pinned
Paused
Agents
54
FLUX VisionReply
馃
FLUX, Image to Texto to Image, VLM
AIQ
aiqcamp
AI & ML interests
None yet
Recent Activity
liked a Space 1 day ago
VIDraft/ai-media liked a model 1 day ago
FINAL-Bench/VKORE-46M-NPU reacted to SeaWolf-AI's post with 馃殌 1 day ago
馃殌 We pretrained a language model from scratch on an inference-only NPU, with real backpropagation.
馃 Model: https://huggingface.co/FINAL-Bench/VKORE-46M-NPU 馃摑 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 路 SwiGLU 路 RMSNorm 路 GQA), char-level, TinyStories.
馃搹 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. 路 Loss trajectories correlate at 0.99 路 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.
馃З One accelerator now covers pretraining 路 LoRA 路 inference 路 quantization. We call this layer VKORE: "CUDA for NPUs."
鈿栵笍 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. 馃檶Organizations
