feat: vllm Model diagnostic test checking long generation quality

docs/adding-new-models.md

@@ -140,4 +140,15 @@ uv run --extra vllm tools/model_diagnostics/1.max_model_len_respected.py Qwen/Qw
 # Generated tokens: 12
 # Total tokens: 20
 # [Qwen/Qwen2.5-1.5B] ALL GOOD!
-```
+```
+
+## [2.long_generation_decode_vs_prefill](https://github.com/NVIDIA/NeMo-RL/blob/main/tools/model_diagnostics/2.long_generation_decode_vs_prefill.py)
+
+Test that vLLM yields near-identical token log-probabilities when comparing decoding with a single prefill pass across multiple prompts.
+
+```sh
+# Run that is expected to pass
+uv run --extra vllm tools/model_diagnostics/2.long_generation_decode_vs_prefill.py Qwen/Qwen2.5-1.5B
+# ...
+# [Qwen/Qwen2.5-1.5B] ALL GOOD!
+```

tools/model_diagnostics/2.long_generation_decode_vs_prefill.py

@@ -0,0 +1,102 @@
+# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import argparse
+
+import torch
+from vllm import LLM, SamplingParams
+
+
+def extract_logprobs(logprobs):
+    output = []
+    for lp in logprobs:
+        if lp is not None:
+            output.append(list(lp.values())[0].logprob)
+    return output
+
+
+def calculate_error(a, b):
+    return torch.exp(torch.abs(a - b)).mean().item()
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--model", type=str, nargs="?", default="nvidia/Nemotron-H-8B-Base-8K"
+    )
+    args = parser.parse_args()
+
+    seed = 0
+
+    sampling_params = SamplingParams(
+        temperature=1.0,
+        top_p=1.0,
+        max_tokens=8192,
+        prompt_logprobs=0,
+        logprobs=0,
+        seed=seed,
+    )
+
+    # Examples as of 0.9.1
+    # model="meta-llama/Meta-Llama-3-8B", # pass
+    # model="nvidia/Nemotron-H-8B-Base-8K", # fail
+    # model="ibm-ai-platform/Bamba-9B-v1", # pass
+    llm = LLM(
+        model=args.model,
+        enforce_eager=True,
+        trust_remote_code=True,
+        enable_prefix_caching=False,
+        enable_chunked_prefill=False,
+        tensor_parallel_size=2,
+        gpu_memory_utilization=0.8,
+        seed=seed,
+    )
+
+    num_batches = 2
+
+    prompts = [
+        "Hello, my name is",
+        "The president of the United States is",
+        "The capital of France is",
+        "The future of AI is",
+    ]
+
+    outputs = llm.generate(prompts * num_batches, sampling_params)
+
+    for i, output in enumerate(outputs):
+        sequence = output.prompt_token_ids + list(output.outputs[0].token_ids)
+        prompt_logprobs = extract_logprobs(output.prompt_logprobs)
+        logprobs = extract_logprobs(output.outputs[0].logprobs)
+        decode_lp = prompt_logprobs + logprobs
+        decode_lp = torch.tensor(decode_lp)
+
+        sampling_params = SamplingParams(
+            temperature=0.0, max_tokens=1, prompt_logprobs=0
+        )
+        score = llm.generate({"prompt_token_ids": sequence}, sampling_params)
+
+        prefill_lp = extract_logprobs(score[0].prompt_logprobs)
+        prefill_lp = torch.tensor(prefill_lp)
+
+        lp_error = calculate_error(decode_lp, prefill_lp)
+        max_abs_error = torch.abs(decode_lp - prefill_lp).max().item()
+        print(
+            f"Processed sequence length {len(sequence)} with lp error {lp_error} and max abs error {max_abs_error}"
+        )
+        assert lp_error < 1.05, f"lp error is higher than expected (1.0636): {lp_error}"
+
+    print(f"[{args.model}] ALL GOOD!")
+
+
+if __name__ == "__main__":
+    main()