coffee/DeepInfer
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepInfer.git synced 2026-10-29 07:17:08 +08:00
Code Issues Packages Projects Releases Wiki Activity
DeepInfer/examples/neuron
History
whn09 cc90926a1c Neuron: full DeepSeek-V4.1-Flash loads and prefills on 64 NeuronCores (EP=64) 
Device-memory fixes found bringing up the released checkpoint (16.45 GiB of
weights per 24 GB core):
- Engram: the Neuron EngramBuffer keeps only this rank's SF rows, so skip the
  all-gather of the replicated SF table (~3 GB per layer, allocated on the
  device -> "Failed to allocate resource").
- gather_tl: index_select / advanced indexing of an int8 table allocates
  several times the table on the device (+4.9 GiB for the 1.3 GB embedding);
  gather through F.embedding on an int32 / fp16 view instead (bit-exact, no
  extra memory).
- deep_gemm FP8/FP4 GEMMs: dequantize the weight in N chunks of 32M elements
  and one expert at a time instead of whole projections / all local experts
  in fp32 (OOM on a fragmented pool); bit-identical results.
- transfer_to_neuron: import torch.utils._device explicitly (missing when
  torch_neuronx is not auto-loaded).

examples/neuron/offline_inference_neuron.py: offline inference in basic mode
(text prompts, no MTP / EPLB, fixed KV pool, max_model_len >= 2048 * 8 for
the candidate-block filter), one process per NeuronCore.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-21 14:57:51 +08:00
..
golden Neuron basic mode — part 3: end-to-end check vs the official reference 2026-10-20 20:20:02 +08:00
check_tiny_checkpoint.py Neuron basic mode — part 3: end-to-end check vs the official reference 2026-10-20 20:20:02 +08:00
compare_golden.py Neuron basic mode — part 3: end-to-end check vs the official reference 2026-10-20 20:20:02 +08:00
device_diff.py Neuron: tiny model on a NeuronCore matches the reference (issues #1, #2, #4) 2026-10-21 13:22:59 +08:00
make_tiny_checkpoint.py Neuron basic mode — part 3: end-to-end check vs the official reference 2026-10-20 20:20:02 +08:00
offline_inference_neuron.py Neuron: full DeepSeek-V4.1-Flash loads and prefills on 64 NeuronCores (EP=64) 2026-10-21 14:57:51 +08:00
probe_layer0.py Neuron: tiny model on a NeuronCore matches the reference (issues #1, #2, #4) 2026-10-21 13:22:59 +08:00
README.md Neuron flash_mla: round attention P to bf16 like the CUDA kernels 2026-10-20 20:48:36 +08:00
ref_kernels_torch.py Neuron basic mode — part 3: end-to-end check vs the official reference 2026-10-20 20:20:02 +08:00
reference_golden.py Neuron basic mode — part 3: end-to-end check vs the official reference 2026-10-20 20:20:02 +08:00
run_deepinfer_tiny.py Neuron: EP > 1 with real neuron collectives (issue #6) 2026-10-21 11:46:59 +08:00
test_ref_kernels_torch.py Neuron basic mode — part 3: end-to-end check vs the official reference 2026-10-20 20:20:02 +08:00

README.md

DeepInfer on AWS Trainium — end-to-end check with a tiny checkpoint

Design and conventions of the port: deepinfer/neuron_ops/NEURON_PORT.md.

The ground truth is DeepSeek's official reference implementation (inference/ in the deepseek-ai/DeepSeek-V4.1-Flash-1017 HF repo) running on CPU, with its TileLang kernels replaced by the pure-torch versions in ref_kernels_torch.py.

Script Purpose
make_tiny_checkpoint.py Random tiny V4.1 checkpoint in the real HF layout (FP8 e4m3 + ue8m0 32x32 scales, FP4 e2m1 experts with 1x32 scales, engram tables)
check_tiny_checkpoint.py DeepInfer loader coverage / shape / dtype checks on the meta device
ref_kernels_torch.py, test_ref_kernels_torch.py Torch replacements for the reference kernel.py
reference_golden.py Reference prefill + greedy decode; saves logits, tokens and per-layer hidden states
run_deepinfer_tiny.py Same prompts through the DeepInfer engine (LLM offline API), teacher-forced
compare_golden.py Candidate vs golden: argmax match, logits rel / cos / KL, per-layer hidden error

Tiny config: 10 layers + 3 nextn, hidden 1024, 8 heads x 512 (MLA, rope 64), 16 experts / top-4, indexer 4 x 128 with top-64, engram on layers 1 and 4, real tokenizer (vocab 129280).

export DEEPINFER_ACCELERATOR=neuron DEEPINFER_NEURON_DEVICE=cpu   # or =neuron for the NeuronCore
python examples/neuron/run_deepinfer_tiny.py --threads 32 --hidden --compare

Results so far (CPU mode, basic mode: eager, no DSpark/MTP, EP=DP=CP=1)

Teacher-forced, prefill + 8 decode steps per prompt; "floor" is the reference compared with itself under a different CPU thread count (FP8 activation quantization amplifies sub-ulp differences).

prompt (tokens) argmax match max logits rel (floor) min cos (floor)
short (11) 9/9 0.054 (0.032) 0.9985 (0.9995)
medium (172, wraps the 128 SWA window) 8/9 * 0.041 (0.039) 0.9992 (0.9993)
long (715, candidate-block filter active) 9/9 0.059 (0.060) 0.9983 (0.9982)

* medium step 6 has a top-1/top-2 margin of 0.0096; the reference flips it against itself too.

probe_layer0.py compares ~35 named ops per layer between the reference and DeepInfer. Layer 0 is bit-exact once three rounding points are aligned:

  • attention P rounded to bf16 before P@V (online softmax, 64-key blocks) — the CUDA FlashMLA kernels and the reference do this; the Neuron flash_mla shim now does too (was fp32, a port bug);
  • SwiGLU output quantized to FP8 straight from fp32 (the reference rounds to bf16 first) and
  • the MoE combine rounding the routed sum to bf16 before adding the shared expert (the reference adds in fp32) — both are DeepInfer's own GPU/Ascend behaviour, kept as is.

The remaining logits difference comes from those two, amplified by FP8 activation quantization layer by layer. Per-step / per-layer numbers before the attention fix: golden/deepinfer_cpu_vs_golden.json.