llama-2-13b-chat-nf4-fp16-upscaled

Model overview

llama-2-13b-chat-nf4-fp16-upscaled is published by arnavgrg as a Llama-based text-generation model. The captured configuration identifies LlamaForCausalLM and Safetensors metadata reports 13015864320 parameters. According to the model card, it derives from Meta's Llama-2-13b-chat after NF4 4-bit quantization via bitsandbytes, upscaled to FP16.

Recorded capabilities

Use cases in the source record

• Conversational text-generation experiments that load the documented FP16 weights with Transformers.
• Quantization-comparison work studying the publisher-described lossy NF4 lineage against the official base.

Limitations and unknowns

• No evaluation results were extracted from this record.
• No context-window value was extracted from this record.
• Quantization quality is publisher-described as lossy and has not been independently verified by Ethen.
• Provider state is historical snapshot data, not independently refreshed current availability.

Source and provenance

Source: arnavgrg/llama-2-13b-chat-nf4-fp16-upscaled

Captured: Unknown. Processed: 2026-09-07T19:34:40.327035+00:00.

This is an upscaled fp16 variant of the original Llama-2-13b-chat base model by Meta after it has been loaded with nf4 4-bit quantization via bitsandbytes. The main idea here is to upscale the linear4bit layers to fp16 so that the quantization/dequantization cost doesn't have to paid for each forward pass at inference time. Note: The quantization operation to nf4 is not lossless, so the model weights for the linear layers are lossy, which means that this model will not work as well as the official base model. To use this model, you can just load it via transformers in fp16: import torch from transformers import AutoModelForCausalLM…