Proto_AGI
mayafree
AI & ML interests
None yet
Recent Activity
reacted to SeaWolf-AI's post with π 3 days ago
π We pretrained a language model from scratch on an inference-only NPU, with real backpropagation.
π€ Model: https://huggingface.co/FINAL-Bench/VKORE-46M-NPU π Technical write-up: https://huggingface.co/blog/FINAL-Bench/backpropagation-on-an-inference-only-npu-pretraini
"NPUs are inference-only, you can't train on them." We built a counterexample: VKORE-46M-NPU, trained from scratch on a first-gen (2025) inference NPU, 32 chips, no pretrained weights. A 46M decoder-only model (RoPE Β· SwiGLU Β· RMSNorm Β· GQA), char-level, TinyStories.
π Why trust it: a falling-loss curve proves nothing. So we trained the same model under identical init, data, and schedule on the NPU and on a GPU reference. Β· Loss trajectories correlate at 0.99 Β· Throughput converged to a comparable level The NPU backward pass and optimizer updates do numerically the same work as the GPU, not an approximation that merely trends down.
𧩠One accelerator now covers pretraining · LoRA · inference · quantization. We call this layer VKORE: "CUDA for NPUs."
βοΈ Public: weights, model card, NPU-vs-GPU comparison. Private (patent-pending, paper to follow): the core training technique. 46M char-level is toy scale on purpose: it proves the training path, not model quality. Larger models and real data are next.
Feedback from folks working on custom autograd backends and low-precision training kernels on non-GPU accelerators is very welcome. π reacted to SeaWolf-AI's post with π 3 days ago
π We pretrained a language model from scratch on an inference-only NPU, with real backpropagation.
π€ Model: https://huggingface.co/FINAL-Bench/VKORE-46M-NPU π Technical write-up: https://huggingface.co/blog/FINAL-Bench/backpropagation-on-an-inference-only-npu-pretraini
"NPUs are inference-only, you can't train on them." We built a counterexample: VKORE-46M-NPU, trained from scratch on a first-gen (2025) inference NPU, 32 chips, no pretrained weights. A 46M decoder-only model (RoPE Β· SwiGLU Β· RMSNorm Β· GQA), char-level, TinyStories.
π Why trust it: a falling-loss curve proves nothing. So we trained the same model under identical init, data, and schedule on the NPU and on a GPU reference. Β· Loss trajectories correlate at 0.99 Β· Throughput converged to a comparable level The NPU backward pass and optimizer updates do numerically the same work as the GPU, not an approximation that merely trends down.
𧩠One accelerator now covers pretraining · LoRA · inference · quantization. We call this layer VKORE: "CUDA for NPUs."
βοΈ Public: weights, model card, NPU-vs-GPU comparison. Private (patent-pending, paper to follow): the core training technique. 46M char-level is toy scale on purpose: it proves the training path, not model quality. Larger models and real data are next.
Feedback from folks working on custom autograd backends and low-precision training kernels on non-GPU accelerators is very welcome. π upvoted an article 3 days ago
Backpropagation on an Inference-Only NPU: Pretraining VKORE-46M-NPU From ScratchOrganizations
Request for Correction Regarding the A.X K2 Entry
2
#1 opened 2 months ago
by 
Minsang