{ "architectures": [ "KimiK3ForConditionalGeneration" ], "auto_map": { "AutoConfig": "configuration_kimi_k3.KimiK3Config", "AutoModel": "modeling_kimi_k3.KimiK3ForConditionalGeneration", "AutoModelForCausalLM": "modeling_kimi_k3.KimiK3ForConditionalGeneration" }, "bos_token_id": 163584, "dtype": "bfloat16", "eos_token_id": 163586, "ignore_index": -100, "image_placeholder": "<|kimi_image_placeholder|>", "media_placeholder_token_id": 163605, "model_type": "kimi_k3", "pad_token_id": 163839, "quantization_config": { "config_groups": { "group_0": { "input_activations": { "dynamic": false, "group_size": 16, "num_bits": 4, "type": "float" }, "targets": [ "language_model.model.layers.1.block_sparse_moe.experts", "language_model.model.layers.1.mlp.experts", "language_model.model.layers.10.block_sparse_moe.experts", "language_model.model.layers.10.mlp.experts", "language_model.model.layers.11.block_sparse_moe.experts", "language_model.model.layers.11.mlp.experts", "language_model.model.layers.12.block_sparse_moe.experts", "language_model.model.layers.12.mlp.experts", "language_model.model.layers.13.block_sparse_moe.experts", "language_model.model.layers.13.mlp.experts", "language_model.model.layers.14.block_sparse_moe.experts", "language_model.model.layers.14.mlp.experts", "language_model.model.layers.15.block_sparse_moe.experts", "language_model.model.layers.15.mlp.experts", "language_model.model.layers.16.block_sparse_moe.experts", "language_model.model.layers.16.mlp.experts", "language_model.model.layers.17.block_sparse_moe.experts", "language_model.model.layers.17.mlp.experts", "language_model.model.layers.18.block_sparse_moe.experts", "language_model.model.layers.18.mlp.experts", "language_model.model.layers.19.block_sparse_moe.experts", "language_model.model.layers.19.mlp.experts", "language_model.model.layers.2.block_sparse_moe.experts", "language_model.model.layers.2.mlp.experts", "language_model.model.layers.20.block_sparse_moe.experts", "language_model.model.layers.20.mlp.experts", "language_model.model.layers.21.block_sparse_moe.experts", "language_model.model.layers.21.mlp.experts", "language_model.model.layers.22.block_sparse_moe.experts", "language_model.model.layers.22.mlp.experts", "language_model.model.layers.23.block_sparse_moe.experts", "language_model.model.layers.23.mlp.experts", "language_model.model.layers.24.block_sparse_moe.experts", "language_model.model.layers.24.mlp.experts", "language_model.model.layers.25.block_sparse_moe.experts", "language_model.model.layers.25.mlp.experts", "language_model.model.layers.26.block_sparse_moe.experts", "language_model.model.layers.26.mlp.experts", "language_model.model.layers.27.block_sparse_moe.experts", "language_model.model.layers.27.mlp.experts", "language_model.model.layers.28.block_sparse_moe.experts", "language_model.model.layers.28.mlp.experts", "language_model.model.layers.29.block_sparse_moe.experts", "language_model.model.layers.29.mlp.experts", "language_model.model.layers.3.block_sparse_moe.experts", "language_model.model.layers.3.mlp.experts", "language_model.model.layers.30.block_sparse_moe.experts", "language_model.model.layers.30.mlp.experts", "language_model.model.layers.31.block_sparse_moe.experts", "language_model.model.layers.31.mlp.experts", "language_model.model.layers.32.block_sparse_moe.experts", "language_model.model.layers.32.mlp.experts", "language_model.model.layers.33.block_sparse_moe.experts", "language_model.model.layers.33.mlp.experts", "language_model.model.layers.34.block_sparse_moe.experts", "language_model.model.layers.34.mlp.experts", "language_model.model.layers.35.block_sparse_moe.experts", "language_model.model.layers.35.mlp.experts", "language_model.model.layers.36.block_sparse_moe.experts", "language_model.model.layers.36.mlp.experts", "language_model.model.layers.37.block_sparse_moe.experts", "language_model.model.layers.37.mlp.experts", "language_model.model.layers.38.block_sparse_moe.experts", "language_model.model.layers.38.mlp.experts", "language_model.model.layers.39.block_sparse_moe.experts", "language_model.model.layers.39.mlp.experts", "language_model.model.layers.4.block_sparse_moe.experts", "language_model.model.layers.4.mlp.experts", "language_model.model.layers.40.block_sparse_moe.experts", "language_model.model.layers.40.mlp.experts", "language_model.model.layers.41.block_sparse_moe.experts", "language_model.model.layers.41.mlp.experts", "language_model.model.layers.42.block_sparse_moe.experts", "language_model.model.layers.42.mlp.experts", "language_model.model.layers.43.block_sparse_moe.experts", "language_model.model.layers.43.mlp.experts", "language_model.model.layers.44.block_sparse_moe.experts", "language_model.model.layers.44.mlp.experts", "language_model.model.layers.45.block_sparse_moe.experts", "language_model.model.layers.45.mlp.experts", "language_model.model.layers.46.block_sparse_moe.experts", "language_model.model.layers.46.mlp.experts", "language_model.model.layers.47.block_sparse_moe.experts", "language_model.model.layers.47.mlp.experts", "language_model.model.layers.48.block_sparse_moe.experts", "language_model.model.layers.48.mlp.experts", "language_model.model.layers.49.block_sparse_moe.experts", "language_model.model.layers.49.mlp.experts", "language_model.model.layers.5.block_sparse_moe.experts", "language_model.model.layers.5.mlp.experts", "language_model.model.layers.50.block_sparse_moe.experts", "language_model.model.layers.50.mlp.experts", "language_model.model.layers.51.block_sparse_moe.experts", "language_model.model.layers.51.mlp.experts", "language_model.model.layers.52.block_sparse_moe.experts", "language_model.model.layers.52.mlp.experts", "language_model.model.layers.53.block_sparse_moe.experts", "language_model.model.layers.53.mlp.experts", "language_model.model.layers.54.block_sparse_moe.experts", "language_model.model.layers.54.mlp.experts", "language_model.model.layers.55.block_sparse_moe.experts", "language_model.model.layers.55.mlp.experts", "language_model.model.layers.56.block_sparse_moe.experts", "language_model.model.layers.56.mlp.experts", "language_model.model.layers.57.block_sparse_moe.experts", "language_model.model.layers.57.mlp.experts", "language_model.model.layers.58.block_sparse_moe.experts", "language_model.model.layers.58.mlp.experts", "language_model.model.layers.59.block_sparse_moe.experts", "language_model.model.layers.59.mlp.experts", "language_model.model.layers.6.block_sparse_moe.experts", "language_model.model.layers.6.mlp.experts", "language_model.model.layers.60.block_sparse_moe.experts", "language_model.model.layers.60.mlp.experts", "language_model.model.layers.61.block_sparse_moe.experts", "language_model.model.layers.61.mlp.experts", "language_model.model.layers.62.block_sparse_moe.experts", "language_model.model.layers.62.mlp.experts", "language_model.model.layers.63.block_sparse_moe.experts", "language_model.model.layers.63.mlp.experts", "language_model.model.layers.64.block_sparse_moe.experts", "language_model.model.layers.64.mlp.experts", "language_model.model.layers.65.block_sparse_moe.experts", "language_model.model.layers.65.mlp.experts", "language_model.model.layers.66.block_sparse_moe.experts", "language_model.model.layers.66.mlp.experts", "language_model.model.layers.67.block_sparse_moe.experts", "language_model.model.layers.67.mlp.experts", "language_model.model.layers.68.block_sparse_moe.experts", "language_model.model.layers.68.mlp.experts", "language_model.model.layers.69.block_sparse_moe.experts", "language_model.model.layers.69.mlp.experts", "language_model.model.layers.7.block_sparse_moe.experts", "language_model.model.layers.7.mlp.experts", "language_model.model.layers.70.block_sparse_moe.experts", "language_model.model.layers.70.mlp.experts", "language_model.model.layers.71.block_sparse_moe.experts", "language_model.model.layers.71.mlp.experts", "language_model.model.layers.72.block_sparse_moe.experts", "language_model.model.layers.72.mlp.experts", "language_model.model.layers.73.block_sparse_moe.experts", "language_model.model.layers.73.mlp.experts", "language_model.model.layers.74.block_sparse_moe.experts", "language_model.model.layers.74.mlp.experts", "language_model.model.layers.75.block_sparse_moe.experts", "language_model.model.layers.75.mlp.experts", "language_model.model.layers.76.block_sparse_moe.experts", "language_model.model.layers.76.mlp.experts", "language_model.model.layers.77.block_sparse_moe.experts", "language_model.model.layers.77.mlp.experts", "language_model.model.layers.78.block_sparse_moe.experts", "language_model.model.layers.78.mlp.experts", "language_model.model.layers.79.block_sparse_moe.experts", "language_model.model.layers.79.mlp.experts", "language_model.model.layers.8.block_sparse_moe.experts", "language_model.model.layers.8.mlp.experts", "language_model.model.layers.80.block_sparse_moe.experts", "language_model.model.layers.80.mlp.experts", "language_model.model.layers.81.block_sparse_moe.experts", "language_model.model.layers.81.mlp.experts", "language_model.model.layers.82.block_sparse_moe.experts", "language_model.model.layers.82.mlp.experts", "language_model.model.layers.83.block_sparse_moe.experts", "language_model.model.layers.83.mlp.experts", "language_model.model.layers.84.block_sparse_moe.experts", "language_model.model.layers.84.mlp.experts", "language_model.model.layers.85.block_sparse_moe.experts", "language_model.model.layers.85.mlp.experts", "language_model.model.layers.86.block_sparse_moe.experts", "language_model.model.layers.86.mlp.experts", "language_model.model.layers.87.block_sparse_moe.experts", "language_model.model.layers.87.mlp.experts", "language_model.model.layers.88.block_sparse_moe.experts", "language_model.model.layers.88.mlp.experts", "language_model.model.layers.89.block_sparse_moe.experts", "language_model.model.layers.89.mlp.experts", "language_model.model.layers.9.block_sparse_moe.experts", "language_model.model.layers.9.mlp.experts", "language_model.model.layers.90.block_sparse_moe.experts", "language_model.model.layers.90.mlp.experts", "language_model.model.layers.91.block_sparse_moe.experts", "language_model.model.layers.91.mlp.experts", "language_model.model.layers.92.block_sparse_moe.experts", "language_model.model.layers.92.mlp.experts", "layers.1.block_sparse_moe.experts", "layers.1.mlp.experts", "layers.10.block_sparse_moe.experts", "layers.10.mlp.experts", "layers.11.block_sparse_moe.experts", "layers.11.mlp.experts", "layers.12.block_sparse_moe.experts", "layers.12.mlp.experts", "layers.13.block_sparse_moe.experts", "layers.13.mlp.experts", "layers.14.block_sparse_moe.experts", "layers.14.mlp.experts", "layers.15.block_sparse_moe.experts", "layers.15.mlp.experts", "layers.16.block_sparse_moe.experts", "layers.16.mlp.experts", "layers.17.block_sparse_moe.experts", "layers.17.mlp.experts", "layers.18.block_sparse_moe.experts", "layers.18.mlp.experts", "layers.19.block_sparse_moe.experts", "layers.19.mlp.experts", "layers.2.block_sparse_moe.experts", "layers.2.mlp.experts", "layers.20.block_sparse_moe.experts", "layers.20.mlp.experts", "layers.21.block_sparse_moe.experts", "layers.21.mlp.experts", "layers.22.block_sparse_moe.experts", "layers.22.mlp.experts", "layers.23.block_sparse_moe.experts", "layers.23.mlp.experts", "layers.24.block_sparse_moe.experts", "layers.24.mlp.experts", "layers.25.block_sparse_moe.experts", "layers.25.mlp.experts", "layers.26.block_sparse_moe.experts", "layers.26.mlp.experts", "layers.27.block_sparse_moe.experts", "layers.27.mlp.experts", "layers.28.block_sparse_moe.experts", "layers.28.mlp.experts", "layers.29.block_sparse_moe.experts", "layers.29.mlp.experts", "layers.3.block_sparse_moe.experts", "layers.3.mlp.experts", "layers.30.block_sparse_moe.experts", "layers.30.mlp.experts", "layers.31.block_sparse_moe.experts", "layers.31.mlp.experts", "layers.32.block_sparse_moe.experts", "layers.32.mlp.experts", "layers.33.block_sparse_moe.experts", "layers.33.mlp.experts", "layers.34.block_sparse_moe.experts", "layers.34.mlp.experts", "layers.35.block_sparse_moe.experts", "layers.35.mlp.experts", "layers.36.block_sparse_moe.experts", "layers.36.mlp.experts", "layers.37.block_sparse_moe.experts", "layers.37.mlp.experts", "layers.38.block_sparse_moe.experts", "layers.38.mlp.experts", "layers.39.block_sparse_moe.experts", "layers.39.mlp.experts", "layers.4.block_sparse_moe.experts", "layers.4.mlp.experts", "layers.40.block_sparse_moe.experts", "layers.40.mlp.experts", "layers.41.block_sparse_moe.experts", "layers.41.mlp.experts", "layers.42.block_sparse_moe.experts", "layers.42.mlp.experts", "layers.43.block_sparse_moe.experts", "layers.43.mlp.experts", "layers.44.block_sparse_moe.experts", "layers.44.mlp.experts", "layers.45.block_sparse_moe.experts", "layers.45.mlp.experts", "layers.46.block_sparse_moe.experts", "layers.46.mlp.experts", "layers.47.block_sparse_moe.experts", "layers.47.mlp.experts", "layers.48.block_sparse_moe.experts", "layers.48.mlp.experts", "layers.49.block_sparse_moe.experts", "layers.49.mlp.experts", "layers.5.block_sparse_moe.experts", "layers.5.mlp.experts", "layers.50.block_sparse_moe.experts", "layers.50.mlp.experts", "layers.51.block_sparse_moe.experts", "layers.51.mlp.experts", "layers.52.block_sparse_moe.experts", "layers.52.mlp.experts", "layers.53.block_sparse_moe.experts", "layers.53.mlp.experts", "layers.54.block_sparse_moe.experts", "layers.54.mlp.experts", "layers.55.block_sparse_moe.experts", "layers.55.mlp.experts", "layers.56.block_sparse_moe.experts", "layers.56.mlp.experts", "layers.57.block_sparse_moe.experts", "layers.57.mlp.experts", "layers.58.block_sparse_moe.experts", "layers.58.mlp.experts", "layers.59.block_sparse_moe.experts", "layers.59.mlp.experts", "layers.6.block_sparse_moe.experts", "layers.6.mlp.experts", "layers.60.block_sparse_moe.experts", "layers.60.mlp.experts", "layers.61.block_sparse_moe.experts", "layers.61.mlp.experts", "layers.62.block_sparse_moe.experts", "layers.62.mlp.experts", "layers.63.block_sparse_moe.experts", "layers.63.mlp.experts", "layers.64.block_sparse_moe.experts", "layers.64.mlp.experts", "layers.65.block_sparse_moe.experts", "layers.65.mlp.experts", "layers.66.block_sparse_moe.experts", "layers.66.mlp.experts", "layers.67.block_sparse_moe.experts", "layers.67.mlp.experts", "layers.68.block_sparse_moe.experts", "layers.68.mlp.experts", "layers.69.block_sparse_moe.experts", "layers.69.mlp.experts", "layers.7.block_sparse_moe.experts", "layers.7.mlp.experts", "layers.70.block_sparse_moe.experts", "layers.70.mlp.experts", "layers.71.block_sparse_moe.experts", "layers.71.mlp.experts", "layers.72.block_sparse_moe.experts", "layers.72.mlp.experts", "layers.73.block_sparse_moe.experts", "layers.73.mlp.experts", "layers.74.block_sparse_moe.experts", "layers.74.mlp.experts", "layers.75.block_sparse_moe.experts", "layers.75.mlp.experts", "layers.76.block_sparse_moe.experts", "layers.76.mlp.experts", "layers.77.block_sparse_moe.experts", "layers.77.mlp.experts", "layers.78.block_sparse_moe.experts", "layers.78.mlp.experts", "layers.79.block_sparse_moe.experts", "layers.79.mlp.experts", "layers.8.block_sparse_moe.experts", "layers.8.mlp.experts", "layers.80.block_sparse_moe.experts", "layers.80.mlp.experts", "layers.81.block_sparse_moe.experts", "layers.81.mlp.experts", "layers.82.block_sparse_moe.experts", "layers.82.mlp.experts", "layers.83.block_sparse_moe.experts", "layers.83.mlp.experts", "layers.84.block_sparse_moe.experts", "layers.84.mlp.experts", "layers.85.block_sparse_moe.experts", "layers.85.mlp.experts", "layers.86.block_sparse_moe.experts", "layers.86.mlp.experts", "layers.87.block_sparse_moe.experts", "layers.87.mlp.experts", "layers.88.block_sparse_moe.experts", "layers.88.mlp.experts", "layers.89.block_sparse_moe.experts", "layers.89.mlp.experts", "layers.9.block_sparse_moe.experts", "layers.9.mlp.experts", "layers.90.block_sparse_moe.experts", "layers.90.mlp.experts", "layers.91.block_sparse_moe.experts", "layers.91.mlp.experts", "layers.92.block_sparse_moe.experts", "layers.92.mlp.experts", "model.layers.1.block_sparse_moe.experts", "model.layers.1.mlp.experts", "model.layers.10.block_sparse_moe.experts", "model.layers.10.mlp.experts", "model.layers.11.block_sparse_moe.experts", "model.layers.11.mlp.experts", "model.layers.12.block_sparse_moe.experts", "model.layers.12.mlp.experts", "model.layers.13.block_sparse_moe.experts", "model.layers.13.mlp.experts", "model.layers.14.block_sparse_moe.experts", "model.layers.14.mlp.experts", "model.layers.15.block