#NPU
Model: thuml/sundial-base-128m (Sundial / Timer 3.0) - generative time-series foundation model, 128M parameters, decoder-only Transformer 12 layers, hidden 768, patch length 16, max context 2880, multi-patch prediction length 720, pre-trained on 1T time points. Licensed Apache-2.0.
Revision: 3212e42564493f520593e5414af4367fc4b49226 (snapshot_download, weight model.safetensors 490MB, FP32).
Task: Time-series forecasting - point and probabilistic generation via TimeFlow (flow-matching). This submission validates zero-shot point forecasting context -> horizon on Ascend NPU.
Architecture: SundialPatchEmbedding + SundialModel (12 decoder layers with RoPE, SDPA) + FlowLoss sampling (SimpleMLPAdaLN, 50 steps).
torch.FloatTensor [batch, context_length] - continuous univariate series, no tokenizer, float32. Deterministic synthetic sample used: batch=2, context_length=512, sine/cosine + Gaussian noise (seed 0), shape [2,512], mean 0.0106, std 0.7103. Axes fixed as [B,T], no channel transpose, no mask.forecast [batch, num_samples, horizon] where horizon=96 (configurable up to 720). Example horizon=96, num_samples=1 -> [2,1,96], float32, finite.input_token_len=16 with masking.~0.7MB, no CSV needed; supports arbitrary [B,T] where T %16 padded left.npu-smi shows 2 devices Ascend910_9362, npu:0 used, 64GB HBM./usr/local/Ascend/cann-8.5.1).HF_ENDPOINT=https://hf-mirror.com via env, no manual proxy.pip install -r requirements.txt
# requirements: torch==2.9.0, torch-npu==2.9.0.post1, transformers==4.40.1, numpy==1.26.4, safetensors, accelerate
# Model auto-downloaded from HF mirror on first run (~490MB), cached via huggingface_hub snapshot_downloadPYTHONPATH=/opt/atomgit/adapt-npu-agent/working_timer/venv/lib/python3.11/site-packages:$PYTHONPATH python inference.py
# or plain: python inference.py (if transformers==4.40.1 installed)thuml/sundial-base-128m (or local /tmp/sundial-base-128m if present) with trust_remote_code=True.npu:0 explicitly, verifies next(model.parameters()).device == npu:0.model.generate(inputs, max_new_tokens=96, num_samples=1) with torch.manual_seed(42) for determinism, torch.npu.synchronize() before/after timing.[2,512] float32, sample seq[0][:5]=[0.1764,0.1653,0.3465,0.5922,0.6685][2,1,96] float32 on npu:0, sample forecast[0,0,:5]=[1.0704,1.0557,1.0188,0.8996,0.8898], mean -0.0730, std 0.6876, min -1.0368, max 1.0704, finite True.npu:0, input npu:0, output npu:0, torch.npu.is_available()==True, device_count=2, Ascend910_9362.3212e425...), same input, same seed 42, same dtype float32, same code path generate. Saved cpu.npy [2,1,96] and npu.npy [2,1,96] compared via scripts/compare_outputs.py --task forecasting --atol 1e-4 --rtol 1e-3 -> PASS. max_abs_error 7.15e-07, mean_abs_error 2.20e-07, finite True, allclose True.get_data ~0.001s).node scripts/render_xterm_evidence.mjs --style raw (not screenshots):

output_token_lens=[720]; longer horizons require autoregressive loop.num_sampling_steps trades quality for speed (not tuned).DynamicCache.seen_tokens and get_usable_length); 4.57+ requires patches.