fix: add dtype guard to FlashAttnBackend capability check

- _backend_supports now rejects fp32 for FlashAttnBackend (flash-attn only supports fp16/bf16), preventing runtime crash on fallback chain - rename test_default_backend_is_torch_native to reflect multi-backend reality - scheduler test fixture uses bf16 model (matches production, avoids unnecessary 3-step fallback chain)
2026-08-07 23:08:22 +08:00
parent 184fbbce5c
commit 1b1f1a0707
3 changed files with 5 additions and 4 deletions
@@ -180,7 +180,7 @@ def test_scheduler_concurrent_get_stats(mock_model_and_tokenizer):
 def _make_real_scheduler(device):
    """Build a scheduler backed by a tiny real model for run_batch tests."""
    cfg = make_rollout_config(max_position_embeddings=64)
-    model = AutoRegressiveLM(cfg).to(device=device).eval()
+    model = AutoRegressiveLM(cfg).to(device=device, dtype=torch.bfloat16).eval()
    tokenizer = FakeTokenizer()
    scheduler = InferenceScheduler(
        model=model,