QuantID PRO
quantid
AI & ML interests
None yet
Recent Activity
liked a Space about 2 hours ago
VIDraft/ai-media liked a model about 2 hours ago
FINAL-Bench/VKORE-46M-NPU reacted to SeaWolf-AI's post with ๐ about 2 hours ago
๐ We pretrained a language model from scratch on an inference-only NPU, with real backpropagation.
๐ค Model: https://huggingface.co/FINAL-Bench/VKORE-46M-NPU ๐ 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 ยท SwiGLU ยท RMSNorm ยท GQA), char-level, TinyStories.
๐ 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. ยท Loss trajectories correlate at 0.99 ยท 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.
๐งฉ One accelerator now covers pretraining ยท LoRA ยท inference ยท quantization. We call this layer VKORE: "CUDA for NPUs."
โ๏ธ 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. ๐Organizations
None yet
Hugging Face Official Benchmarks: Map and Participation
#11 opened 8 days ago
by quantid
Darwin-180B-RSI
#2 opened 15 days ago
by quantid
Darwin-180B-RSI
#2 opened 15 days ago
by quantid