_sparse_moe.experts", "model.layers.15.mlp.experts", "model.layers.16.block_sparse_moe.experts", "model.layers.16.mlp.experts", "model.layers.17.block_sparse_moe.experts", "model.layers.17.mlp.experts", "model.layers.18.block_sparse_moe.experts", "model.layers.18.mlp.experts", "model.layers.19.block_sparse_moe.experts", "model.layers.19.mlp.experts", "model.layers.2.block_sparse_moe.experts", "model.layers.2.mlp.experts", "model.layers.20.block_sparse_moe.experts", "model.layers.20.mlp.experts", "model.layers.21.block_sparse_moe.experts", "model.layers.21.mlp.experts", "model.layers.22.block_sparse_moe.experts", "model.layers.22.mlp.experts", "model.layers.23.block_sparse_moe.experts", "model.layers.23.mlp.experts", "model.layers.24.block_sparse_moe.experts", "model.layers.24.mlp.experts", "model.layers.25.block_sparse_moe.experts", "model.layers.25.mlp.experts", "model.layers.26.block_sparse_moe.experts", "model.layers.26.mlp.experts", "model.layers.27.block_sparse_moe.experts", "model.layers.27.mlp.experts", "model.layers.28.block_sparse_moe.experts", "model.layers.28.mlp.experts", "model.layers.29.block_sparse_moe.experts", "model.layers.29.mlp.experts", "model.layers.3.block_sparse_moe.experts", "model.layers.3.mlp.experts", "model.layers.30.block_sparse_moe.experts", "model.layers.30.mlp.experts", "model.layers.31.block_sparse_moe.experts", "model.layers.31.mlp.experts", "model.layers.32.block_sparse_moe.experts", "model.layers.32.mlp.experts", "model.layers.33.block_sparse_moe.experts", "model.layers.33.mlp.experts", "model.layers.34.block_sparse_moe.experts", "model.layers.34.mlp.experts", "model.layers.35.block_sparse_moe.experts", "model.layers.35.mlp.experts", "model.layers.36.block_sparse_moe.experts", "model.layers.36.mlp.experts", "model.layers.37.block_sparse_moe.experts", "model.layers.37.mlp.experts", "model.layers.38.block_sparse_moe.experts", "model.layers.38.mlp.experts", "model.layers.39.block_sparse_moe.experts", "model.layers.39.mlp.experts", "model.layers.4.block_sparse_moe.experts", "model.layers.4.mlp.experts", "model.layers.40.block_sparse_moe.experts", "model.layers.40.mlp.experts", "model.layers.41.block_sparse_moe.experts", "model.layers.41.mlp.experts", "model.layers.42.block_sparse_moe.experts", "model.layers.42.mlp.experts", "model.layers.43.block_sparse_moe.experts", "model.layers.43.mlp.experts", "model.layers.44.block_sparse_moe.experts", "model.layers.44.mlp.experts", "model.layers.45.block_sparse_moe.experts", "model.layers.45.mlp.experts", "model.layers.46.block_sparse_moe.experts", "model.layers.46.mlp.experts", "model.layers.47.block_sparse_moe.experts", "model.layers.47.mlp.experts", "model.layers.48.block_sparse_moe.experts", "model.layers.48.mlp.experts", "model.layers.49.block_sparse_moe.experts", "model.layers.49.mlp.experts", "model.layers.5.block_sparse_moe.experts", "model.layers.5.mlp.experts", "model.layers.50.block_sparse_moe.experts", "model.layers.50.mlp.experts", "model.layers.51.block_sparse_moe.experts", "model.layers.51.mlp.experts", "model.layers.52.block_sparse_moe.experts", "model.layers.52.mlp.experts", "model.layers.53.block_sparse_moe.experts", "model.layers.53.mlp.experts", "model.layers.54.block_sparse_moe.experts", "model.layers.54.mlp.experts", "model.layers.55.block_sparse_moe.experts", "model.layers.55.mlp.experts", "model.layers.56.block_sparse_moe.experts", "model.layers.56.mlp.experts", "model.layers.57.block_sparse_moe.experts", "model.layers.57.mlp.experts", "model.layers.58.block_sparse_moe.experts", "model.layers.58.mlp.experts", "model.layers.59.block_sparse_moe.experts", "model.layers.59.mlp.experts", "model.layers.6.block_sparse_moe.experts", "model.layers.6.mlp.experts", "model.layers.60.block_sparse_moe.experts", "model.layers.60.mlp.experts", "model.layers.61.block_sparse_moe.experts", "model.layers.61.mlp.experts", "model.layers.62.block_sparse_moe.experts", "model.layers.62.mlp.experts", "model.layers.63.block_sparse_moe.experts", "model.layers.63.mlp.experts", "model.layers.64.block_sparse_moe.experts", "model.layers.64.mlp.experts", "model.layers.65.block_sparse_moe.experts", "model.layers.65.mlp.experts", "model.layers.66.block_sparse_moe.experts", "model.layers.66.mlp.experts", "model.layers.67.block_sparse_moe.experts", "model.layers.67.mlp.experts", "model.layers.68.block_sparse_moe.experts", "model.layers.68.mlp.experts", "model.layers.69.block_sparse_moe.experts", "model.layers.69.mlp.experts", "model.layers.7.block_sparse_moe.experts", "model.layers.7.mlp.experts", "model.layers.70.block_sparse_moe.experts", "model.layers.70.mlp.experts", "model.layers.71.block_sparse_moe.experts", "model.layers.71.mlp.experts", "model.layers.72.block_sparse_moe.experts", "model.layers.72.mlp.experts", "model.layers.73.block_sparse_moe.experts", "model.layers.73.mlp.experts", "model.layers.74.block_sparse_moe.experts", "model.layers.74.mlp.experts", "model.layers.75.block_sparse_moe.experts", "model.layers.75.mlp.experts", "model.layers.76.block_sparse_moe.experts", "model.layers.76.mlp.experts", "model.layers.77.block_sparse_moe.experts", "model.layers.77.mlp.experts", "model.layers.78.block_sparse_moe.experts", "model.layers.78.mlp.experts", "model.layers.79.block_sparse_moe.experts", "model.layers.79.mlp.experts", "model.layers.8.block_sparse_moe.experts", "model.layers.8.mlp.experts", "model.layers.80.block_sparse_moe.experts", "model.layers.80.mlp.experts", "model.layers.81.block_sparse_moe.experts", "model.layers.81.mlp.experts", "model.layers.82.block_sparse_moe.experts", "model.layers.82.mlp.experts", "model.layers.83.block_sparse_moe.experts", "model.layers.83.mlp.experts", "model.layers.84.block_sparse_moe.experts", "model.layers.84.mlp.experts", "model.layers.85.block_sparse_moe.experts", "model.layers.85.mlp.experts", "model.layers.86.block_sparse_moe.experts", "model.layers.86.mlp.experts", "model.layers.87.block_sparse_moe.experts", "model.layers.87.mlp.experts", "model.layers.88.block_sparse_moe.experts", "model.layers.88.mlp.experts", "model.layers.89.block_sparse_moe.experts", "model.layers.89.mlp.experts", "model.layers.9.block_sparse_moe.experts", "model.layers.9.mlp.experts", "model.layers.90.block_sparse_moe.experts", "model.layers.90.mlp.experts", "model.layers.91.block_sparse_moe.experts", "model.layers.91.mlp.experts", "model.layers.92.block_sparse_moe.experts", "model.layers.92.mlp.experts" ], "weights": { "dynamic": false, "group_size": 16, "num_bits": 4, "type": "float" } }, "group_1": { "quant_algo": "FP8_PB_WO", "targets": [ "language_model.model.layers.0.self_attn.b_proj", "language_model.model.layers.0.self_attn.f_a_proj", "language_model.model.layers.0.self_attn.f_b_proj", "language_model.model.layers.0.self_attn.fused_qkvg_proj", "language_model.model.layers.0.self_attn.g_proj", "language_model.model.layers.0.self_attn.in_proj_qkvgfab", "language_model.model.layers.0.self_attn.k_proj", "language_model.model.layers.0.self_attn.o_proj", "language_model.model.layers.0.self_attn.q_proj", "language_model.model.layers.0.self_attn.v_proj", "language_model.model.layers.1.self_attn.b_proj", "language_model.model.layers.1.self_attn.f_a_proj", "language_model.model.layers.1.self_attn.f_b_proj", "language_model.model.layers.1.self_attn.fused_qkvg_proj", "language_model.model.layers.1.self_attn.g_proj", "language_model.model.layers.1.self_attn.in_proj_qkvgfab", "language_model.model.layers.1.self_attn.k_proj", "language_model.model.layers.1.self_attn.o_proj", "language_model.model.layers.1.self_attn.q_proj", "language_model.model.layers.1.self_attn.v_proj", "language_model.model.layers.10.self_attn.b_proj", "language_model.model.layers.10.self_attn.f_a_proj", "language_model.model.layers.10.self_attn.f_b_proj", "language_model.model.layers.10.self_attn.fused_qkvg_proj", "language_model.model.layers.10.self_attn.g_proj", "language_model.model.layers.10.self_attn.in_proj_qkvgfab", "language_model.model.layers.10.self_attn.k_proj", "language_model.model.layers.10.self_attn.o_proj", "language_model.model.layers.10.self_attn.q_proj", "language_model.model.layers.10.self_attn.v_proj", "language_model.model.layers.11.self_attn.fused_qkv_a_proj", "language_model.model.layers.11.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.11.self_attn.fused_qkvg_proj", "language_model.model.layers.11.self_attn.g_proj", "language_model.model.layers.11.self_attn.in_proj_qkvgfab", "language_model.model.layers.11.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.11.self_attn.kv_b_proj", "language_model.model.layers.11.self_attn.o_proj", "language_model.model.layers.11.self_attn.q_a_proj", "language_model.model.layers.11.self_attn.q_b_proj", "language_model.model.layers.12.self_attn.b_proj", "language_model.model.layers.12.self_attn.f_a_proj", "language_model.model.layers.12.self_attn.f_b_proj", "language_model.model.layers.12.self_attn.fused_qkvg_proj", "language_model.model.layers.12.self_attn.g_proj", "language_model.model.layers.12.self_attn.in_proj_qkvgfab", "language_model.model.layers.12.self_attn.k_proj", "language_model.model.layers.12.self_attn.o_proj", "language_model.model.layers.12.self_attn.q_proj", "language_model.model.layers.12.self_attn.v_proj", "language_model.model.layers.13.self_attn.b_proj", "language_model.model.layers.13.self_attn.f_a_proj", "language_model.model.layers.13.self_attn.f_b_proj", "language_model.model.layers.13.self_attn.fused_qkvg_proj", "language_model.model.layers.13.self_attn.g_proj", "language_model.model.layers.13.self_attn.in_proj_qkvgfab", "language_model.model.layers.13.self_attn.k_proj", "language_model.model.layers.13.self_attn.o_proj", "language_model.model.layers.13.self_attn.q_proj", "language_model.model.layers.13.self_attn.v_proj", "language_model.model.layers.14.self_attn.b_proj", "language_model.model.layers.14.self_attn.f_a_proj", "language_model.model.layers.14.self_attn.f_b_proj", "language_model.model.layers.14.self_attn.fused_qkvg_proj", "language_model.model.layers.14.self_attn.g_proj", "language_model.model.layers.14.self_attn.in_proj_qkvgfab", "language_model.model.layers.14.self_attn.k_proj", "language_model.model.layers.14.self_attn.o_proj", "language_model.model.layers.14.self_attn.q_proj", "language_model.model.layers.14.self_attn.v_proj", "language_model.model.layers.15.self_attn.fused_qkv_a_proj", "language_model.model.layers.15.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.15.self_attn.fused_qkvg_proj", "language_model.model.layers.15.self_attn.g_proj", "language_model.model.layers.15.self_attn.in_proj_qkvgfab", "language_model.model.layers.15.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.15.self_attn.kv_b_proj", "language_model.model.layers.15.self_attn.o_proj", "language_model.model.layers.15.self_attn.q_a_proj", "language_model.model.layers.15.self_attn.q_b_proj", "language_model.model.layers.16.self_attn.b_proj", "language_model.model.layers.16.self_attn.f_a_proj", "language_model.model.layers.16.self_attn.f_b_proj", "language_model.model.layers.16.self_attn.fused_qkvg_proj", "language_model.model.layers.16.self_attn.g_proj", "language_model.model.layers.16.self_attn.in_proj_qkvgfab", "language_model.model.layers.16.self_attn.k_proj", "language_model.model.layers.16.self_attn.o_proj", "language_model.model.layers.16.self_attn.q_proj", "language_model.model.layers.16.self_attn.v_proj", "language_model.model.layers.17.self_attn.b_proj", "language_model.model.layers.17.self_attn.f_a_proj", "language_model.model.layers.17.self_attn.f_b_proj", "language_model.model.layers.17.self_attn.fused_qkvg_proj", "language_model.model.layers.17.self_attn.g_proj", "language_model.model.layers.17.self_attn.in_proj_qkvgfab", "language_model.model.layers.17.self_attn.k_proj", "language_model.model.layers.17.self_attn.o_proj", "language_model.model.layers.17.self_attn.q_proj", "language_model.model.layers.17.self_attn.v_proj", "language_model.model.layers.18.self_attn.b_proj", "language_model.model.layers.18.self_attn.f_a_proj", "language_model.model.layers.18.self_attn.f_b_proj", "language_model.model.layers.18.self_attn.fused_qkvg_proj", "language_model.model.layers.18.self_attn.g_proj", "language_model.model.layers.18.self_attn.in_proj_qkvgfab", "language_model.model.layers.18.self_attn.k_proj", "language_model.model.layers.18.self_attn.o_proj", "language_model.model.layers.18.self_attn.q_proj", "language_model.model.layers.18.self_attn.v_proj", "language_model.model.layers.19.self_attn.fused_qkv_a_proj", "language_model.model.layers.19.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.19.self_attn.fused_qkvg_proj", "language_model.model.layers.19.self_attn.g_proj", "language_model.model.layers.19.self_attn.in_proj_qkvgfab", "language_model.model.layers.19.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.19.self_attn.kv_b_proj", "language_model.model.layers.19.self_attn.o_proj", "language_model.model.layers.19.self_attn.q_a_proj", "language_model.model.layers.19.self_attn.q_b_proj", "language_model.model.layers.2.self_attn.b_proj", "language_model.model.layers.2.self_attn.f_a_proj", "language_model.model.layers.2.self_attn.f_b_proj", "language_model.model.layers.2.self_attn.fused_qkvg_proj", "language_model.model.layers.2.self_attn.g_proj", "language_model.model.layers.2.self_attn.in_proj_qkvgfab", "language_model.model.layers.2.self_attn.k_proj", "language_model.model.layers.2.self_attn.o_proj", "language_model.model.layers.2.self_attn.q_proj", "language_model.model.layers.2.self_attn.v_proj", "language_model.model.layers.20.self_attn.b_proj", "language_model.model.layers.20.self_attn.f_a_proj", "language_model.model.layers.20.self_attn.f_b_proj", "language_model.model.layers.20.self_attn.fused_qkvg_proj", "language_model.model.layers.20.self_attn.g_proj", "language_model.model.layers.20.self_attn.in_proj_qkvgfab", "language_model.model.layers.20.self_attn.k_proj", "language_model.model.layers.20.self_attn.o_proj", "language_model.model.layers.20.self_attn.q_proj", "language_model.model.layers.20.self_attn.v_proj", "language_model.model.layers.21.self_attn.b_proj", "language_model.model.layers.21.self_attn.f_a_proj", "language_model.model.layers.21.self_attn.f_b_proj", "language_model.model.layers.21.self_attn.fused_qkvg_proj", "language_model.model.layers.21.self_attn.g_proj", "language_model.model.layers.21.self_attn.in_proj_qkvgfab", "language_model.model.layers.21.self_attn.k_proj", "language_model.model.layers.21.self_attn.o_proj", "language_model.model.layers.21.self_attn.q_proj", "language_model.model.layers.21.self_attn.v_proj", "language_model.model.layers.22.self_attn.b_proj", "language_model.model.layers.22.self_attn.f_a_proj", "language_model.model.layers.22.self_attn.f_b_proj", "language_model.model.layers.22.self_attn.fused_qkvg_proj", "language_model.model.layers.22.self_attn.g_proj", "language_model.model.layers.22.self_attn.in_proj_qkvgfab", "language_model.model.layers.22.self_attn.k_proj", "language_model.model.layers.22.self_attn.o_proj", "language_model.model.layers.22.self_attn.q_proj", "language_model.model.layers.22.self_attn.v_proj", "language_model.model.layers.23.self_attn.fused_qkv_a_proj", "language_model.model.layers.23.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.23.self_attn.fused_qkvg_proj", "language_model.model.layers.23.self_attn.g_proj", "language_model.model.layers.23.self_attn.in_proj_qkvgfab", "language_model.model.layers.23.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.23.self_attn.kv_b_proj", "language_model.model.layers.23.self_attn.o_proj", "language_model.model.layers.23.self_attn.q_a_proj", "language_model.model.layers.23.self_attn.q_b_proj", "language_model.model.layers.24.self_attn.b_proj", "language_model.model.layers.24.self_attn.f_a_proj", "language_model.model.layers.24.self_attn.f_b_proj", "language_model.model.layers.24.self_attn.fused_qkvg_proj", "language_model.model.layers.24.self_attn.g_proj", "language_model.model.layers.24.self_attn.in_proj_qkvgfab", "language_model.model.layers.24.self_attn.k_proj", "language_model.model.layers.24.self_attn.o_proj", "language_model.model.layers.24.self_attn.q_proj", "language_model.model.layers.24.self_attn.v_proj", "language_model.model.layers.25.self_attn.b_proj", "language_model.model.layers.25.self_attn.f_a_proj", "language_model.model.layers.25.self_attn.f_b_proj", "language_model.model.layers.25.self_attn.fused_qkvg_proj", "language_model.model.layers.25.self_attn.g_proj", "language_model.model.layers.25.self_attn.in_proj_qkvgfab", "language_model.model.layers.25.self_attn.k_proj", "language_model.model.layers.25.self_attn.o_proj", "language_model.model.layers.25.self_attn.q_proj", "language_model.model.layers.25.self_attn.v_proj", "language_model.model.layers.26.self_attn.b_proj", "language_model.model.layers.26.self_attn.f_a_proj", "language_model.model.layers.26.self_attn.f_b_proj", "language_model.model.layers.26.self_attn.fused_qkvg_proj", "language_model.model.layers.26.self_attn.g_proj", "language_model.model.layers.26.self_attn.in_proj_qkvgfab", "language_model.model.layers.26.self_attn.k_proj", "language_model.model.layers.26.self_attn.o_proj", "language_model.model.layers.26.self_attn.q_proj", "language_model.model.layers.26.self_attn.v_proj", "language_model.model.layers.27.self_attn.fused_qkv_a_proj", "language_model.model.layers.27.