HARP Quantized Models Quantized Llama 2 (7B/13B/70B) at 2-bit via HARP, a learnable orthogonal preprocessor for extreme LLM quantization. EMNLP 2026. brain-lab/Llama-2-7b-QuIP-HARP-2Bit Text Generation • 0.6B • Updated Aug 30 • 19 brain-lab/Llama-2-13b-QuIP-HARP-2Bit Text Generation • 0.9B • Updated Aug 30 • 17 • 1 brain-lab/Llama-2-70b-QuIP-HARP-2Bit Text Generation • 3B • Updated Aug 30 • 18
Gradient-Faithful Surrogates paper models Final checkpoints of Llama3 and Qwen3 models from Gradient-Faithful Surrogates for KV-Cache Quantization Recovery paper brain-lab/kivi-llama-gradient-faithful 8B • Updated Aug 19 • 7 brain-lab/qjl-llama-gradient-faithful 8B • Updated Aug 19 • 4 brain-lab/kivi-qwen3-gradient-faithful 4B • Updated Aug 19 • 7 brain-lab/qjl-qwen3-gradient-faithful 4B • Updated Aug 19 • 4
HARP Quantized Models Quantized Llama 2 (7B/13B/70B) at 2-bit via HARP, a learnable orthogonal preprocessor for extreme LLM quantization. EMNLP 2026. brain-lab/Llama-2-7b-QuIP-HARP-2Bit Text Generation • 0.6B • Updated Aug 30 • 19 brain-lab/Llama-2-13b-QuIP-HARP-2Bit Text Generation • 0.9B • Updated Aug 30 • 17 • 1 brain-lab/Llama-2-70b-QuIP-HARP-2Bit Text Generation • 3B • Updated Aug 30 • 18
Gradient-Faithful Surrogates paper models Final checkpoints of Llama3 and Qwen3 models from Gradient-Faithful Surrogates for KV-Cache Quantization Recovery paper brain-lab/kivi-llama-gradient-faithful 8B • Updated Aug 19 • 7 brain-lab/qjl-llama-gradient-faithful 8B • Updated Aug 19 • 4 brain-lab/kivi-qwen3-gradient-faithful 4B • Updated Aug 19 • 7 brain-lab/qjl-qwen3-gradient-faithful 4B • Updated Aug 19 • 4
