[Core] Update PyTorch to 2.14.0, torchvision to 0.29.0, triton to 3.8.0 (release channel) (#52183)

Signed-off-by: Andrey Talman <atalman@users.noreply.github.com>
Signed-off-by: Andrey Talman <atalman@fb.com>
Co-authored-by: Andrey Talman <atalman@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Andrey Talman
2026-10-14 00:17:59 -07:00
committed by GitHub
co-authored by Andrey Talman Claude Opus 5.5
parent ec317a4618
commit 30e79c3a76
16 changed files with 36 additions and 27 deletions
@@ -342,10 +342,13 @@ def test_full_cuda_graph(
example_prompts, max_tokens, num_logprobs
)
# Leave headroom for the HF model that ran earlier in this process on a
# 32.5 GiB MIG.
with vllm_runner(
model,
max_num_seqs=MAX_NUM_SEQS,
attention_backend=ATTN_BACKEND,
gpu_memory_utilization=0.9,
enable_chunked_prefill=True,
) as vllm_model:
vllm_outputs = vllm_model.generate_greedy_logprobs(
@@ -74,6 +74,12 @@ def test_kv_sharing_fast_prefill(
"ROCm: torch.compile produces incorrect output for gemma-3n's GELU "
"with tanh approximation. Use enforce_eager=True instead."
)
if not enforce_eager and not use_v2_model_runner and current_platform.is_cuda():
pytest.skip(
"cublasCreate fails with CUBLAS_STATUS_ALLOC_FAILED during memory "
"profiling with torch 2.14: "
"https://github.com/pytorch/pytorch/issues/199944"
)
sampling_params = SamplingParams(temperature=0.0, max_tokens=100)
compilation_config = CompilationConfig(