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.27.self_attn.fused_qkvg_proj", "language_model.model.layers.27.self_attn.g_proj", "language_model.model.layers.27.self_attn.in_proj_qkvgfab", "language_model.model.layers.27.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.27.self_attn.kv_b_proj", "language_model.model.layers.27.self_attn.o_proj", "language_model.model.layers.27.self_attn.q_a_proj", "language_model.model.layers.27.self_attn.q_b_proj", "language_model.model.layers.28.self_attn.b_proj", "language_model.model.layers.28.self_attn.f_a_proj", "language_model.model.layers.28.self_attn.f_b_proj", "language_model.model.layers.28.self_attn.fused_qkvg_proj", "language_model.model.layers.28.self_attn.g_proj", "language_model.model.layers.28.self_attn.in_proj_qkvgfab", "language_model.model.layers.28.self_attn.k_proj", "language_model.model.layers.28.self_attn.o_proj", "language_model.model.layers.28.self_attn.q_proj", "language_model.model.layers.28.self_attn.v_proj", "language_model.model.layers.29.self_attn.b_proj", "language_model.model.layers.29.self_attn.f_a_proj", "language_model.model.layers.29.self_attn.f_b_proj", "language_model.model.layers.29.self_attn.fused_qkvg_proj", "language_model.model.layers.29.self_attn.g_proj", "language_model.model.layers.29.self_attn.in_proj_qkvgfab", "language_model.model.layers.29.self_attn.k_proj", "language_model.model.layers.29.self_attn.o_proj", "language_model.model.layers.29.self_attn.q_proj", "language_model.model.layers.29.self_attn.v_proj", "language_model.model.layers.3.self_attn.fused_qkv_a_proj", "language_model.model.layers.3.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.3.self_attn.fused_qkvg_proj", "language_model.model.layers.3.self_attn.g_proj", "language_model.model.layers.3.self_attn.in_proj_qkvgfab", "language_model.model.layers.3.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.3.self_attn.kv_b_proj", "language_model.model.layers.3.self_attn.o_proj", "language_model.model.layers.3.self_attn.q_a_proj", "language_model.model.layers.3.self_attn.q_b_proj", "language_model.model.layers.30.self_attn.b_proj", "language_model.model.layers.30.self_attn.f_a_proj", "language_model.model.layers.30.self_attn.f_b_proj", "language_model.model.layers.30.self_attn.fused_qkvg_proj", "language_model.model.layers.30.self_attn.g_proj", "language_model.model.layers.30.self_attn.in_proj_qkvgfab", "language_model.model.layers.30.self_attn.k_proj", "language_model.model.layers.30.self_attn.o_proj", "language_model.model.layers.30.self_attn.q_proj", "language_model.model.layers.30.self_attn.v_proj", "language_model.model.layers.31.self_attn.fused_qkv_a_proj", "language_model.model.layers.31.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.31.self_attn.fused_qkvg_proj", "language_model.model.layers.31.self_attn.g_proj", "language_model.model.layers.31.self_attn.in_proj_qkvgfab", "language_model.model.layers.31.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.31.self_attn.kv_b_proj", "language_model.model.layers.31.self_attn.o_proj", "language_model.model.layers.31.self_attn.q_a_proj", "language_model.model.layers.31.self_attn.q_b_proj", "language_model.model.layers.32.self_attn.b_proj", "language_model.model.layers.32.self_attn.f_a_proj", "language_model.model.layers.32.self_attn.f_b_proj", "language_model.model.layers.32.self_attn.fused_qkvg_proj", "language_model.model.layers.32.self_attn.g_proj", "language_model.model.layers.32.self_attn.in_proj_qkvgfab", "language_model.model.layers.32.self_attn.k_proj", "language_model.model.layers.32.self_attn.o_proj", "language_model.model.layers.32.self_attn.q_proj", "language_model.model.layers.32.self_attn.v_proj", "language_model.model.layers.33.self_attn.b_proj", "language_model.model.layers.33.self_attn.f_a_proj", "language_model.model.layers.33.self_attn.f_b_proj", "language_model.model.layers.33.self_attn.fused_qkvg_proj", "language_model.model.layers.33.self_attn.g_proj", "language_model.model.layers.33.self_attn.in_proj_qkvgfab", "language_model.model.layers.33.self_attn.k_proj", "language_model.model.layers.33.self_attn.o_proj", "language_model.model.layers.33.self_attn.q_proj", "language_model.model.layers.33.self_attn.v_proj", "language_model.model.layers.34.self_attn.b_proj", "language_model.model.layers.34.self_attn.f_a_proj", "language_model.model.layers.34.self_attn.f_b_proj", "language_model.model.layers.34.self_attn.fused_qkvg_proj", "language_model.model.layers.34.self_attn.g_proj", "language_model.model.layers.34.self_attn.in_proj_qkvgfab", "language_model.model.layers.34.self_attn.k_proj", "language_model.model.layers.34.self_attn.o_proj", "language_model.model.layers.34.self_attn.q_proj", "language_model.model.layers.34.self_attn.v_proj", "language_model.model.layers.35.self_attn.fused_qkv_a_proj", "language_model.model.layers.35.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.35.self_attn.fused_qkvg_proj", "language_model.model.layers.35.self_attn.g_proj", "language_model.model.layers.35.self_attn.in_proj_qkvgfab", "language_model.model.layers.35.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.35.self_attn.kv_b_proj", "language_model.model.layers.35.self_attn.o_proj", "language_model.model.layers.35.self_attn.q_a_proj", "language_model.model.layers.35.self_attn.q_b_proj", "language_model.model.layers.36.self_attn.b_proj", "language_model.model.layers.36.self_attn.f_a_proj", "language_model.model.layers.36.self_attn.f_b_proj", "language_model.model.layers.36.self_attn.fused_qkvg_proj", "language_model.model.layers.36.self_attn.g_proj", "language_model.model.layers.36.self_attn.in_proj_qkvgfab", "language_model.model.layers.36.self_attn.k_proj", "language_model.model.layers.36.self_attn.o_proj", "language_model.model.layers.36.self_attn.q_proj", "language_model.model.layers.36.self_attn.v_proj", "language_model.model.layers.37.self_attn.b_proj", "language_model.model.layers.37.self_attn.f_a_proj", "language_model.model.layers.37.self_attn.f_b_proj", "language_model.model.layers.37.self_attn.fused_qkvg_proj", "language_model.model.layers.37.self_attn.g_proj", "language_model.model.layers.37.self_attn.in_proj_qkvgfab", "language_model.model.layers.37.self_attn.k_proj", "language_model.model.layers.37.self_attn.o_proj", "language_model.model.layers.37.self_attn.q_proj", "language_model.model.layers.37.self_attn.v_proj", "language_model.model.layers.38.self_attn.b_proj", "language_model.model.layers.38.self_attn.f_a_proj", "language_model.model.layers.38.self_attn.f_b_proj", "language_model.model.layers.38.self_attn.fused_qkvg_proj", "language_model.model.layers.38.self_attn.g_proj", "language_model.model.layers.38.self_attn.in_proj_qkvgfab", "language_model.model.layers.38.self_attn.k_proj", "language_model.model.layers.38.self_attn.o_proj", "language_model.model.layers.38.self_attn.q_proj", "language_model.model.layers.38.self_attn.v_proj", "language_model.model.layers.39.self_attn.fused_qkv_a_proj", "language_model.model.layers.39.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.39.self_attn.fused_qkvg_proj", "language_model.model.layers.39.self_attn.g_proj", "language_model.model.layers.39.self_attn.in_proj_qkvgfab", "language_model.model.layers.39.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.39.self_attn.kv_b_proj", "language_model.model.layers.39.self_attn.o_proj", "language_model.model.layers.39.self_attn.q_a_proj", "language_model.model.layers.39.self_attn.q_b_proj", "language_model.model.layers.4.self_attn.b_proj", "language_model.model.layers.4.self_attn.f_a_proj", "language_model.model.layers.4.self_attn.f_b_proj", "language_model.model.layers.4.self_attn.fused_qkvg_proj", "language_model.model.layers.4.self_attn.g_proj", "language_model.model.layers.4.self_attn.in_proj_qkvgfab", "language_model.model.layers.4.self_attn.k_proj", "language_model.model.layers.4.self_attn.o_proj", "language_model.model.layers.4.self_attn.q_proj", "language_model.model.layers.4.self_attn.v_proj", "language_model.model.layers.40.self_attn.b_proj", "language_model.model.layers.40.self_attn.f_a_proj", "language_model.model.layers.40.self_attn.f_b_proj", "language_model.model.layers.40.self_attn.fused_qkvg_proj", "language_model.model.layers.40.self_attn.g_proj", "language_model.model.layers.40.self_attn.in_proj_qkvgfab", "language_model.model.layers.40.self_attn.k_proj", "language_model.model.layers.40.self_attn.o_proj", "language_model.model.layers.40.self_attn.q_proj", "language_model.model.layers.40.self_attn.v_proj", "language_model.model.layers.41.self_attn.b_proj", "language_model.model.layers.41.self_attn.f_a_proj", "language_model.model.layers.41.self_attn.f_b_proj", "language_model.model.layers.41.self_attn.fused_qkvg_proj", "language_model.model.layers.41.self_attn.g_proj", "language_model.model.layers.41.self_attn.in_proj_qkvgfab", "language_model.model.layers.41.self_attn.k_proj", "language_model.model.layers.41.self_attn.o_proj", "language_model.model.layers.41.self_attn.q_proj", "language_model.model.layers.41.self_attn.v_proj", "language_model.model.layers.42.self_attn.b_proj", "language_model.model.layers.42.self_attn.f_a_proj", "language_model.model.layers.42.self_attn.f_b_proj", "language_model.model.layers.42.self_attn.fused_qkvg_proj", "language_model.model.layers.42.self_attn.g_proj", "language_model.model.layers.42.self_attn.in_proj_qkvgfab", "language_model.model.layers.42.self_attn.k_proj", "language_model.model.layers.42.self_attn.o_proj", "language_model.model.layers.42.self_attn.q_proj", "language_model.model.layers.42.self_attn.v_proj", "language_model.model.layers.43.self_attn.fused_qkv_a_proj", "language_model.model.layers.43.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.43.self_attn.fused_qkvg_proj", "language_model.model.layers.43.self_attn.g_proj", "language_model.model.layers.43.self_attn.in_proj_qkvgfab", "language_model.model.layers.43.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.43.self_attn.kv_b_proj", "language_model.model.layers.43.self_attn.o_proj", "language_model.model.layers.43.self_attn.q_a_proj", "language_model.model.layers.43.self_attn.q_b_proj", "language_model.model.layers.44.self_attn.b_proj", "language_model.model.layers.44.self_attn.f_a_proj", "language_model.model.layers.44.self_attn.f_b_proj", "language_model.model.layers.44.self_attn.fused_qkvg_proj", "language_model.model.layers.44.self_attn.g_proj", "language_model.model.layers.44.self_attn.in_proj_qkvgfab", "language_model.model.layers.44.self_attn.k_proj", "language_model.model.layers.44.self_attn.o_proj", "language_model.model.layers.44.self_attn.q_proj", "language_model.model.layers.44.self_attn.v_proj", "language_model.model.layers.45.self_attn.b_proj", "language_model.model.layers.45.self_attn.f_a_proj", "language_model.model.layers.45.self_attn.f_b_proj", "language_model.model.layers.45.self_attn.fused_qkvg_proj", "language_model.model.layers.45.self_attn.g_proj", "language_model.model.layers.45.self_attn.in_proj_qkvgfab", "language_model.model.layers.45.self_attn.k_proj", "language_model.model.layers.45.self_attn.o_proj", "language_model.model.layers.45.self_attn.q_proj", "language_model.model.layers.45.self_attn.v_proj", "language_model.model.layers.46.self_attn.b_proj", "language_model.model.layers.46.self_attn.f_a_proj", "language_model.model.layers.46.self_attn.f_b_proj", "language_model.model.layers.46.self_attn.fused_qkvg_proj", "language_model.model.layers.46.self_attn.g_proj", "language_model.model.layers.46.self_attn.in_proj_qkvgfab", "language_model.model.layers.46.self_attn.k_proj", "language_model.model.layers.46.self_attn.o_proj", "language_model.model.layers.46.self_attn.q_proj", "language_model.model.layers.46.self_attn.v_proj", "language_model.model.layers.47.self_attn.fused_qkv_a_proj", "language_model.model.layers.47.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.47.self_attn.fused_qkvg_proj", "language_model.model.layers.47.self_attn.g_proj", "language_model.model.layers.47.self_attn.in_proj_qkvgfab", "language_model.model.layers.47.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.47.self_attn.kv_b_proj", "language_model.model.layers.47.self_attn.o_proj", "language_model.model.layers.47.self_attn.q_a_proj", "language_model.model.layers.47.self_attn.q_b_proj", "language_model.model.layers.48.self_attn.b_proj", "language_model.model.layers.48.self_attn.f_a_proj", "language_model.model.layers.48.self_attn.f_b_proj", "language_model.model.layers.48.self_attn.fused_qkvg_proj", "language_model.model.layers.48.self_attn.g_proj", "language_model.model.layers.48.self_attn.in_proj_qkvgfab", "language_model.model.layers.48.self_attn.k_proj", "language_model.model.layers.48.self_attn.o_proj", "language_model.model.layers.48.self_attn.q_proj", "language_model.model.layers.48.self_attn.v_proj", "language_model.model.layers.49.self_attn.b_proj", "language_model.model.layers.49.self_attn.f_a_proj", "language_model.model.layers.49.self_attn.f_b_proj", "language_model.model.layers.49.self_attn.fused_qkvg_proj", "language_model.model.layers.49.self_attn.g_proj", "language_model.model.layers.49.self_attn.in_proj_qkvgfab", "language_model.model.layers.49.self_attn.k_proj", "language_model.model.layers.49.self_attn.o_proj", "language_model.model.layers.49.self_attn.q_proj", "language_model.model.layers.49.self_attn.v_proj", "language_model.model.layers.5.self_attn.b_proj", "language_model.model.layers.5.self_attn.f_a_proj", "language_model.model.layers.5.self_attn.f_b_proj", "language_model.model.layers.5.self_attn.fused_qkvg_proj", "language_model.model.layers.5.self_attn.g_proj", "language_model.model.layers.5.self_attn.in_proj_qkvgfab", "language_model.model.layers.5.self_attn.k_proj", "language_model.model.layers.5.self_attn.o_proj", "language_model.model.layers.5.self_attn.q_proj", "language_model.model.layers.5.self_attn.v_proj", "language_model.model.layers.50.self_attn.b_proj", "language_model.model.layers.50.self_attn.f_a_proj", "language_model.model.layers.50.self_attn.f_b_proj", "language_model.model.layers.50.self_attn.fused_qkvg_proj", "language_model.model.layers.50.self_attn.g_proj", "language_model.model.layers.50.self_attn.in_proj_qkvgfab", "language_model.model.layers.50.self_attn.k_proj", "language_model.model.layers.50.self_attn.o_proj", "language_model.model.layers.50.self_attn.q_proj", "language_model.model.layers.50.self_attn.v_proj", "language_model.model.layers.51.self_attn.fused_qkv_a_proj", "language_model.model.layers.51.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.51.self_attn.fused_qkvg_proj", "language_model.model.layers.51.self_attn.g_proj", "language_model.model.layers.51.self_attn.in_proj_qkvgfab", "language_model.model.layers.51.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.51.self_attn.kv_b_proj", "language_model.model.layers.51.self_attn.o_proj", "language_model.model.layers.51.self_attn.q_a_proj", "language_model.model.layers.51.self_attn.q_b_proj", "language_model.model.layers.52.self_attn.b_proj", "language_model.model.layers.52.self_attn.f_a_proj", "language_model.model.layers.52.self_attn.f_b_proj", "language_model.model.layers.52.self_attn.fused_qkvg_proj", "language_model.model.layers.52.self_attn.g_proj", "language_model.model.layers.52.self_attn.in_proj_qkvgfab", "language_model.model.layers.52.self_attn.k_proj", "language_model.model.layers.52.self_attn.o_proj", "language_model.model.layers.52.self_attn.q_proj", "language_model.model.layers.52.self_attn.v_proj", "language_model.model.layers.53.self_attn.b_proj", "language_model.model.layers.53.self_attn.f_a_proj", "language_model.model.layers.53.self_attn.f_b_proj", "language_model.model.layers.53.self_attn.fused_qkvg_proj", "language_model.model.layers.53.self_attn.g_proj", "language_model.model.layers.53.self_attn.in_proj_qkvgfab", "language_model.model.layers.53.self_attn.k_proj", "language_model.model.layers.53.self_attn.o_proj", "language_model.model.layers.53.self_attn.q_proj", "language_model.model.layers.53.self_attn.v_proj", "language_model.model.layers.54.self_attn.b_proj", "language_model.model.layers.54.self_attn.f_a_proj", "language_model.model.layers.54.self_attn.f_b_proj", "language_model.model.layers.54.self_attn.fused_qkvg_proj", "language_model.model.layers.54.self_attn.g_proj", "language_model.model.layers.54.self_attn.in_proj_qkvgfab", "language_model.model.layers.54.self_attn.k_proj", "language_model.model.layers.54.self_attn.o_proj", "language_model.model.layers.54.self_attn.q_proj", "language_model.model.layers.54.self_attn.v_proj", "language_model.model.layers.55.self_attn.fused_qkv_a_proj", "language_model.model.layers.55.