fix(model): replace INVALID_TOKEN sentinel with real first-token decode in prefill for whisper models

[rebel-eunji]

🚀 Summary of Changes

Problem

When running whisper models through vLLM serve , and a client POSTs to /v1/audio/transcriptions, vLLM raises the following error:

torch._dynamo.exc.TorchRuntimeError: Dynamo failed to run FX node with fake tensors:
call_method masked_fill_(*(FakeTensor(..., size=(1, 51966)),
                           FakeTensor(..., size=(1, 51866), dtype=torch.bool), -inf), **{}):
got RuntimeError('Attempting to broadcast a dimension of length 51866 at -1!
                  Mismatching argument at index 1 had torch.Size([1, 51866]);
                  but expected shape should be broadcastable to [1, 51966]')

from user code:
  File ".../vllm_rbln/v1/sample/rbln_sampler.py", line 275, in forward
    logits = self.apply_logits_processors(...)
  File ".../vllm/v1/sample/sampler.py", line 288, in apply_logits_processors
    logits.masked_fill_(sampling_metadata.allowed_token_ids_mask, float("-inf"))

Solution

Replace the INVALID_TOKEN sentinel trick in Whisper prefill: now run the encoder and the first decoder step with
decoder_start_token_id to return real first-token logits to the sampler.

---

📌 Related Issues / Tickets

• Resolves #
• Related to #

---

✅ Type of Change

• 🚀 Release (release)
• ✨ Feature (feature)
• 🧠 Model support (model)
• 🧬 Core engine changes (core)
• 🛠 Bug fix (fix)
• ⚙️ Performance improvement (perf)
• 🔁 Refactor or code cleanup (refactor)
• 📄 Documentation (docs)
• ❓ Other (other): please describe

---

🧪 How to Test

1. Run ...
2. Verify output: ...
3. Edge case tested: ...

---

📸 Screenshots / Logs (if applicable)

---

📋 Checklist

• PR title follows Conventional Commits format
• This PR is linked to an existing issue
• The test method is described, and the expected result is clearly stated
• Relevant documentation has been updated (if applicable)

---

💬 Notes

---

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!