๐ข Open-sourcing the Sarvam 30B and 105B models! Trained from scratch with all data, model research and inference optimisation done in-house, these models punch above their weight in most global benchmarks plus excel in Indian languages.
Get the weights at Hugging Face and AIKosh. Thanks to the good folks at SGLang for day 0 support, vLLM support coming soon. Links, benchmark scores, examples, and more in our blog - https://t.co/DcCG3zlN8p
AMD Ryzen chief teases they are bringing back Zen 3 because of the DDR5 shortage.
This is on the heels of nVidia rumored to be bringing back the 3060.
We going backwards like itโs 2020. ๐ญ
@PINTO03091 My goal is to find the fastest model for C++/TensorRT inference with dynamic batching, strictly under an Apache 2.0 license. Which models should i be looking into as a potential solution ?
@PINTO03091 Greetings! I would like your opinion on selecting models that are compatible with Tensorrt dynamic batch size. I want to deploy a nano model in c++ Tensorrt for basic detection tasks such as vehicles (1/3)
@PINTO03091 I've had good success with RF-DETR Nano for vehicle detection, but I'm unable to export it with dynamic batches. My understanding is that many DETR-based architectures have internal complexities that make dynamic batching for TensorRT deployment a challenge. Any Advice?
@PINTO03091 Understood. I will have to retrain and try. Where can i find this deimv2_hgnetv2_n_wholebody34 weights? or can i retrain with official weights as base ?
@PINTO03091 When exporting with its post-processor, we got a single label_xyxy_score output of shape [N, 300, 6], treating this as [class_id, x_min, y_min, x_max, y_max, score] with normalized [0,1] bbox coordinates and handling all post-processing externally.