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.55.self_attn.fused_qkvg_proj", "language_model.model.layers.55.self_attn.g_proj", "language_model.model.layers.55.self_attn.in_proj_qkvgfab", "language_model.model.layers.55.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.55.self_attn.kv_b_proj", "language_model.model.layers.55.self_attn.o_proj", "language_model.model.layers.55.self_attn.q_a_proj", "language_model.model.layers.55.self_attn.q_b_proj", "language_model.model.layers.56.self_attn.b_proj", "language_model.model.layers.56.self_attn.f_a_proj", "language_model.model.layers.56.self_attn.f_b_proj", "language_model.model.layers.56.self_attn.fused_qkvg_proj", "language_model.model.layers.56.self_attn.g_proj", "language_model.model.layers.56.self_attn.in_proj_qkvgfab", "language_model.model.layers.56.self_attn.k_proj", "language_model.model.layers.56.self_attn.o_proj", "language_model.model.layers.56.self_attn.q_proj", "language_model.model.layers.56.self_attn.v_proj", "language_model.model.layers.57.self_attn.b_proj", "language_model.model.layers.57.self_attn.f_a_proj", "language_model.model.layers.57.self_attn.f_b_proj", "language_model.model.layers.57.self_attn.fused_qkvg_proj", "language_model.model.layers.57.self_attn.g_proj", "language_model.model.layers.57.self_attn.in_proj_qkvgfab", "language_model.model.layers.57.self_attn.k_proj", "language_model.model.layers.57.self_attn.o_proj", "language_model.model.layers.57.self_attn.q_proj", "language_model.model.layers.57.self_attn.v_proj", "language_model.model.layers.58.self_attn.b_proj", "language_model.model.layers.58.self_attn.f_a_proj", "language_model.model.layers.58.self_attn.f_b_proj", "language_model.model.layers.58.self_attn.fused_qkvg_proj", "language_model.model.layers.58.self_attn.g_proj", "language_model.model.layers.58.self_attn.in_proj_qkvgfab", "language_model.model.layers.58.self_attn.k_proj", "language_model.model.layers.58.self_attn.o_proj", "language_model.model.layers.58.self_attn.q_proj", "language_model.model.layers.58.self_attn.v_proj", "language_model.model.layers.59.self_attn.fused_qkv_a_proj", "language_model.model.layers.59.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.59.self_attn.fused_qkvg_proj", "language_model.model.layers.59.self_attn.g_proj", "language_model.model.layers.59.self_attn.in_proj_qkvgfab", "language_model.model.layers.59.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.59.self_attn.kv_b_proj", "language_model.model.layers.59.self_attn.o_proj", "language_model.model.layers.59.self_attn.q_a_proj", "language_model.model.layers.59.self_attn.q_b_proj", "language_model.model.layers.6.self_attn.b_proj", "language_model.model.layers.6.self_attn.f_a_proj", "language_model.model.layers.6.self_attn.f_b_proj", "language_model.model.layers.6.self_attn.fused_qkvg_proj", "language_model.model.layers.6.self_attn.g_proj", "language_model.model.layers.6.self_attn.in_proj_qkvgfab", "language_model.model.layers.6.self_attn.k_proj", "language_model.model.layers.6.self_attn.o_proj", "language_model.model.layers.6.self_attn.q_proj", "language_model.model.layers.6.self_attn.v_proj", "language_model.model.layers.60.self_attn.b_proj", "language_model.model.layers.60.self_attn.f_a_proj", "language_model.model.layers.60.self_attn.f_b_proj", "language_model.model.layers.60.self_attn.fused_qkvg_proj", "language_model.model.layers.60.self_attn.g_proj", "language_model.model.layers.60.self_attn.in_proj_qkvgfab", "language_model.model.layers.60.self_attn.k_proj", "language_model.model.layers.60.self_attn.o_proj", "language_model.model.layers.60.self_attn.q_proj", "language_model.model.layers.60.self_attn.v_proj", "language_model.model.layers.61.self_attn.b_proj", "language_model.model.layers.61.self_attn.f_a_proj", "language_model.model.layers.61.self_attn.f_b_proj", "language_model.model.layers.61.self_attn.fused_qkvg_proj", "language_model.model.layers.61.self_attn.g_proj", "language_model.model.layers.61.self_attn.in_proj_qkvgfab", "language_model.model.layers.61.self_attn.k_proj", "language_model.model.layers.61.self_attn.o_proj", "language_model.model.layers.61.self_attn.q_proj", "language_model.model.layers.61.self_attn.v_proj", "language_model.model.layers.62.self_attn.b_proj", "language_model.model.layers.62.self_attn.f_a_proj", "language_model.model.layers.62.self_attn.f_b_proj", "language_model.model.layers.62.self_attn.fused_qkvg_proj", "language_model.model.layers.62.self_attn.g_proj", "language_model.model.layers.62.self_attn.in_proj_qkvgfab", "language_model.model.layers.62.self_attn.k_proj", "language_model.model.layers.62.self_attn.o_proj", "language_model.model.layers.62.self_attn.q_proj", "language_model.model.layers.62.self_attn.v_proj", "language_model.model.layers.63.self_attn.fused_qkv_a_proj", "language_model.model.layers.63.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.63.self_attn.fused_qkvg_proj", "language_model.model.layers.63.self_attn.g_proj", "language_model.model.layers.63.self_attn.in_proj_qkvgfab", "language_model.model.layers.63.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.63.self_attn.kv_b_proj", "language_model.model.layers.63.self_attn.o_proj", "language_model.model.layers.63.self_attn.q_a_proj", "language_model.model.layers.63.self_attn.q_b_proj", "language_model.model.layers.64.self_attn.b_proj", "language_model.model.layers.64.self_attn.f_a_proj", "language_model.model.layers.64.self_attn.f_b_proj", "language_model.model.layers.64.self_attn.fused_qkvg_proj", "language_model.model.layers.64.self_attn.g_proj", "language_model.model.layers.64.self_attn.in_proj_qkvgfab", "language_model.model.layers.64.self_attn.k_proj", "language_model.model.layers.64.self_attn.o_proj", "language_model.model.layers.64.self_attn.q_proj", "language_model.model.layers.64.self_attn.v_proj", "language_model.model.layers.65.self_attn.b_proj", "language_model.model.layers.65.self_attn.f_a_proj", "language_model.model.layers.65.self_attn.f_b_proj", "language_model.model.layers.65.self_attn.fused_qkvg_proj", "language_model.model.layers.65.self_attn.g_proj", "language_model.model.layers.65.self_attn.in_proj_qkvgfab", "language_model.model.layers.65.self_attn.k_proj", "language_model.model.layers.65.self_attn.o_proj", "language_model.model.layers.65.self_attn.q_proj", "language_model.model.layers.65.self_attn.v_proj", "language_model.model.layers.66.self_attn.b_proj", "language_model.model.layers.66.self_attn.f_a_proj", "language_model.model.layers.66.self_attn.f_b_proj", "language_model.model.layers.66.self_attn.fused_qkvg_proj", "language_model.model.layers.66.self_attn.g_proj", "language_model.model.layers.66.self_attn.in_proj_qkvgfab", "language_model.model.layers.66.self_attn.k_proj", "language_model.model.layers.66.self_attn.o_proj", "language_model.model.layers.66.self_attn.q_proj", "language_model.model.layers.66.self_attn.v_proj", "language_model.model.layers.67.self_attn.fused_qkv_a_proj", "language_model.model.layers.67.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.67.self_attn.fused_qkvg_proj", "language_model.model.layers.67.self_attn.g_proj", "language_model.model.layers.67.self_attn.in_proj_qkvgfab", "language_model.model.layers.67.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.67.self_attn.kv_b_proj", "language_model.model.layers.67.self_attn.o_proj", "language_model.model.layers.67.self_attn.q_a_proj", "language_model.model.layers.67.self_attn.q_b_proj", "language_model.model.layers.68.self_attn.b_proj", "language_model.model.layers.68.self_attn.f_a_proj", "language_model.model.layers.68.self_attn.f_b_proj", "language_model.model.layers.68.self_attn.fused_qkvg_proj", "language_model.model.layers.68.self_attn.g_proj", "language_model.model.layers.68.self_attn.in_proj_qkvgfab", "language_model.model.layers.68.self_attn.k_proj", "language_model.model.layers.68.self_attn.o_proj", "language_model.model.layers.68.self_attn.q_proj", "language_model.model.layers.68.self_attn.v_proj", "language_model.model.layers.69.self_attn.b_proj", "language_model.model.layers.69.self_attn.f_a_proj", "language_model.model.layers.69.self_attn.f_b_proj", "language_model.model.layers.69.self_attn.fused_qkvg_proj", "language_model.model.layers.69.self_attn.g_proj", "language_model.model.layers.69.self_attn.in_proj_qkvgfab", "language_model.model.layers.69.self_attn.k_proj", "language_model.model.layers.69.self_attn.o_proj", "language_model.model.layers.69.self_attn.q_proj", "language_model.model.layers.69.self_attn.v_proj", "language_model.model.layers.7.self_attn.fused_qkv_a_proj", "language_model.model.layers.7.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.7.self_attn.fused_qkvg_proj", "language_model.model.layers.7.self_attn.g_proj", "language_model.model.layers.7.self_attn.in_proj_qkvgfab", "language_model.model.layers.7.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.7.self_attn.kv_b_proj", "language_model.model.layers.7.self_attn.o_proj", "language_model.model.layers.7.self_attn.q_a_proj", "language_model.model.layers.7.self_attn.q_b_proj", "language_model.model.layers.70.self_attn.b_proj", "language_model.model.layers.70.self_attn.f_a_proj", "language_model.model.layers.70.self_attn.f_b_proj", "language_model.model.layers.70.self_attn.fused_qkvg_proj", "language_model.model.layers.70.self_attn.g_proj", "language_model.model.layers.70.self_attn.in_proj_qkvgfab", "language_model.model.layers.70.self_attn.k_proj", "language_model.model.layers.70.self_attn.o_proj", "language_model.model.layers.70.self_attn.q_proj", "language_model.model.layers.70.self_attn.v_proj", "language_model.model.layers.71.self_attn.fused_qkv_a_proj", "language_model.model.layers.71.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.71.self_attn.fused_qkvg_proj", "language_model.model.layers.71.self_attn.g_proj", "language_model.model.layers.71.self_attn.in_proj_qkvgfab", "language_model.model.layers.71.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.71.self_attn.kv_b_proj", "language_model.model.layers.71.self_attn.o_proj", "language_model.model.layers.71.self_attn.q_a_proj", "language_model.model.layers.71.self_attn.q_b_proj", "language_model.model.layers.72.self_attn.b_proj", "language_model.model.layers.72.self_attn.f_a_proj", "language_model.model.layers.72.self_attn.f_b_proj", "language_model.model.layers.72.self_attn.fused_qkvg_proj", "language_model.model.layers.72.self_attn.g_proj", "language_model.model.layers.72.self_attn.in_proj_qkvgfab", "language_model.model.layers.72.self_attn.k_proj", "language_model.model.layers.72.self_attn.o_proj", "language_model.model.layers.72.self_attn.q_proj", "language_model.model.layers.72.self_attn.v_proj", "language_model.model.layers.73.self_attn.b_proj", "language_model.model.layers.73.self_attn.f_a_proj", "language_model.model.layers.73.self_attn.f_b_proj", "language_model.model.layers.73.self_attn.fused_qkvg_proj", "language_model.model.layers.73.self_attn.g_proj", "language_model.model.layers.73.self_attn.in_proj_qkvgfab", "language_model.model.layers.73.self_attn.k_proj", "language_model.model.layers.73.self_attn.o_proj", "language_model.model.layers.73.self_attn.q_proj", "language_model.model.layers.73.self_attn.v_proj", "language_model.model.layers.74.self_attn.b_proj", "language_model.model.layers.74.self_attn.f_a_proj", "language_model.model.layers.74.self_attn.f_b_proj", "language_model.model.layers.74.self_attn.fused_qkvg_proj", "language_model.model.layers.74.self_attn.g_proj", "language_model.model.layers.74.self_attn.in_proj_qkvgfab", "language_model.model.layers.74.self_attn.k_proj", "language_model.model.layers.74.self_attn.o_proj", "language_model.model.layers.74.self_attn.q_proj", "language_model.model.layers.74.self_attn.v_proj", "language_model.model.layers.75.self_attn.fused_qkv_a_proj", "language_model.model.layers.75.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.75.self_attn.fused_qkvg_proj", "language_model.model.layers.75.self_attn.g_proj", "language_model.model.layers.75.self_attn.in_proj_qkvgfab", "language_model.model.layers.75.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.75.self_attn.kv_b_proj", "language_model.model.layers.75.self_attn.o_proj", "language_model.model.layers.75.self_attn.q_a_proj", "language_model.model.layers.75.self_attn.q_b_proj", "language_model.model.layers.76.self_attn.b_proj", "language_model.model.layers.76.self_attn.f_a_proj", "language_model.model.layers.76.self_attn.f_b_proj", "language_model.model.layers.76.self_attn.fused_qkvg_proj", "language_model.model.layers.76.self_attn.g_proj", "language_model.model.layers.76.self_attn.in_proj_qkvgfab", "language_model.model.layers.76.self_attn.k_proj", "language_model.model.layers.76.self_attn.o_proj", "language_model.model.layers.76.self_attn.q_proj", "language_model.model.layers.76.self_attn.v_proj", "language_model.model.layers.77.self_attn.b_proj", "language_model.model.layers.77.self_attn.f_a_proj", "language_model.model.layers.77.self_attn.f_b_proj", "language_model.model.layers.77.self_attn.fused_qkvg_proj", "language_model.model.layers.77.self_attn.g_proj", "language_model.model.layers.77.self_attn.in_proj_qkvgfab", "language_model.model.layers.77.self_attn.k_proj", "language_model.model.layers.77.self_attn.o_proj", "language_model.model.layers.77.self_attn.q_proj", "language_model.model.layers.77.self_attn.v_proj", "language_model.model.layers.78.self_attn.b_proj", "language_model.model.layers.78.self_attn.f_a_proj", "language_model.model.layers.78.self_attn.f_b_proj", "language_model.model.layers.78.self_attn.fused_qkvg_proj", "language_model.model.layers.78.self_attn.g_proj", "language_model.model.layers.78.self_attn.in_proj_qkvgfab", "language_model.model.layers.78.self_attn.k_proj", "language_model.model.layers.78.self_attn.o_proj", "language_model.model.layers.78.self_attn.q_proj", "language_model.model.layers.78.self_attn.v_proj", "language_model.model.layers.79.self_attn.fused_qkv_a_proj", "language_model.model.layers.79.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.79.self_attn.fused_qkvg_proj", "language_model.model.layers.79.self_attn.g_proj", "language_model.model.layers.79.self_attn.in_proj_qkvgfab", "language_model.model.layers.79.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.79.self_attn.kv_b_proj", "language_model.model.layers.79.self_attn.o_proj", "language_model.model.layers.79.self_attn.q_a_proj", "language_model.model.layers.79.self_attn.q_b_proj", "language_model.model.layers.8.self_attn.b_proj", "language_model.model.layers.8.self_attn.f_a_proj", "language_model.model.layers.8.self_attn.f_b_proj", "language_model.model.layers.8.self_attn.fused_qkvg_proj", "language_model.model.layers.8.self_attn.g_proj", "language_model.model.layers.8.self_attn.in_proj_qkvgfab", "language_model.model.layers.8.self_attn.k_proj", "language_model.model.layers.8.self_attn.o_proj", "language_model.model.layers.8.self_attn.q_proj", "language_model.model.layers.8.self_attn.v_proj", "language_model.model.layers.80.self_attn.b_proj", "language_model.model.layers.80.self_attn.f_a_proj", "language_model.model.layers.80.self_attn.f_b_proj", "language_model.model.layers.80.self_attn.fused_qkvg_proj", "language_model.model.layers.80.self_attn.g_proj", "language_model.model.layers.80.self_attn.in_proj_qkvgfab", "language_model.model.layers.80.self_attn.k_proj", "language_model.model.layers.80.self_attn.o_proj", "language_model.model.layers.80.self_attn.q_proj", "language_model.model.layers.80.self_attn.v_proj", "language_model.model.layers.81.self_attn.b_proj", "language_model.model.layers.81.self_attn.f_a_proj", "language_model.model.layers.81.self_attn.f_b_proj", "language_model.model.layers.81.self_attn.fused_qkvg_proj", "language_model.model.layers.81.self_attn.g_proj", "language_model.model.layers.81.self_attn.in_proj_qkvgfab", "language_model.model.layers.81.self_attn.k_proj", "language_model.model.layers.81.self_attn.o_proj", "language_model.model.layers.81.self_attn.q_proj", "language_model.model.layers.81.self_attn.v_proj", "language_model.model.layers.82.self_attn.b_proj", "language_model.model.layers.82.self_attn.f_a_proj", "language_model.model.layers.82.self_attn.f_b_proj", "language_model.model.layers.82.self_attn.fused_qkvg_proj", "language_model.model.layers.82.self_attn.g_proj", "language_model.model.layers.82.self_attn.in_proj_qkvgfab", "language_model.model.layers.82.self_attn.k_proj", "language_model.model.layers.82.self_attn.o_proj", "language_model.model.layers.82.self_attn.q_proj", "language_model.model.layers.82.self_attn.v_proj", "language_model.model.layers.83.self_attn.fused_qkv_a_proj", "language_model.model.layers.83.