts, sampling_config)
+        matches = 0
+        misses = 0
+        for ref_output, spec_output in zip(ref_outputs, spec_outputs):
+            if ref_output.outputs[0].text == spec_output.outputs[0].text:
+                matches += 1
+            else:
+                misses += 1
+                print(f"ref_output: {ref_output.outputs[0].text}")
+                print(f"spec_output: {spec_output.outputs[0].text}")
+
+        # Heuristic: expect at least 66% of the prompts to match exactly
+        # Upon failure, inspect the outputs to check for inaccuracy.
+        assert matches > int(0.66 * len(ref_outputs))
+        del spec_llm