
Post-Training Quantization—INT8 Compression for Faster Inference
Delivery in
4 days
- Views 3
Amount of days required to complete work for this Offer as set by the freelancer.
Rating of the Offer as calculated from other buyers' reviews.
Average time for the freelancer to first reply on the workstream after purchase or contact on this Offer.
What you get with this Offer
I will apply post-training INT8 quantization to your model — calibrating the quantization parameters on a representative dataset, applying symmetric or asymmetric quantization per layer, and delivering an INT8 model with quality evaluation and inference latency benchmarking. Post-training INT8 quantization is the most reliable quantization approach for production deployment — it requires no retraining, calibration takes minutes rather than hours, and INT8 quantization of most model architectures produces less than 1% quality degradation while reducing model size by 2x and improving inference throughput by 1.5-2x on hardware with INT8 arithmetic acceleration.
The quantization covers calibration dataset selection, per-layer quantization calibration, INT8 model generation, quality evaluation on your task, latency benchmarking comparing INT8 against FP32 and FP16, and delivery of the quantized model.
The quantization covers calibration dataset selection, per-layer quantization calibration, INT8 model generation, quality evaluation on your task, latency benchmarking comparing INT8 against FP32 and FP16, and delivery of the quantized model.
What the Freelancer needs to start the work
Please share your model (Hugging Face or local checkpoint), your calibration dataset, your deployment hardware, your task and quality evaluation, and your latency requirements.
We collect cookies to enable the proper functioning and security of our website, and to enhance your experience. By clicking on 'Accept All Cookies', you consent to the use of these cookies. You can change your 'Cookies Settings' at any time. For more information, please read ourCookie Policy
Cookie Settings
Accept All Cookies