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.83.self_attn.fused_qkvg_proj", "language_model.model.layers.83.self_attn.g_proj", "language_model.model.layers.83.self_attn.in_proj_qkvgfab", "language_model.model.layers.83.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.83.self_attn.kv_b_proj", "language_model.model.layers.83.self_attn.o_proj", "language_model.model.layers.83.self_attn.q_a_proj", "language_model.model.layers.83.self_attn.q_b_proj", "language_model.model.layers.84.self_attn.b_proj", "language_model.model.layers.84.self_attn.f_a_proj", "language_model.model.layers.84.self_attn.f_b_proj", "language_model.model.layers.84.self_attn.fused_qkvg_proj", "language_model.model.layers.84.self_attn.g_proj", "language_model.model.layers.84.self_attn.in_proj_qkvgfab", "language_model.model.layers.84.self_attn.k_proj", "language_model.model.layers.84.self_attn.o_proj", "language_model.model.layers.84.self_attn.q_proj", "language_model.model.layers.84.self_attn.v_proj", "language_model.model.layers.85.self_attn.b_proj", "language_model.model.layers.85.self_attn.f_a_proj", "language_model.model.layers.85.self_attn.f_b_proj", "language_model.model.layers.85.self_attn.fused_qkvg_proj", "language_model.model.layers.85.self_attn.g_proj", "language_model.model.layers.85.self_attn.in_proj_qkvgfab", "language_model.model.layers.85.self_attn.k_proj", "language_model.model.layers.85.self_attn.o_proj", "language_model.model.layers.85.self_attn.q_proj", "language_model.model.layers.85.self_attn.v_proj", "language_model.model.layers.86.self_attn.b_proj", "language_model.model.layers.86.self_attn.f_a_proj", "language_model.model.layers.86.self_attn.f_b_proj", "language_model.model.layers.86.self_attn.fused_qkvg_proj", "language_model.model.layers.86.self_attn.g_proj", "language_model.model.layers.86.self_attn.in_proj_qkvgfab", "language_model.model.layers.86.self_attn.k_proj", "language_model.model.layers.86.self_attn.o_proj", "language_model.model.layers.86.self_attn.q_proj", "language_model.model.layers.86.self_attn.v_proj", "language_model.model.layers.87.self_attn.fused_qkv_a_proj", "language_model.model.layers.87.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.87.self_attn.fused_qkvg_proj", "language_model.model.layers.87.self_attn.g_proj", "language_model.model.layers.87.self_attn.in_proj_qkvgfab", "language_model.model.layers.87.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.87.self_attn.kv_b_proj", "language_model.model.layers.87.self_attn.o_proj", "language_model.model.layers.87.self_attn.q_a_proj", "language_model.model.layers.87.self_attn.q_b_proj", "language_model.model.layers.88.self_attn.b_proj", "language_model.model.layers.88.self_attn.f_a_proj", "language_model.model.layers.88.self_attn.f_b_proj", "language_model.model.layers.88.self_attn.fused_qkvg_proj", "language_model.model.layers.88.self_attn.g_proj", "language_model.model.layers.88.self_attn.in_proj_qkvgfab", "language_model.model.layers.88.self_attn.k_proj", "language_model.model.layers.88.self_attn.o_proj", "language_model.model.layers.88.self_attn.q_proj", "language_model.model.layers.88.self_attn.v_proj", "language_model.model.layers.89.self_attn.b_proj", "language_model.model.layers.89.self_attn.f_a_proj", "language_model.model.layers.89.self_attn.f_b_proj", "language_model.model.layers.89.self_attn.fused_qkvg_proj", "language_model.model.layers.89.self_attn.g_proj", "language_model.model.layers.89.self_attn.in_proj_qkvgfab", "language_model.model.layers.89.self_attn.k_proj", "language_model.model.layers.89.self_attn.o_proj", "language_model.model.layers.89.self_attn.q_proj", "language_model.model.layers.89.self_attn.v_proj", "language_model.model.layers.9.self_attn.b_proj", "language_model.model.layers.9.self_attn.f_a_proj", "language_model.model.layers.9.self_attn.f_b_proj", "language_model.model.layers.9.self_attn.fused_qkvg_proj", "language_model.model.layers.9.self_attn.g_proj", "language_model.model.layers.9.self_attn.in_proj_qkvgfab", "language_model.model.layers.9.self_attn.k_proj", "language_model.model.layers.9.self_attn.o_proj", "language_model.model.layers.9.self_attn.q_proj", "language_model.model.layers.9.self_attn.v_proj", "language_model.model.layers.90.self_attn.b_proj", "language_model.model.layers.90.self_attn.f_a_proj", "language_model.model.layers.90.self_attn.f_b_proj", "language_model.model.layers.90.self_attn.fused_qkvg_proj", "language_model.model.layers.90.self_attn.g_proj", "language_model.model.layers.90.self_attn.in_proj_qkvgfab", "language_model.model.layers.90.self_attn.k_proj", "language_model.model.layers.90.self_attn.o_proj", "language_model.model.layers.90.self_attn.q_proj", "language_model.model.layers.90.self_attn.v_proj", "language_model.model.layers.91.self_attn.fused_qkv_a_proj", "language_model.model.layers.91.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.91.self_attn.fused_qkvg_proj", "language_model.model.layers.91.self_attn.g_proj", "language_model.model.layers.91.self_attn.in_proj_qkvgfab", "language_model.model.layers.91.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.91.self_attn.kv_b_proj", "language_model.model.layers.91.self_attn.o_proj", "language_model.model.layers.91.self_attn.q_a_proj", "language_model.model.layers.91.self_attn.q_b_proj", "language_model.model.layers.92.self_attn.fused_qkv_a_proj", "language_model.model.layers.92.self_attn.fused_qkv_a_proj_with_mqa", "language_model.model.layers.92.self_attn.fused_qkvg_proj", "language_model.model.layers.92.self_attn.g_proj", "language_model.model.layers.92.self_attn.in_proj_qkvgfab", "language_model.model.layers.92.self_attn.kv_a_proj_with_mqa", "language_model.model.layers.92.self_attn.kv_b_proj", "language_model.model.layers.92.self_attn.o_proj", "language_model.model.layers.92.self_attn.q_a_proj", "language_model.model.layers.92.self_attn.q_b_proj", "layers.0.self_attn.b_proj", "layers.0.self_attn.f_a_proj", "layers.0.self_attn.f_b_proj", "layers.0.self_attn.fused_qkvg_proj", "layers.0.self_attn.g_proj", "layers.0.self_attn.in_proj_qkvgfab", "layers.0.self_attn.k_proj", "layers.0.self_attn.o_proj", "layers.0.self_attn.q_proj", "layers.0.self_attn.v_proj", "layers.1.self_attn.b_proj", "layers.1.self_attn.f_a_proj", "layers.1.self_attn.f_b_proj", "layers.1.self_attn.fused_qkvg_proj", "layers.1.self_attn.g_proj", "layers.1.self_attn.in_proj_qkvgfab", "layers.1.self_attn.k_proj", "layers.1.self_attn.o_proj", "layers.1.self_attn.q_proj", "layers.1.self_attn.v_proj", "layers.10.self_attn.b_proj", "layers.10.self_attn.f_a_proj", "layers.10.self_attn.f_b_proj", "layers.10.self_attn.fused_qkvg_proj", "layers.10.self_attn.g_proj", "layers.10.self_attn.in_proj_qkvgfab", "layers.10.self_attn.k_proj", "layers.10.self_attn.o_proj", "layers.10.self_attn.q_proj", "layers.10.self_attn.v_proj", "layers.11.self_attn.fused_qkv_a_proj", "layers.11.self_attn.fused_qkv_a_proj_with_mqa", "layers.11.self_attn.fused_qkvg_proj", "layers.11.self_attn.g_proj", "layers.11.self_attn.in_proj_qkvgfab", "layers.11.self_attn.kv_a_proj_with_mqa", "layers.11.self_attn.kv_b_proj", "layers.11.self_attn.o_proj", "layers.11.self_attn.q_a_proj", "layers.11.self_attn.q_b_proj", "layers.12.self_attn.b_proj", "layers.12.self_attn.f_a_proj", "layers.12.self_attn.f_b_proj", "layers.12.self_attn.fused_qkvg_proj", "layers.12.self_attn.g_proj", "layers.12.self_attn.in_proj_qkvgfab", "layers.12.self_attn.k_proj", "layers.12.self_attn.o_proj", "layers.12.self_attn.q_proj", "layers.12.self_attn.v_proj", "layers.13.self_attn.b_proj", "layers.13.self_attn.f_a_proj", "layers.13.self_attn.f_b_proj", "layers.13.self_attn.fused_qkvg_proj", "layers.13.self_attn.g_proj", "layers.13.self_attn.in_proj_qkvgfab", "layers.13.self_attn.k_proj", "layers.13.self_attn.o_proj", "layers.13.self_attn.q_proj", "layers.13.self_attn.v_proj", "layers.14.self_attn.b_proj", "layers.14.self_attn.f_a_proj", "layers.14.self_attn.f_b_proj", "layers.14.self_attn.fused_qkvg_proj", "layers.14.self_attn.g_proj", "layers.14.self_attn.in_proj_qkvgfab", "layers.14.self_attn.k_proj", "layers.14.self_attn.o_proj", "layers.14.self_attn.q_proj", "layers.14.self_attn.v_proj", "layers.15.self_attn.fused_qkv_a_proj", "layers.15.self_attn.fused_qkv_a_proj_with_mqa", "layers.15.self_attn.fused_qkvg_proj", "layers.15.self_attn.g_proj", "layers.15.self_attn.in_proj_qkvgfab", "layers.15.self_attn.kv_a_proj_with_mqa", "layers.15.self_attn.kv_b_proj", "layers.15.self_attn.o_proj", "layers.15.self_attn.q_a_proj", "layers.15.self_attn.q_b_proj", "layers.16.self_attn.b_proj", "layers.16.self_attn.f_a_proj", "layers.16.self_attn.f_b_proj", "layers.16.self_attn.fused_qkvg_proj", "layers.16.self_attn.g_proj", "layers.16.self_attn.in_proj_qkvgfab", "layers.16.self_attn.k_proj", "layers.16.self_attn.o_proj", "layers.16.self_attn.q_proj", "layers.16.self_attn.v_proj", "layers.17.self_attn.b_proj", "layers.17.self_attn.f_a_proj", "layers.17.self_attn.f_b_proj", "layers.17.self_attn.fused_qkvg_proj", "layers.17.self_attn.g_proj", "layers.17.self_attn.in_proj_qkvgfab", "layers.17.self_attn.k_proj", "layers.17.self_attn.o_proj", "layers.17.self_attn.q_proj", "layers.17.self_attn.v_proj", "layers.18.self_attn.b_proj", "layers.18.self_attn.f_a_proj", "layers.18.self_attn.f_b_proj", "layers.18.self_attn.fused_qkvg_proj", "layers.18.self_attn.g_proj", "layers.18.self_attn.in_proj_qkvgfab", "layers.18.self_attn.k_proj", "layers.18.self_attn.o_proj", "layers.18.self_attn.q_proj", "layers.18.self_attn.v_proj", "layers.19.self_attn.fused_qkv_a_proj", "layers.19.self_attn.fused_qkv_a_proj_with_mqa", "layers.19.self_attn.fused_qkvg_proj", "layers.19.self_attn.g_proj", "layers.19.self_attn.in_proj_qkvgfab", "layers.19.self_attn.kv_a_proj_with_mqa", "layers.19.self_attn.kv_b_proj", "layers.19.self_attn.o_proj", "layers.19.self_attn.q_a_proj", "layers.19.self_attn.q_b_proj", "layers.2.self_attn.b_proj", "layers.2.self_attn.f_a_proj", "layers.2.self_attn.f_b_proj", "layers.2.self_attn.fused_qkvg_proj", "layers.2.self_attn.g_proj", "layers.2.self_attn.in_proj_qkvgfab", "layers.2.self_attn.k_proj", "layers.2.self_attn.o_proj", "layers.2.self_attn.q_proj", "layers.2.self_attn.v_proj", "layers.20.self_attn.b_proj", "layers.20.self_attn.f_a_proj", "layers.20.self_attn.f_b_proj", "layers.20.self_attn.fused_qkvg_proj", "layers.20.self_attn.g_proj", "layers.20.self_attn.in_proj_qkvgfab", "layers.20.self_attn.k_proj", "layers.20.self_attn.o_proj", "layers.20.self_attn.q_proj", "layers.20.self_attn.v_proj", "layers.21.self_attn.b_proj", "layers.21.self_attn.f_a_proj", "layers.21.self_attn.f_b_proj", "layers.21.self_attn.fused_qkvg_proj", "layers.21.self_attn.g_proj", "layers.21.self_attn.in_proj_qkvgfab", "layers.21.self_attn.k_proj", "layers.21.self_attn.o_proj", "layers.21.self_attn.q_proj", "layers.21.self_attn.v_proj", "layers.22.self_attn.b_proj", "layers.22.self_attn.f_a_proj", "layers.22.self_attn.f_b_proj", "layers.22.self_attn.fused_qkvg_proj", "layers.22.self_attn.g_proj", "layers.22.self_attn.in_proj_qkvgfab", "layers.22.self_attn.k_proj", "layers.22.self_attn.o_proj", "layers.22.self_attn.q_proj", "layers.22.self_attn.v_proj", "layers.23.self_attn.fused_qkv_a_proj", "layers.23.self_attn.fused_qkv_a_proj_with_mqa", "layers.23.self_attn.fused_qkvg_proj", "layers.23.self_attn.g_proj", "layers.23.self_attn.in_proj_qkvgfab", "layers.23.self_attn.kv_a_proj_with_mqa", "layers.23.self_attn.kv_b_proj", "layers.23.self_attn.o_proj", "layers.23.self_attn.q_a_proj", "layers.23.self_attn.q_b_proj", "layers.24.self_attn.b_proj", "layers.24.self_attn.f_a_proj", "layers.24.self_attn.f_b_proj", "layers.24.self_attn.fused_qkvg_proj", "layers.24.self_attn.g_proj", "layers.24.self_attn.in_proj_qkvgfab", "layers.24.self_attn.k_proj", "layers.24.self_attn.o_proj", "layers.24.self_attn.q_proj", "layers.24.self_attn.v_proj", "layers.25.self_attn.b_proj", "layers.25.self_attn.f_a_proj", "layers.25.self_attn.f_b_proj", "layers.25.self_attn.fused_qkvg_proj", "layers.25.self_attn.g_proj", "layers.25.self_attn.in_proj_qkvgfab", "layers.25.self_attn.k_proj", "layers.25.self_attn.o_proj", "layers.25.self_attn.q_proj", "layers.25.self_attn.v_proj", "layers.26.self_attn.b_proj", "layers.26.self_attn.f_a_proj", "layers.26.self_attn.f_b_proj", "layers.26.self_attn.fused_qkvg_proj", "layers.26.self_attn.g_proj", "layers.26.self_attn.in_proj_qkvgfab", "layers.26.self_attn.k_proj", "layers.26.self_attn.o_proj", "layers.26.self_attn.q_proj", "layers.26.self_attn.v_proj", "layers.27.self_attn.fused_qkv_a_proj", "layers.27.self_attn.fused_qkv_a_proj_with_mqa", "layers.27.self_attn.fused_qkvg_proj", "layers.27.self_attn.g_proj", "layers.27.self_attn.in_proj_qkvgfab", "layers.27.self_attn.kv_a_proj_with_mqa", "layers.27.self_attn.kv_b_proj", "layers.27.self_attn.o_proj", "layers.27.self_attn.q_a_proj", "layers.27.self_attn.q_b_proj", "layers.28.self_attn.b_proj", "layers.28.self_attn.f_a_proj", "layers.28.self_attn.f_b_proj", "layers.28.self_attn.fused_qkvg_proj", "layers.28.self_attn.g_proj", "layers.28.self_attn.in_proj_qkvgfab", "layers.28.self_attn.k_proj", "layers.28.self_attn.o_proj", "layers.28.self_attn.q_proj", "layers.28.self_attn.v_proj", "layers.29.self_attn.b_proj", "layers.29.self_attn.f_a_proj", "layers.29.self_attn.f_b_proj", "layers.29.self_attn.fused_qkvg_proj", "layers.29.self_attn.g_proj", "layers.29.self_attn.in_proj_qkvgfab", "layers.29.self_attn.k_proj", "layers.29.self_attn.o_proj", "layers.29.self_attn.q_proj", "layers.29.self_attn.v_proj", "layers.3.self_attn.fused_qkv_a_proj", "layers.3.self_attn.fused_qkv_a_proj_with_mqa", "layers.3.self_attn.fused_qkvg_proj", "layers.3.self_attn.g_proj", "layers.3.self_attn.in_proj_qkvgfab", "layers.3.self_attn.kv_a_proj_with_mqa", "layers.3.self_attn.kv_b_proj", "layers.3.self_attn.o_proj", "layers.3.self_attn.q_a_proj", "layers.3.self_attn.q_b_proj", "layers.30.self_attn.b_proj", "layers.30.self_attn.f_a_proj", "layers.30.self_attn.f_b_proj", "layers.30.self_attn.fused_qkvg_proj", "layers.30.self_attn.g_proj", "layers.30.self_attn.in_proj_qkvgfab", "layers.30.self_attn.k_proj", "layers.30.self_attn.o_proj", "layers.30.self_attn.q_proj", "layers.30.self_attn.v_proj", "layers.31.self_attn.fused_qkv_a_proj", "layers.31.self_attn.fused_qkv_a_proj_with_mqa", "layers.31.self_attn.fused_qkvg_proj", "layers.31.self_attn.g_proj", "layers.31.self_attn.in_proj_qkvgfab", "layers.31.self_attn.kv_a_proj_with_mqa", "layers.31.self_attn.kv_b_proj", "layers.31.self_attn.o_proj", "layers.31.self_attn.q_a_proj", "layers.31.self_attn.q_b_proj", "layers.32.self_attn.b_proj", "layers.32.self_attn.f_a_proj", "layers.32.self_attn.f_b_proj", "layers.32.self_attn.fused_qkvg_proj", "layers.32.self_attn.g_proj", "layers.32.self_attn.in_proj_qkvgfab", "layers.32.self_attn.k_proj", "layers.32.self_attn.o_proj", "layers.32.self_attn.q_proj", "layers.32.self_attn.v_proj", "layers.33.self_attn.b_proj", "layers.33.self_attn.f_a_proj", "layers.33.self_attn.f_b_proj", "layers.33.self_attn.fused_qkvg_proj", "layers.33.self_attn.g_proj", "layers.33.self_attn.in_proj_qkvgfab", "layers.33.self_attn.k_proj", "layers.33.self_attn.o_proj", "layers.33.self_attn.q_proj", "layers.33.self_attn.v_proj", "layers.34.self_attn.b_proj", "layers.34.self_attn.f_a_proj", "layers.34.self_attn.f_b_proj", "layers.34.self_attn.fused_qkvg_proj", "layers.34.self_attn.g_proj", "layers.34.self_attn.in_proj_qkvgfab", "layers.34.self_attn.k_proj", "layers.34.self_attn.o_proj", "layers.34.self_attn.q_proj", "layers.34.self_attn.v_proj", "layers.35.self_attn.fused_qkv_a_proj", "layers.35.self_attn.fused_qkv_a_proj_with_mqa", "layers.35.self_attn.fused_qkvg_proj", "layers.35.self_attn.g_proj", "layers.35.self_attn.in_proj_qkvgfab", "layers.35.self_attn.kv_a_proj_with_mqa", "layers.35.self_attn.kv_b_proj", "layers.35.self_attn.o_proj", "layers.35.self_attn.q_a_proj", "layers.35.self_attn.q_b_proj", "layers.36.self_attn.b_proj", "layers.36.self_attn.f_a_proj", "layers.36.self_attn.f_b_proj", "layers.36.self_attn.fused_qkvg_proj", "layers.36.self_attn.g_proj", "layers.36.self_attn.in_proj_qkvgfab", "layers.36.self_attn.k_proj", "layers.36.self_attn.o_proj", "layers.36.self_attn.q_proj", "layers.36.self_attn.v_proj", "layers.37.self_attn.b_proj", "layers.37.self_attn.f_a_proj", "layers.37.self_attn.f_b_proj", "layers.37.self_attn.fused_qkvg_proj", "layers.37.self_attn.g_proj", "layers.37.self_attn.in_proj_qkvgfab", "layers.37.self_attn.k_proj", "layers.37.self_attn.o_proj", "layers.37.self_attn.q_proj", "layers.37.self_attn.v_proj", "layers.38.self_attn.b_proj", "layers.38.self_attn.f_a_proj", "layers.38.self_attn.f_b_proj", "layers.38.self_attn.fused_qkvg_proj", "layers.38.self_attn.g_proj", "layers.38.self_attn.in_proj_qkvgfab", "layers.38.self_attn.k_proj", "layers.38.self_attn.o_proj", "layers.38.self_attn.q_proj", "layers.38.self_attn.v_proj", "layers.39.self_attn.fused_qkv_a_proj", "layers.39.self_attn.fused_qkv_a_proj_with_mqa", "layers.39.self_attn.fused_qkvg_proj", "layers.39.self_attn.g_proj", "layers.39.self_attn.in_proj_qkvgfab", "layers.39.self_attn.kv_a_proj_with_mqa", "layers.39.self_attn.kv_b_proj", "layers.39.self_attn.o_proj", "layers.39.self_attn.q_a_proj", "layers.39.self_attn.q_b_proj", "layers.4.self_attn.b_proj", "layers.4.self_attn.f_a_proj", "layers.4.self_attn.f_b_proj", "layers.4.self_attn.fused_qkvg_proj", "layers.4.self_attn.g_proj", "layers.4.self_attn.in_proj_qkvgfab", "layers.4.self_attn.k_proj", "layers.4.self_attn.o_proj", "layers.4.self_attn.q_proj", "layers.4.self_attn.v_proj", "layers.40.self_attn.b_proj", "layers.40.self_attn.f_a_proj", "layers.40.self_attn.f_b_proj", "layers.40.self_attn.fused_qkvg_proj", "layers.40.self_attn.g_proj", "layers.40.self_attn.in_proj_qkvgfab", "layers.40.self_attn.k_proj", "layers.40.self_attn.o_proj", "layers.40.self_attn.q_proj", "layers.40.self_attn.v_proj", "layers.41.self_attn.b_proj", "layers.41.self_attn.f_a_proj", "layers.41.self_attn.f_b_proj", "layers.41.self_attn.fused_qkvg_proj", "layers.41.self_attn.g_proj", "layers.41.self_attn.in_proj_qkvgfab", "layers.41.self_attn.k_proj", "layers.41.self_attn.o_proj", "layers.41.self_attn.q_proj", "layers.41.self_attn.v_proj", "layers.42.self_attn.b_proj", "layers.42.self_attn.f_a_proj", "layers.42.self_attn.f_b_proj", "layers.42.self_attn.fused_qkvg_proj", "layers.42.self_attn.g_proj", "layers.42.self_attn.in_proj_qkvgfab", "layers.42.self_attn.k_proj", "layers.42.self_attn.o_proj", "layers.42.self_attn.q_proj", "layers.42.self_attn.v_proj", "layers.43.self_attn.fused_qkv_a_proj", "layers.43.self_attn.fused_qkv_a_proj_with_mqa", "layers.43.self_attn.fused_qkvg_proj", "layers.43.self_attn.g_proj", "layers.43.self_attn.in_proj_qkvgfab", "layers.43.self_attn.kv_a_proj_with_mqa", "layers.43.self_attn.kv_b_proj", "layers.43.self_attn.o_proj", "layers.43.self_attn.q_a_proj", "layers.43.self_attn.q_b_proj", "layers.44.self_attn.b_proj", "layers.44.self_attn.f_a_proj", "layers.44.self_attn.f_b_proj", "layers.44.self_attn.fused_qkvg_proj", "layers.44.self_attn.g_proj", "layers.44.self_attn.in_proj_qkvgfab", "layers.44.self_attn.k_proj", "layers.44.self_attn.o_proj", "layers.44.self_attn.q_proj", "layers.44.self_attn.v_proj", "layers.45.self_attn.b_proj", "layers.45.self_attn.f_a_proj", "layers.45.self_attn.f_b_proj", "layers.45.self_attn.fused_qkvg_proj", "layers.45.self_attn.g_proj", "layers.45.self_attn.in_proj_qkvgfab", "layers.45.self_attn.k_proj", "layers.45.self_attn.o_proj", "layers.45.self_attn.q_proj", "layers.45.self_attn.v_proj", "layers.46.self_attn.b_proj", "layers.46.self_attn.f_a_proj", "layers.46.self_attn.f_b_proj", "layers.46.self_attn.fused_qkvg_proj", "layers.46.self_attn.g_proj", "layers.46.self_attn.in_proj_qkvgfab", "layers.46.self_attn.k_proj", "layers.46.self_attn.o_proj", "layers.46.self_attn.q_proj", "layers.46.self_attn.v_proj", "layers.47.self_attn.fused_qkv_a_proj", "layers.47.self_attn.fused_qkv_a_proj_with_mqa", "layers.47.self_attn.fused_qkvg_proj", "layers.47.self_attn.g_proj", "layers.47.self_attn.in_proj_qkvgfab", "layers.47.self_attn.kv_a_proj_with_mqa", "layers.47.self_attn.kv_b_proj", "layers.47.self_attn.o_proj", "layers.47.self_attn.q_a_proj", "layers.47.self_attn.q_b_proj", "layers.48.self_attn.b_proj", "layers.48.self_attn.f_a_proj", "layers.48.self_attn.f_b_proj", "layers.48.self_attn.fused_qkvg_proj", "layers.48.self_attn.g_proj", "layers.48.self_attn.in_proj_qkvgfab", "layers.48.self_attn.k_proj", "layers.48.self_attn.o_proj", "layers.48.self_attn.q_proj", "layers.48.self_attn.v_proj", "layers.49.self_attn.b_proj", "layers.49.self_attn.f_a_proj", "layers.49.self_attn.f_b_proj", "layers.49.self_attn.fused_qkvg_proj", "layers.49.self_attn.g_proj", "layers.49.self_attn.in_proj_qkvgfab", "layers.49.self_attn.k_proj", "layers.49.self_attn.o_proj", "layers.49.self_attn.q_proj", "layers.49.self_attn.v_proj", "layers.5.self_attn.b_proj", "layers.5.self_attn.f_a_proj", "layers.5.self_attn.f_b_proj", "layers.5.self_attn.fused_qkvg_proj", "layers.5.self_attn.g_proj", "layers.5.self_attn.in_proj_qkvgfab", "layers.5.self_attn.k_proj", "layers.5.self_attn.o_proj", "layers.5.self_attn.q_proj", "layers.5.self_attn.v_proj", "layers.50.self_attn.b_proj", "layers.50.self_attn.f_a_proj", "layers.50.self_attn.f_b_proj", "layers.50.self_attn.fused_qkvg_proj", "layers.50.self_attn.g_proj", "layers.50.self_attn.in_proj_qkvgfab", "layers.50.self_attn.k_proj", "layers.50.self_attn.o_proj", "layers.50.self_attn.q_proj", "layers.50.self_attn.v_proj", "layers.51.self_attn.fused_qkv_a_proj", "layers.51.self_attn.fused_qkv_a_proj_with_mqa", "layers.51.self_attn.fused_qkvg_proj", "layers.51.self_attn.g_proj", "layers.51.self_attn.in_proj_qkvgfab", "layers.51.self_attn.kv_a_proj_with_mqa", "layers.51.self_attn.kv_b_proj", "layers.51.self_attn.o_proj", "layers.51.self_attn.q_a_proj", "layers.51.self_attn.q_b_proj", "layers.52.self_attn.b_proj", "layers.52.self_attn.f_a_proj", "layers.52.self_attn.f_b_proj", "layers.52.self_attn.fused_qkvg_proj", "layers.52.self_attn.g_proj", "layers.52.self_attn.in_proj_qkvgfab", "layers.52.self_attn.k_proj", "layers.52.self_attn.o_proj", "layers.52.self_attn.q_proj", "layers.52.self_attn.v_proj", "layers.53.self_attn.b_proj", "layers.53.self_attn.f_a_proj", "layers.53.self_attn.f_b_proj", "layers.53.self_attn.fused_qkvg_proj", "layers.53.self_attn.g_proj", "layers.53.self_attn.in_proj_qkvgfab", "layers.53.self_attn.k_proj", "layers.53.self_attn.o_proj", "layers.53.self_attn.q_proj", "layers.53.self_attn.v_proj", "layers.54.self_attn.b_proj", "layers.54.self_attn.f_a_proj", "layers.54.self_attn.f_b_proj", "layers.54.self_attn.fused_qkvg_proj", "layers.54.self_attn.g_proj", "layers.54.self_attn.in_proj_qkvgfab", "layers.54.self_attn.k_proj", "layers.54.self_attn.o_proj", "layers.54.self_attn.q_proj", "layers.54.self_attn.v_proj", "layers.55.self_attn.fused_qkv_a_proj", "layers.55.self_attn.fused_qkv_a_proj_with_mqa", "layers.55.self_attn.fused_qkvg_proj", "layers.55.self_attn.g_proj", "layers.55.self_attn.in_proj_qkvgfab", "layers.55.self_attn.kv_a_proj_with_mqa", "layers.55.self_attn.kv_b_proj", "layers.55.self_attn.o_proj", "layers.55.self_attn.q_a_proj", "layers.55.self_attn.q_b_proj", "layers.56.self_attn.b_proj", "layers.56.self_attn.f_a_proj", "layers.56.self_attn.f_b_proj", "layers.56.self_attn.fused_qkvg_proj", "layers.56.self_attn.g_proj", "layers.56.self_attn.in_proj_qkvgfab", "layers.56.self_attn.k_proj", "layers.56.self_attn.o_proj", "layers.56.self_attn.q_proj", "layers.56.self_attn.v_proj", "layers.57.self_attn.b_proj", "layers.57.self_attn.f_a_proj", "layers.57.self_attn.f_b_proj", "layers.57.self_attn.fused_qkvg_proj", "layers.57.self_attn.g_proj", "layers.57.self_attn.in_proj_qkvgfab", "layers.57.self_attn.k_proj", "layers.57.self_attn.o_proj", "layers.57.self_attn.q_proj", "layers.57.self_attn.v_proj", "layers.58.self_attn.b_proj", "layers.58.self_attn.f_a_proj", "layers.58.self_attn.f_b_proj", "layers.58.self_attn.fused_qkvg_proj", "layers.58.self_attn.g_proj", "layers.58.self_attn.in_proj_qkvgfab", "layers.58.self_attn.k_proj", "layers.58.self_attn.o_proj", "layers.58.self_attn.q_proj", "layers.58.self_attn.v_proj", "layers.59.self_attn.fused_qkv_a_proj", "layers.59.self_attn.fused_qkv_a_proj_with_mqa", "layers.59.self_attn.fused_qkvg_proj", "layers.59.self_attn.g_proj", "layers.59.self_attn.in_proj_qkvgfab", "layers.59.self_attn.kv_a_proj_with_mqa", "layers.59.self_attn.kv_b_proj", "layers.59.self_attn.o_proj", "layers.59.self_attn.q_a_proj", "layers.59.self_attn.q_b_proj", "layers.6.self_attn.b_proj", "layers.6.self_attn.f_a_proj", "layers.6.self_attn.f_b_proj", "layers.6.self_attn.fused_qkvg_proj", "layers.6.self_attn.g_proj", "layers.6.self_attn.in_proj_qkvgfab", "layers.6.self_attn.k_proj", "layers.6.self_attn.o_proj", "layers.6.self_attn.q_proj", "layers.6.self_attn.v_proj", "layers.60.self_attn.b_proj", "layers.60.self_attn.f_a_proj", "layers.60.self_attn.f_b_proj", "layers.60.self_attn.fused_qkvg_proj", "layers.60.self_attn.g_proj", "layers.60.self_attn.in_proj_qkvgfab", "layers.60.self_attn.k_proj", "layers.60.self_attn.o_proj", "layers.60.self_attn.q_proj", "layers.60.self_attn.v_proj", "layers.61.self_attn.b_proj", "layers.61.self_attn.f_a_proj", "layers.61.self_attn.f_b_proj", "layers.61.self_attn.fused_qkvg_proj", "layers.61.self_attn.g_proj", "layers.61.self_attn.in_proj_qkvgfab", "layers.61.self_attn.k_proj", "layers.61.self_attn.o_proj", "layers.61.self_attn.q_proj", "layers.61.self_attn.v_proj", "layers.62.self_attn.b_proj", "layers.62.self_attn.f_a_proj", "layers.62.self_attn.f_b_proj", "layers.62.self_attn.fused_qkvg_proj", "layers.62.self_attn.g_proj", "layers.62.self_attn.in_proj_qkvgfab", "layers.62.self_attn.k_proj", "layers.62.self_attn.o_proj", "layers.62.self_attn.q_proj", "layers.62.self_attn.v_proj", "layers.63.self_attn.fused_qkv_a_proj", "layers.63.self_attn.fused_qkv_a_proj_with_mqa", "layers.63.self_attn.fused_qkvg_proj", "layers.63.self_attn.g_proj", "layers.63.self_attn.in_proj_qkvgfab", "layers.63.self_attn.kv_a_proj_with_mqa", "layers.63.self_attn.kv_b_proj", "layers.63.self_attn.o_proj", "layers.63.self_attn.q_a_proj", "layers.63.self_attn.q_b_proj", "layers.64.self_attn.b_proj", "layers.64.self_attn.f_a_proj", "layers.64.self_attn.f_b_proj", "layers.64.self_attn.fused_qkvg_proj", "layers.64.self_attn.g_proj", "layers.64.self_attn.in_proj_qkvgfab", "layers.64.self_attn.k_proj", "layers.64.self_attn.o_proj", "layers.64.self_attn.q_proj", "layers.64.self_attn.v_proj", "layers.65.self_attn.b_proj", "layers.65.self_attn.f_a_proj", "layers.65.self_attn.f_b_proj", "layers.65.self_attn.fused_qkvg_proj", "layers.65.self_attn.g_proj", "layers.65.self_attn.in_proj_qkvgfab", "layers.65.self_attn.k_proj", "layers.65.self_attn.o_proj", "layers.65.self_attn.q_proj", "layers.65.self_attn.v_proj", "layers.66.self_attn.b_proj", "layers.66.self_attn.f_a_proj", "layers.66.self_attn.f_b_proj", "layers.66.self_attn.fused_qkvg_proj", "layers.66.self_attn.g_proj", "layers.66.self_attn.in_proj_qkvgfab", "layers.66.self_attn.k_proj", "layers.66.self_attn.o_proj", "layers.66.self_attn.q_proj", "layers.66.self_attn.v_proj", "layers.67.self_attn.fused_qkv_a_proj", "layers.67.self_attn.fused_qkv_a_proj_with_mqa", "layers.67.self_attn.fused_qkvg_proj", "layers.67.self_attn.g_proj", "layers.67.self_attn.in_proj_qkvgfab", "layers.67.self_attn.kv_a_proj_with_mqa", "layers.67.self_attn.kv_b_proj", "layers.67.self_attn.o_proj", "layers.67.self_attn.q_a_proj", "layers.67.self_attn.q_b_proj", "layers.68.self_attn.b_proj", "layers.68.self_attn.f_a_proj", "layers.68.self_attn.f_b_proj", "layers.68.self_attn.fused_qkvg_proj", "layers.68.self_attn.g_proj", "layers.68.self_attn.in_proj_qkvgfab", "layers.68.self_attn.k_proj", "layers.68.self_attn.o_proj", "layers.68.self_attn.q_proj", "layers.68.self_attn.v_proj", "layers.69.self_attn.b_proj", "layers.69.self_attn.f_a_proj", "layers.69.self_attn.f_b_proj", "layers.69.self_attn.fused_qkvg_proj", "layers.69.self_attn.g_proj", "layers.69.self_attn.in_proj_qkvgfab", "layers.69.self_attn.k_proj", "layers.69.self_attn.o_proj", "layers.69.self_attn.q_proj", "layers.69.self_attn.v_proj", "layers.7.self_attn.fused_qkv_a_proj", "layers.7.self_attn.fused_qkv_a_proj_with_mqa", "layers.7.self_attn.fused_qkvg_proj", "layers.7.self_attn.g_proj", "layers.7.self_attn.in_proj_qkvgfab", "layers.7.self_attn.kv_a_proj_with_mqa", "layers.7.self_attn.kv_b_proj", "layers.7.self_attn.o_proj", "layers.7.self_attn.q_a_proj", "layers.7.self_attn.q_b_proj", "layers.70.self_attn.b_proj", "layers.70.self_attn.f_a_proj", "layers.70.self_attn.f_b_proj", "layers.70.self_attn.fused_qkvg_proj", "layers.70.self_attn.g_proj", "layers.70.self_attn.in_proj_qkvgfab", "layers.70.self_attn.k_proj", "layers.70.self_attn.o_proj", "layers.70.self_attn.q_proj", "layers.70.self_attn.v_proj", "layers.71.self_attn.fused_qkv_a_proj", "layers.71.self_attn.fused_qkv_a_proj_with_mqa", "layers.71.self_attn.fused_qkvg_proj", "layers.71.self_attn.g_proj", "layers.71.self_attn.in_proj_qkvgfab", "layers.71.self_attn.kv_a_proj_with_mqa", "layers.71.self_attn.kv_b_proj", "layers.71.self_attn.o_proj", "layers.71.self_attn.q_a_proj", "layers.71.self_attn.q_b_proj", "layers.72.self_attn.b_proj", "layers.72.self_attn.f_a_proj", "layers.72.self_attn.f_b_proj", "layers.72.self_attn.fused_qkvg_proj", "layers.72.self_attn.g_proj", "layers.72.self_attn.in_proj_qkvgfab", "layers.72.self_attn.k_proj", "layers.72.self_attn.o_proj", "layers.72.self_attn.q_proj", "layers.72.self_attn.v_proj", "layers.73.self_attn.b_proj", "layers.73.self_attn.f_a_proj", "layers.73.self_attn.f_b_proj", "layers.73.self_attn.fused_qkvg_proj", "layers.73.self_attn.g_proj", "layers.73.self_attn.in_proj_qkvgfab", "layers.73.self_attn.k_proj", "layers.73.self_attn.o_proj", "layers.73.self_attn.q_proj", "layers.73.self_attn.v_proj", "layers.74.self_attn.b_proj", "layers.74.self_attn.f_a_proj", "layers.74.self_attn.f_b_proj", "layers.74.self_attn.fused_qkvg_proj", "layers.74.self_attn.g_proj", "layers.74.self_attn.in_proj_qkvgfab", "layers.74.self_attn.k_proj", "layers.74.self_attn.o_proj", "layers.74.self_attn.q_proj", "layers.74.self_attn.v_proj", "layers.75.self_attn.fused_qkv_a_proj", "layers.75.self_attn.fused_qkv_a_proj_with_mqa", "layers.75.self_attn.fused_qkvg_proj", "layers.75.self_attn.g_proj", "layers.75.self_attn.in_proj_qkvgfab", "layers.75.self_attn.kv_a_proj_with_mqa", "layers.75.self_attn.kv_b_proj", "layers.75.self_attn.o_proj", "layers.75.self_attn.q_a_proj", "layers.75.self_attn.q_b_proj", "layers.76.self_attn.b_proj", "layers.76.self_attn.f_a_proj", "layers.76.self_attn.f_b_proj", "layers.76.self_attn.fused_qkvg_proj", "layers.76.self_attn.g_proj", "layers.76.self_attn.in_proj_qkvgfab", "layers.76.self_attn.k_proj", "layers.76.self_attn.o_proj", "layers.76.self_attn.q_proj", "layers.76.self_attn.v_proj", "layers.77.self_attn.b_proj", "layers.77.self_attn.f_a_proj", "layers.77.self_attn.f_b_proj", "layers.77.self_attn.fused_qkvg_proj", "layers.77.self_attn.g_proj", "layers.77.self_attn.in_proj_qkvgfab", "layers.77.self_attn.k_proj", "layers.77.self_attn.o_proj", "layers.77.self_attn.q_proj", "layers.77.self_attn.v_proj", "layers.78.self_attn.b_proj", "layers.78.self_attn.f_a_proj", "layers.78.self_attn.f_b_proj", "layers.78.self_attn.fused_qkvg_proj", "layers.78.self_attn.g_proj", "layers.78.self_attn.in_proj_qkvgfab", "layers.78.self_attn.k_proj", "layers.78.self_attn.o_proj", "layers.78.self_attn.q_proj", "layers.78.self_attn.v_proj", "layers.79.self_attn.fused_qkv_a_proj", "layers.79.self_attn.fused_qkv_a_proj_with_mqa", "layers.79.self_attn.fused_qkvg_proj", "layers.79.self_attn.g_proj", "layers.79.self_attn.in_proj_qkvgfab", "layers.79.self_attn.kv_a_proj_with_mqa", "layers.79.self_attn.kv_b_proj", "layers.79.self_attn.o_proj", "layers.79.self_attn.q_a_proj", "layers.79.self_attn.q_b_proj", "layers.8.self_attn.b_proj", "layers.8.self_attn.f_a_proj", "layers.8.self_attn.f_b_proj", "layers.8.self_attn.fused_qkvg_proj", "layers.8.self_attn.g_proj", "layers.8.self_attn.in_proj_qkvgfab", "layers.8.self_attn.k_proj", "layers.8.self_attn.o_proj", "layers.8.self_attn.q_proj", "layers.8.self_attn.v_proj", "layers.80.self_attn.b_proj", "layers.80.self_attn.f_a_proj", "layers.80.self_attn.f_b_proj", "layers.80.self_attn.fused_qkvg_proj", "layers.80.self_attn.g_proj", "layers.80.self_attn.in_proj_qkvgfab", "layers.80.self_attn.k_proj", "layers.80.self_attn.o_proj", "layers.80.self_attn.q_proj", "layers.80.self_attn.v_proj", "layers.81.self_attn.b_proj", "layers.81.self_attn.f_a_proj", "layers.81.self_attn.f_b_proj", "layers.81.self_attn.fused_qkvg_proj", "layers.81.self_attn.g_proj", "layers.81.self_attn.in_proj_qkvgfab", "layers.81.self_attn.k_proj", "layers.81.self_attn.o_proj", "layers.81.self_attn.q_proj", "layers.81.self_attn.v_proj", "layers.82.self_attn.b_proj", "layers.82.self_attn.f_a_proj", "layers.82.self_attn.f_b_proj", "layers.82.self_attn.fused_qkvg_proj", "layers.82.self_attn.g_proj", "layers.82.self_attn.in_proj_qkvgfab", "layers.82.self_attn.k_proj", "layers.82.self_attn.o_proj", "layers.82.self_attn.q_proj", "layers.82.self_attn.v_proj", "layers.83.self_attn.fused_qkv_a_proj", "layers.83.self_attn.fused_qkv_a_proj_with_mqa", "layers.83.self_attn.fused_qkvg_proj", "layers.83.self_attn.g_proj", "layers.83.self_attn.in_proj_qkvgfab", "layers.83.self_attn.kv_a_proj_with_mqa", "layers.83.self_attn.kv_b_proj", "layers.83.self_attn.o_proj", "layers.83.self_attn.q_a_proj", "layers.83.self_attn.q_b_proj", "layers.84.self_attn.b_proj", "layers.84.self_attn.f_a_proj", "layers.84.self_attn.f_b_proj", "layers.84.self_attn.fused_qkvg_proj", "layers.84.self_attn.g_proj", "layers.84.self_attn.in_proj_qkvgfab", "layers.84.self_attn.k_proj", "layers.84.self_attn.o_proj", "layers.84.self_attn.q_proj", "layers.84.self_attn.v_proj", "layers.85.self_attn.b_proj", "layers.85.self_attn.f_a_proj", "layers.85.self_attn.f_b_proj", "layers.85.self_attn.fused_qkvg_proj", "layers.85.self_attn.g_proj", "layers.85.self_attn.in_proj_qkvgfab", "layers.85.self_attn.k_proj", "layers.85.self_attn.o_proj", "layers.85.self_attn.q_proj", "layers.85.self_attn.v_proj", "layers.86.self_attn.b_proj", "layers.86.self_attn.f_a_proj", "layers.86.self_attn.f_b_proj", "layers.86.self_attn.fused_qkvg_proj", "layers.86.self_attn.g_proj", "layers.86.self_attn.in_proj_qkvgfab", "layers.86.self_attn.k_proj", "layers.86.self_attn.o_proj", "layers.86.self_attn.q_proj", "layers.86.self_attn.v_proj", "layers.87.self_attn.fused_qkv_a_proj", "layers.87.self_attn.fused_qkv_a_proj_with_mqa", "layers.87.self_attn.fused_qkvg_proj", "layers.87.self_attn.g_proj", "layers.87.self_attn.in_proj_qkvgfab", "layers.87.self_attn.kv_a_proj_with_mqa", "layers.87.self_attn.kv_b_proj", "layers.87.self_attn.o_proj", "layers.87.self_attn.q_a_proj", "layers.87.self_attn.q_b_proj", "layers.88.self_attn.b_proj", "layers.88.self_attn.f_a_proj", "layers.88.self_attn.f_b_proj", "layers.88.self_attn.fused_qkvg_proj", "layers.88.self_attn.g_proj", "layers.88.self_attn.in_proj_qkvgfab", "layers.88.self_attn.k_proj", "layers.88.self_attn.o_proj", "layers.88.self_attn.q_proj", "layers.88.self_attn.v_proj", "layers.89.self_attn.b_proj", "layers.89.self_attn.f_a_proj", "layers.89.self_attn.f_b_proj", "layers.89.self_attn.fused_qkvg_proj", "layers.89.self_attn.g_proj", "layers.89.self_attn.in_proj_qkvgfab", "layers.89.self_attn.k_proj", "layers.89.self_attn.o_proj", "layers.89.self_attn.q_proj", "layers.89.self_attn.v_proj", "layers.9.self_attn.b_proj", "layers.9.self_attn.f_a_proj", "layers.9.self_attn.f_b_proj", "layers.9.self_attn.fused_qkvg_proj", "layers.9.self_attn.g_proj", "layers.9.self_attn.in_proj_qkvgfab", "layers.9.self_attn.k_proj", "layers.9.self_attn.o_proj", "layers.9.self_attn.q_proj", "layers.9.self_attn.v_proj", "layers.90.self_attn.b_proj", "layers.90.self_attn.f_a_proj", "layers.90.self_attn.f_b_proj", "layers.90.self_attn.fused_qkvg_proj", "layers.90.self_attn.g_proj", "layers.90.self_attn.in_proj_qkvgfab", "layers.90.self_attn.k_proj", "layers.90.self_attn.o_proj", "layers.90.self_attn.q_proj", "layers.90.self_attn.v_proj", "layers.91.self_attn.fused_qkv_a_proj", "layers.91.self_attn.fused_qkv_a_proj_with_mqa", "layers.91.self_attn.fused_qkvg_proj", "layers.91.self_attn.g_proj", "layers.91.self_attn.in_proj_qkvgfab", "layers.91.self_attn.kv_a_proj_with_mqa", "layers.91.self_attn.kv_b_proj", "layers.91.self_attn.o_proj", "layers.91.self_attn.q_a_proj", "layers.91.self_attn.q_b_proj", "layers.92.self_attn.fused_qkv_a_proj", "layers.92.self_attn.fused_qkv_a_proj_with_mqa", "layers.92.self_attn.fused_qkvg_proj", "layers.92.self_attn.g_proj", "layers.92.self_attn.in_proj_qkvgfab", "layers.92.self_attn.kv_a_proj_with_mqa", "layers.92.self_attn.kv_b_proj", "layers.92.self_attn.o_proj", "layers.92.self_attn.q_a_proj", "layers.92.self_attn.q_b_proj", "model.layers.0.self_attn.b_proj", "model.layers.0.self_attn.f_a_proj", "model.layers.0.self_attn.f_b_proj", "model.layers.0.self_attn.fused_qkvg_proj", "model.layers.0.self_attn.g_proj", "model.layers.0.self_attn.in_proj_qkvgfab", "model.layers.0.self_attn.k_proj", "model.layers.0.self_attn.o_proj", "model.layers.0.self_attn.q_proj", "model.layers.0.self_attn.v_proj", "model.layers.1.self_attn.b_proj", "model.layers.1.self_attn.f_a_proj", "model.layers.1.self_attn.f_b_proj", "model.layers.1.self_attn.fused_qkvg_proj", "model.layers.1.self_attn.g_proj", "model.layers.1.self_attn.in_proj_qkvgfab", "model.layers.1.self_attn.k_proj", "model.layers.1.self_attn.o_proj", "model.layers.1.self_attn.q_proj", "model.layers.1.self_attn.v_proj", "model.layers.10.self_attn.b_proj", "model.layers.10.self_attn.f_a_proj", "model.layers.10.self_attn.f_b_proj", "model.layers.10.self_attn.fused_qkvg_proj", "model.layers.10.self_attn.g_proj", "model.layers.10.self_attn.in_proj_qkvgfab", "model.layers.10.self_attn.k_proj", "model.layers.10.self_attn.o_proj", "model.layers.10.self_attn.q_proj", "model.layers.10.self_attn.v_proj", "model.layers.11.self_attn.fused_qkv_a_proj", "model.layers.11.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.11.self_attn.fused_qkvg_proj", "model.layers.11.self_attn.g_proj", "model.layers.11.self_attn.in_proj_qkvgfab", "model.layers.11.self_attn.kv_a_proj_with_mqa", "model.layers.11.self_attn.kv_b_proj", "model.layers.11.self_attn.o_proj", "model.layers.11.self_attn.q_a_proj", "model.layers.11.self_attn.q_b_proj", "model.layers.12.self_attn.b_proj", "model.layers.12.self_attn.f_a_proj", "model.layers.12.self_attn.f_b_proj", "model.layers.12.self_attn.fused_qkvg_proj", "model.layers.12.self_attn.g_proj", "model.layers.12.self_attn.in_proj_qkvgfab", "model.layers.12.self_attn.k_proj", "model.layers.12.self_attn.o_proj", "model.layers.12.self_attn.q_proj", "model.layers.12.self_attn.v_proj", "model.layers.13.self_attn.b_proj", "model.layers.13.self_attn.f_a_proj", "model.layers.13.self_attn.f_b_proj", "model.layers.13.self_attn.fused_qkvg_proj", "model.layers.13.self_attn.g_proj", "model.layers.13.self_attn.in_proj_qkvgfab", "model.layers.13.self_attn.k_proj", "model.layers.13.self_attn.o_proj", "model.layers.13.self_attn.q_proj", "model.layers.13.self_attn.v_proj", "model.layers.14.self_attn.b_proj", "model.layers.14.self_attn.f_a_proj", "model.layers.14.self_attn.f_b_proj", "model.layers.14.self_attn.fused_qkvg_proj", "model.layers.14.self_attn.g_proj", "model.layers.14.self_attn.in_proj_qkvgfab", "model.layers.14.self_attn.k_proj", "model.layers.14.self_attn.o_proj", "model.layers.14.self_attn.q_proj", "model.layers.14.self_attn.v_proj", "model.layers.15.self_attn.fused_qkv_a_proj", "model.layers.15.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.15.self_attn.fused_qkvg_proj", "model.layers.15.self_attn.g_proj", "model.layers.15.self_attn.in_proj_qkvgfab", "model.layers.15.self_attn.kv_a_proj_with_mqa", "model.layers.15.self_attn.kv_b_proj", "model.layers.15.self_attn.o_proj", "model.layers.15.self_attn.q_a_proj", "model.layers.15.self_attn.q_b_proj", "model.layers.16.self_attn.b_proj", "model.layers.16.self_attn.f_a_proj", "model.layers.16.self_attn.f_b_proj", "model.layers.16.self_attn.fused_qkvg_proj", "model.layers.16.self_attn.g_proj", "model.layers.16.self_attn.in_proj_qkvgfab", "model.layers.16.self_attn.k_proj", "model.layers.16.self_attn.o_proj", "model.layers.16.self_attn.q_proj", "model.layers.16.self_attn.v_proj", "model.layers.17.self_attn.b_proj", "model.layers.17.self_attn.f_a_proj", "model.layers.17.self_attn.f_b_proj", "model.layers.17.self_attn.fused_qkvg_proj", "model.layers.17.self_attn.g_proj", "model.layers.17.self_attn.in_proj_qkvgfab", "model.layers.17.self_attn.k_proj", "model.layers.17.self_attn.o_proj", "model.layers.17.self_attn.q_proj", "model.layers.17.self_attn.v_proj", "model.layers.18.self_attn.b_proj", "model.layers.18.self_attn.f_a_proj", "model.layers.18.self_attn.f_b_proj", "model.layers.18.self_attn.fused_qkvg_proj", "model.layers.18.self_attn.g_proj", "model.layers.18.self_attn.in_proj_qkvgfab", "model.layers.18.self_attn.k_proj", "model.layers.18.self_attn.o_proj", "model.layers.18.self_attn.q_proj", "model.layers.18.self_attn.v_proj", "model.layers.19.self_attn.fused_qkv_a_proj", "model.layers.19.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.19.self_attn.fused_qkvg_proj", "model.layers.19.self_attn.g_proj", "model.layers.19.self_attn.in_proj_qkvgfab", "model.layers.19.self_attn.kv_a_proj_with_mqa", "model.layers.19.self_attn.kv_b_proj", "model.layers.19.self_attn.o_proj", "model.layers.19.self_attn.q_a_proj", "model.layers.19.self_attn.q_b_proj", "model.layers.2.self_attn.b_proj", "model.layers.2.self_attn.f_a_proj", "model.layers.2.self_attn.f_b_proj", "model.layers.2.self_attn.fused_qkvg_proj", "model.layers.2.self_attn.g_proj", "model.layers.2.self_attn.in_proj_qkvgfab", "model.layers.2.self_attn.k_proj", "model.layers.2.self_attn.o_proj", "model.layers.2.self_attn.q_proj", "model.layers.2.self_attn.v_proj", "model.layers.20.self_attn.b_proj", "model.layers.20.self_attn.f_a_proj", "model.layers.20.self_attn.f_b_proj", "model.layers.20.self_attn.fused_qkvg_proj", "model.layers.20.self_attn.g_proj", "model.layers.20.self_attn.in_proj_qkvgfab", "model.layers.20.self_attn.k_proj", "model.layers.20.self_attn.o_proj", "model.layers.20.self_attn.q_proj", "model.layers.20.self_attn.v_proj", "model.layers.21.self_attn.b_proj", "model.layers.21.self_attn.f_a_proj", "model.layers.21.self_attn.f_b_proj", "model.layers.21.self_attn.fused_qkvg_proj", "model.layers.21.self_attn.g_proj", "model.layers.21.self_attn.in_proj_qkvgfab", "model.layers.21.self_attn.k_proj", "model.layers.21.self_attn.o_proj", "model.layers.21.self_attn.q_proj", "model.layers.21.self_attn.v_proj", "model.layers.22.self_attn.b_proj", "model.layers.22.self_attn.f_a_proj", "model.layers.22.self_attn.f_b_proj", "model.layers.22.self_attn.fused_qkvg_proj", "model.layers.22.self_attn.g_proj", "model.layers.22.self_attn.in_proj_qkvgfab", "model.layers.22.self_attn.k_proj", "model.layers.22.self_attn.o_proj", "model.layers.22.self_attn.q_proj", "model.layers.22.self_attn.v_proj", "model.layers.23.self_attn.fused_qkv_a_proj", "model.layers.23.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.23.self_attn.fused_qkvg_proj", "model.layers.23.self_attn.g_proj", "model.layers.23.self_attn.in_proj_qkvgfab", "model.layers.23.self_attn.kv_a_proj_with_mqa", "model.layers.23.self_attn.kv_b_proj", "model.layers.23.self_attn.o_proj", "model.layers.23.self_attn.q_a_proj", "model.layers.23.self_attn.q_b_proj", "model.layers.24.self_attn.b_proj", "model.layers.24.self_attn.f_a_proj", "model.layers.24.self_attn.f_b_proj", "model.layers.24.self_attn.fused_qkvg_proj", "model.layers.24.self_attn.g_proj", "model.layers.24.self_attn.in_proj_qkvgfab", "model.layers.24.self_attn.k_proj", "model.layers.24.self_attn.o_proj", "model.layers.24.self_attn.q_proj", "model.layers.24.self_attn.v_proj", "model.layers.25.self_attn.b_proj", "model.layers.25.self_attn.f_a_proj", "model.layers.25.self_attn.f_b_proj", "model.layers.25.self_attn.fused_qkvg_proj", "model.layers.25.self_attn.g_proj", "model.layers.25.self_attn.in_proj_qkvgfab", "model.layers.25.self_attn.k_proj", "model.layers.25.self_attn.o_proj", "model.layers.25.self_attn.q_proj", "model.layers.25.self_attn.v_proj", "model.layers.26.self_attn.b_proj", "model.layers.26.self_attn.f_a_proj", "model.layers.26.self_attn.f_b_proj", "model.layers.26.self_attn.fused_qkvg_proj", "model.layers.26.self_attn.g_proj", "model.layers.26.self_attn.in_proj_qkvgfab", "model.layers.26.self_attn.k_proj", "model.layers.26.self_attn.o_proj", "model.layers.26.self_attn.q_proj", "model.layers.26.self_attn.v_proj", "model.layers.27.self_attn.fused_qkv_a_proj", "model.layers.27.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.27.self_attn.fused_qkvg_proj", "model.layers.27.self_attn.g_proj", "model.layers.27.self_attn.in_proj_qkvgfab", "model.layers.27.self_attn.kv_a_proj_with_mqa", "model.layers.27.self_attn.kv_b_proj", "model.layers.27.self_attn.o_proj", "model.layers.27.self_attn.q_a_proj", "model.layers.27.self_attn.q_b_proj", "model.layers.28.self_attn.b_proj", "model.layers.28.self_attn.f_a_proj", "model.layers.28.self_attn.f_b_proj", "model.layers.28.self_attn.fused_qkvg_proj", "model.layers.28.self_attn.g_proj", "model.layers.28.self_attn.in_proj_qkvgfab", "model.layers.28.self_attn.k_proj", "model.layers.28.self_attn.o_proj", "model.layers.28.self_attn.q_proj", "model.layers.28.self_attn.v_proj", "model.layers.29.self_attn.b_proj", "model.layers.29.self_attn.f_a_proj", "model.layers.29.self_attn.f_b_proj", "model.layers.29.self_attn.fused_qkvg_proj", "model.layers.29.self_attn.g_proj", "model.layers.29.self_attn.in_proj_qkvgfab", "model.layers.29.self_attn.k_proj", "model.layers.29.self_attn.o_proj", "model.layers.29.self_attn.q_proj", "model.layers.29.self_attn.v_proj", "model.layers.3.self_attn.fused_qkv_a_proj", "model.layers.3.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.3.self_attn.fused_qkvg_proj", "model.layers.3.self_attn.g_proj", "model.layers.3.self_attn.in_proj_qkvgfab", "model.layers.3.self_attn.kv_a_proj_with_mqa", "model.layers.3.self_attn.kv_b_proj", "model.layers.3.self_attn.o_proj", "model.layers.3.self_attn.q_a_proj", "model.layers.3.self_attn.q_b_proj", "model.layers.30.self_attn.b_proj", "model.layers.30.self_attn.f_a_proj", "model.layers.30.self_attn.f_b_proj", "model.layers.30.self_attn.fused_qkvg_proj", "model.layers.30.self_attn.g_proj", "model.layers.30.self_attn.in_proj_qkvgfab", "model.layers.30.self_attn.k_proj", "model.layers.30.self_attn.o_proj", "model.layers.30.self_attn.q_proj", "model.layers.30.self_attn.v_proj", "model.layers.31.self_attn.fused_qkv_a_proj", "model.layers.31.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.31.self_attn.fused_qkvg_proj", "model.layers.31.self_attn.g_proj", "model.layers.31.self_attn.in_proj_qkvgfab", "model.layers.31.self_attn.kv_a_proj_with_mqa", "model.layers.31.self_attn.kv_b_proj", "model.layers.31.self_attn.o_proj", "model.layers.31.self_attn.q_a_proj", "model.layers.31.self_attn.q_b_proj", "model.layers.32.self_attn.b_proj", "model.layers.32.self_attn.f_a_proj", "model.layers.32.self_attn.f_b_proj", "model.layers.32.self_attn.fused_qkvg_proj", "model.layers.32.self_attn.g_proj", "model.layers.32.self_attn.in_proj_qkvgfab", "model.layers.32.self_attn.k_proj", "model.layers.32.self_attn.o_proj", "model.layers.32.self_attn.q_proj", "model.layers.32.self_attn.v_proj", "model.layers.33.self_attn.b_proj", "model.layers.33.self_attn.f_a_proj", "model.layers.33.self_attn.f_b_proj", "model.layers.33.self_attn.fused_qkvg_proj", "model.layers.33.self_attn.g_proj", "model.layers.33.self_attn.in_proj_qkvgfab", "model.layers.33.self_attn.k_proj", "model.layers.33.self_attn.o_proj", "model.layers.33.self_attn.q_proj", "model.layers.33.self_attn.v_proj", "model.layers.34.self_attn.b_proj", "model.layers.34.self_attn.f_a_proj", "model.layers.34.self_attn.f_b_proj", "model.layers.34.self_attn.fused_qkvg_proj", "model.layers.34.self_attn.g_proj", "model.layers.34.self_attn.in_proj_qkvgfab", "model.layers.34.self_attn.k_proj", "model.layers.34.self_attn.o_proj", "model.layers.34.self_attn.q_proj", "model.layers.34.self_attn.v_proj", "model.layers.35.self_attn.fused_qkv_a_proj", "model.layers.35.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.35.self_attn.fused_qkvg_proj", "model.layers.35.self_attn.g_proj", "model.layers.35.self_attn.in_proj_qkvgfab", "model.layers.35.self_attn.kv_a_proj_with_mqa", "model.layers.35.self_attn.kv_b_proj", "model.layers.35.self_attn.o_proj", "model.layers.35.self_attn.q_a_proj", "model.layers.35.self_attn.q_b_proj", "model.layers.36.self_attn.b_proj", "model.layers.36.self_attn.f_a_proj", "model.layers.36.self_attn.f_b_proj", "model.layers.36.self_attn.fused_qkvg_proj", "model.layers.36.self_attn.g_proj", "model.layers.36.self_attn.in_proj_qkvgfab", "model.layers.36.self_attn.k_proj", "model.layers.36.self_attn.o_proj", "model.layers.36.self_attn.q_proj", "model.layers.36.self_attn.v_proj", "model.layers.37.self_attn.b_proj", "model.layers.37.self_attn.f_a_proj", "model.layers.37.self_attn.f_b_proj", "model.layers.37.self_attn.fused_qkvg_proj", "model.layers.37.self_attn.g_proj", "model.layers.37.self_attn.in_proj_qkvgfab", "model.layers.37.self_attn.k_proj", "model.layers.37.self_attn.o_proj", "model.layers.37.self_attn.q_proj", "model.layers.37.self_attn.v_proj", "model.layers.38.self_attn.b_proj", "model.layers.38.self_attn.f_a_proj", "model.layers.38.self_attn.f_b_proj", "model.layers.38.self_attn.fused_qkvg_proj", "model.layers.38.self_attn.g_proj", "model.layers.38.self_attn.in_proj_qkvgfab", "model.layers.38.self_attn.k_proj", "model.layers.38.self_attn.o_proj", "model.layers.38.self_attn.q_proj", "model.layers.38.self_attn.v_proj", "model.layers.39.self_attn.fused_qkv_a_proj", "model.layers.39.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.39.self_attn.fused_qkvg_proj", "model.layers.39.self_attn.g_proj", "model.layers.39.self_attn.in_proj_qkvgfab", "model.layers.39.self_attn.kv_a_proj_with_mqa", "model.layers.39.self_attn.kv_b_proj", "model.layers.39.self_attn.o_proj", "model.layers.39.self_attn.q_a_proj", "model.layers.39.self_attn.q_b_proj", "model.layers.4.self_attn.b_proj", "model.layers.4.self_attn.f_a_proj", "model.layers.4.self_attn.f_b_proj", "model.layers.4.self_attn.fused_qkvg_proj", "model.layers.4.self_attn.g_proj", "model.layers.4.self_attn.in_proj_qkvgfab", "model.layers.4.self_attn.k_proj", "model.layers.4.self_attn.o_proj", "model.layers.4.self_attn.q_proj", "model.layers.4.self_attn.v_proj", "model.layers.40.self_attn.b_proj", "model.layers.40.self_attn.f_a_proj", "model.layers.40.self_attn.f_b_proj", "model.layers.40.self_attn.fused_qkvg_proj", "model.layers.40.self_attn.g_proj", "model.layers.40.self_attn.in_proj_qkvgfab", "model.layers.40.self_attn.k_proj", "model.layers.40.self_attn.o_proj", "model.layers.40.self_attn.q_proj", "model.layers.40.self_attn.v_proj", "model.layers.41.self_attn.b_proj", "model.layers.41.self_attn.f_a_proj", "model.layers.41.self_attn.f_b_proj", "model.layers.41.self_attn.fused_qkvg_proj", "model.layers.41.self_attn.g_proj", "model.layers.41.self_attn.in_proj_qkvgfab", "model.layers.41.self_attn.k_proj", "model.layers.41.self_attn.o_proj", "model.layers.41.self_attn.q_proj", "model.layers.41.self_attn.v_proj", "model.layers.42.self_attn.b_proj", "model.layers.42.self_attn.f_a_proj", "model.layers.42.self_attn.f_b_proj", "model.layers.42.self_attn.fused_qkvg_proj", "model.layers.42.self_attn.g_proj", "model.layers.42.self_attn.in_proj_qkvgfab", "model.layers.42.self_attn.k_proj", "model.layers.42.self_attn.o_proj", "model.layers.42.self_attn.q_proj", "model.layers.42.self_attn.v_proj", "model.layers.43.self_attn.fused_qkv_a_proj", "model.layers.43.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.43.self_attn.fused_qkvg_proj", "model.layers.43.self_attn.g_proj", "model.layers.43.self_attn.in_proj_qkvgfab", "model.layers.43.self_attn.kv_a_proj_with_mqa", "model.layers.43.self_attn.kv_b_proj", "model.layers.43.self_attn.o_proj", "model.layers.43.self_attn.q_a_proj", "model.layers.43.self_attn.q_b_proj", "model.layers.44.self_attn.b_proj", "model.layers.44.self_attn.f_a_proj", "model.layers.44.self_attn.f_b_proj", "model.layers.44.self_attn.fused_qkvg_proj", "model.layers.44.self_attn.g_proj", "model.layers.44.self_attn.in_proj_qkvgfab", "model.layers.44.self_attn.k_proj", "model.layers.44.self_attn.o_proj", "model.layers.44.self_attn.q_proj", "model.layers.44.self_attn.v_proj", "model.layers.45.self_attn.b_proj", "model.layers.45.self_attn.f_a_proj", "model.layers.45.self_attn.f_b_proj", "model.layers.45.self_attn.fused_qkvg_proj", "model.layers.45.self_attn.g_proj", "model.layers.45.self_attn.in_proj_qkvgfab", "model.layers.45.self_attn.k_proj", "model.layers.45.self_attn.o_proj", "model.layers.45.self_attn.q_proj", "model.layers.45.self_attn.v_proj", "model.layers.46.self_attn.b_proj", "model.layers.46.self_attn.f_a_proj", "model.layers.46.self_attn.f_b_proj", "model.layers.46.self_attn.fused_qkvg_proj", "model.layers.46.self_attn.g_proj", "model.layers.46.self_attn.in_proj_qkvgfab", "model.layers.46.self_attn.k_proj", "model.layers.46.self_attn.o_proj", "model.layers.46.self_attn.q_proj", "model.layers.46.self_attn.v_proj", "model.layers.47.self_attn.fused_qkv_a_proj", "model.layers.47.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.47.self_attn.fused_qkvg_proj", "model.layers.47.self_attn.g_proj", "model.layers.47.self_attn.in_proj_qkvgfab", "model.layers.47.self_attn.kv_a_proj_with_mqa", "model.layers.47.self_attn.kv_b_proj", "model.layers.47.self_attn.o_proj", "model.layers.47.self_attn.q_a_proj", "model.layers.47.self_attn.q_b_proj", "model.layers.48.self_attn.b_proj", "model.layers.48.self_attn.f_a_proj", "model.layers.48.self_attn.f_b_proj", "model.layers.48.self_attn.fused_qkvg_proj", "model.layers.48.self_attn.g_proj", "model.layers.48.self_attn.in_proj_qkvgfab", "model.layers.48.self_attn.k_proj", "model.layers.48.self_attn.o_proj", "model.layers.48.self_attn.q_proj", "model.layers.48.self_attn.v_proj", "model.layers.49.self_attn.b_proj", "model.layers.49.self_attn.f_a_proj", "model.layers.49.self_attn.f_b_proj", "model.layers.49.self_attn.fused_qkvg_proj", "model.layers.49.self_attn.g_proj", "model.layers.49.self_attn.in_proj_qkvgfab", "model.layers.49.self_attn.k_proj", "model.layers.49.self_attn.o_proj", "model.layers.49.self_attn.q_proj", "model.layers.49.self_attn.v_proj", "model.layers.5.self_attn.b_proj", "model.layers.5.self_attn.f_a_proj", "model.layers.5.self_attn.f_b_proj", "model.layers.5.self_attn.fused_qkvg_proj", "model.layers.5.self_attn.g_proj", "model.layers.5.self_attn.in_proj_qkvgfab", "model.layers.5.self_attn.k_proj", "model.layers.5.self_attn.o_proj", "model.layers.5.self_attn.q_proj", "model.layers.5.self_attn.v_proj", "model.layers.50.self_attn.b_proj", "model.layers.50.self_attn.f_a_proj", "model.layers.50.self_attn.f_b_proj", "model.layers.50.self_attn.fused_qkvg_proj", "model.layers.50.self_attn.g_proj", "model.layers.50.self_attn.in_proj_qkvgfab", "model.layers.50.self_attn.k_proj", "model.layers.50.self_attn.o_proj", "model.layers.50.self_attn.q_proj", "model.layers.50.self_attn.v_proj", "model.layers.51.self_attn.fused_qkv_a_proj", "model.layers.51.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.51.self_attn.fused_qkvg_proj", "model.layers.51.self_attn.g_proj", "model.layers.51.self_attn.in_proj_qkvgfab", "model.layers.51.self_attn.kv_a_proj_with_mqa", "model.layers.51.self_attn.kv_b_proj", "model.layers.51.self_attn.o_proj", "model.layers.51.self_attn.q_a_proj", "model.layers.51.self_attn.q_b_proj", "model.layers.52.self_attn.b_proj", "model.layers.52.self_attn.f_a_proj", "model.layers.52.self_attn.f_b_proj", "model.layers.52.self_attn.fused_qkvg_proj", "model.layers.52.self_attn.g_proj", "model.layers.52.self_attn.in_proj_qkvgfab", "model.layers.52.self_attn.k_proj", "model.layers.52.self_attn.o_proj", "model.layers.52.self_attn.q_proj", "model.layers.52.self_attn.v_proj", "model.layers.53.self_attn.b_proj", "model.layers.53.self_attn.f_a_proj", "model.layers.53.self_attn.f_b_proj", "model.layers.53.self_attn.fused_qkvg_proj", "model.layers.53.self_attn.g_proj", "model.layers.53.self_attn.in_proj_qkvgfab", "model.layers.53.self_attn.k_proj", "model.layers.53.self_attn.o_proj", "model.layers.53.self_attn.q_proj", "model.layers.53.self_attn.v_proj", "model.layers.54.self_attn.b_proj", "model.layers.54.self_attn.f_a_proj", "model.layers.54.self_attn.f_b_proj", "model.layers.54.self_attn.fused_qkvg_proj", "model.layers.54.self_attn.g_proj", "model.layers.54.self_attn.in_proj_qkvgfab", "model.layers.54.self_attn.k_proj", "model.layers.54.self_attn.o_proj", "model.layers.54.self_attn.q_proj", "model.layers.54.self_attn.v_proj", "model.layers.55.self_attn.fused_qkv_a_proj", "model.layers.55.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.55.self_attn.fused_qkvg_proj", "model.layers.55.self_attn.g_proj", "model.layers.55.self_attn.in_proj_qkvgfab", "model.layers.55.self_attn.kv_a_proj_with_mqa", "model.layers.55.self_attn.kv_b_proj", "model.layers.55.self_attn.o_proj", "model.layers.55.self_attn.q_a_proj", "model.layers.55.self_attn.q_b_proj", "model.layers.56.self_attn.b_proj", "model.layers.56.self_attn.f_a_proj", "model.layers.56.self_attn.f_b_proj", "model.layers.56.self_attn.fused_qkvg_proj", "model.layers.56.self_attn.g_proj", "model.layers.56.self_attn.in_proj_qkvgfab", "model.layers.56.self_attn.k_proj", "model.layers.56.self_attn.o_proj", "model.layers.56.self_attn.q_proj", "model.layers.56.self_attn.v_proj", "model.layers.57.self_attn.b_proj", "model.layers.57.self_attn.f_a_proj", "model.layers.57.self_attn.f_b_proj", "model.layers.57.self_attn.fused_qkvg_proj", "model.layers.57.self_attn.g_proj", "model.layers.57.self_attn.in_proj_qkvgfab", "model.layers.57.self_attn.k_proj", "model.layers.57.self_attn.o_proj", "model.layers.57.self_attn.q_proj", "model.layers.57.self_attn.v_proj", "model.layers.58.self_attn.b_proj", "model.layers.58.self_attn.f_a_proj", "model.layers.58.self_attn.f_b_proj", "model.layers.58.self_attn.fused_qkvg_proj", "model.layers.58.self_attn.g_proj", "model.layers.58.self_attn.in_proj_qkvgfab", "model.layers.58.self_attn.k_proj", "model.layers.58.self_attn.o_proj", "model.layers.58.self_attn.q_proj", "model.layers.58.self_attn.v_proj", "model.layers.59.self_attn.fused_qkv_a_proj", "model.layers.59.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.59.self_attn.fused_qkvg_proj", "model.layers.59.self_attn.g_proj", "model.layers.59.self_attn.in_proj_qkvgfab", "model.layers.59.self_attn.kv_a_proj_with_mqa", "model.layers.59.self_attn.kv_b_proj", "model.layers.59.self_attn.o_proj", "model.layers.59.self_attn.q_a_proj", "model.layers.59.self_attn.q_b_proj", "model.layers.6.self_attn.b_proj", "model.layers.6.self_attn.f_a_proj", "model.layers.6.self_attn.f_b_proj", "model.layers.6.self_attn.fused_qkvg_proj", "model.layers.6.self_attn.g_proj", "model.layers.6.self_attn.in_proj_qkvgfab", "model.layers.6.self_attn.k_proj", "model.layers.6.self_attn.o_proj", "model.layers.6.self_attn.q_proj", "model.layers.6.self_attn.v_proj", "model.layers.60.self_attn.b_proj", "model.layers.60.self_attn.f_a_proj", "model.layers.60.self_attn.f_b_proj", "model.layers.60.self_attn.fused_qkvg_proj", "model.layers.60.self_attn.g_proj", "model.layers.60.self_attn.in_proj_qkvgfab", "model.layers.60.self_attn.k_proj", "model.layers.60.self_attn.o_proj", "model.layers.60.self_attn.q_proj", "model.layers.60.self_attn.v_proj", "model.layers.61.self_attn.b_proj", "model.layers.61.self_attn.f_a_proj", "model.layers.61.self_attn.f_b_proj", "model.layers.61.self_attn.fused_qkvg_proj", "model.layers.61.self_attn.g_proj", "model.layers.61.self_attn.in_proj_qkvgfab", "model.layers.61.self_attn.k_proj", "model.layers.61.self_attn.o_proj", "model.layers.61.self_attn.q_proj", "model.layers.61.self_attn.v_proj", "model.layers.62.self_attn.b_proj", "model.layers.62.self_attn.f_a_proj", "model.layers.62.self_attn.f_b_proj", "model.layers.62.self_attn.fused_qkvg_proj", "model.layers.62.self_attn.g_proj", "model.layers.62.self_attn.in_proj_qkvgfab", "model.layers.62.self_attn.k_proj", "model.layers.62.self_attn.o_proj", "model.layers.62.self_attn.q_proj", "model.layers.62.self_attn.v_proj", "model.layers.63.self_attn.fused_qkv_a_proj", "model.layers.63.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.63.self_attn.fused_qkvg_proj", "model.layers.63.self_attn.g_proj", "model.layers.63.self_attn.in_proj_qkvgfab", "model.layers.63.self_attn.kv_a_proj_with_mqa", "model.layers.63.self_attn.kv_b_proj", "model.layers.63.self_attn.o_proj", "model.layers.63.self_attn.q_a_proj", "model.layers.63.self_attn.q_b_proj", "model.layers.64.self_attn.b_proj", "model.layers.64.self_attn.f_a_proj", "model.layers.64.self_attn.f_b_proj", "model.layers.64.self_attn.fused_qkvg_proj", "model.layers.64.self_attn.g_proj", "model.layers.64.self_attn.in_proj_qkvgfab", "model.layers.64.self_attn.k_proj", "model.layers.64.self_attn.o_proj", "model.layers.64.self_attn.q_proj", "model.layers.64.self_attn.v_proj", "model.layers.65.self_attn.b_proj", "model.layers.65.self_attn.f_a_proj", "model.layers.65.self_attn.f_b_proj", "model.layers.65.self_attn.fused_qkvg_proj", "model.layers.65.self_attn.g_proj", "model.layers.65.self_attn.in_proj_qkvgfab", "model.layers.65.self_attn.k_proj", "model.layers.65.self_attn.o_proj", "model.layers.65.self_attn.q_proj", "model.layers.65.self_attn.v_proj", "model.layers.66.self_attn.b_proj", "model.layers.66.self_attn.f_a_proj", "model.layers.66.self_attn.f_b_proj", "model.layers.66.self_attn.fused_qkvg_proj", "model.layers.66.self_attn.g_proj", "model.layers.66.self_attn.in_proj_qkvgfab", "model.layers.66.self_attn.k_proj", "model.layers.66.self_attn.o_proj", "model.layers.66.self_attn.q_proj", "model.layers.66.self_attn.v_proj", "model.layers.67.self_attn.fused_qkv_a_proj", "model.layers.67.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.67.self_attn.fused_qkvg_proj", "model.layers.67.self_attn.g_proj", "model.layers.67.self_attn.in_proj_qkvgfab", "model.layers.67.self_attn.kv_a_proj_with_mqa", "model.layers.67.self_attn.kv_b_proj", "model.layers.67.self_attn.o_proj", "model.layers.67.self_attn.q_a_proj", "model.layers.67.self_attn.q_b_proj", "model.layers.68.self_attn.b_proj", "model.layers.68.self_attn.f_a_proj", "model.layers.68.self_attn.f_b_proj", "model.layers.68.self_attn.fused_qkvg_proj", "model.layers.68.self_attn.g_proj", "model.layers.68.self_attn.in_proj_qkvgfab", "model.layers.68.self_attn.k_proj", "model.layers.68.self_attn.o_proj", "model.layers.68.self_attn.q_proj", "model.layers.68.self_attn.v_proj", "model.layers.69.self_attn.b_proj", "model.layers.69.self_attn.f_a_proj", "model.layers.69.self_attn.f_b_proj", "model.layers.69.self_attn.fused_qkvg_proj", "model.layers.69.self_attn.g_proj", "model.layers.69.self_attn.in_proj_qkvgfab", "model.layers.69.self_attn.k_proj", "model.layers.69.self_attn.o_proj", "model.layers.69.self_attn.q_proj", "model.layers.69.self_attn.v_proj", "model.layers.7.self_attn.fused_qkv_a_proj", "model.layers.7.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.7.self_attn.fused_qkvg_proj", "model.layers.7.self_attn.g_proj", "model.layers.7.self_attn.in_proj_qkvgfab", "model.layers.7.self_attn.kv_a_proj_with_mqa", "model.layers.7.self_attn.kv_b_proj", "model.layers.7.self_attn.o_proj", "model.layers.7.self_attn.q_a_proj", "model.layers.7.self_attn.q_b_proj", "model.layers.70.self_attn.b_proj", "model.layers.70.self_attn.f_a_proj", "model.layers.70.self_attn.f_b_proj", "model.layers.70.self_attn.fused_qkvg_proj", "model.layers.70.self_attn.g_proj", "model.layers.70.self_attn.in_proj_qkvgfab", "model.layers.70.self_attn.k_proj", "model.layers.70.self_attn.o_proj", "model.layers.70.self_attn.q_proj", "model.layers.70.self_attn.v_proj", "model.layers.71.self_attn.fused_qkv_a_proj", "model.layers.71.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.71.self_attn.fused_qkvg_proj", "model.layers.71.self_attn.g_proj", "model.layers.71.self_attn.in_proj_qkvgfab", "model.layers.71.self_attn.kv_a_proj_with_mqa", "model.layers.71.self_attn.kv_b_proj", "model.layers.71.self_attn.o_proj", "model.layers.71.self_attn.q_a_proj", "model.layers.71.self_attn.q_b_proj", "model.layers.72.self_attn.b_proj", "model.layers.72.self_attn.f_a_proj", "model.layers.72.self_attn.f_b_proj", "model.layers.72.self_attn.fused_qkvg_proj", "model.layers.72.self_attn.g_proj", "model.layers.72.self_attn.in_proj_qkvgfab", "model.layers.72.self_attn.k_proj", "model.layers.72.self_attn.o_proj", "model.layers.72.self_attn.q_proj", "model.layers.72.self_attn.v_proj", "model.layers.73.self_attn.b_proj", "model.layers.73.self_attn.f_a_proj", "model.layers.73.self_attn.f_b_proj", "model.layers.73.self_attn.fused_qkvg_proj", "model.layers.73.self_attn.g_proj", "model.layers.73.self_attn.in_proj_qkvgfab", "model.layers.73.self_attn.k_proj", "model.layers.73.self_attn.o_proj", "model.layers.73.self_attn.q_proj", "model.layers.73.self_attn.v_proj", "model.layers.74.self_attn.b_proj", "model.layers.74.self_attn.f_a_proj", "model.layers.74.self_attn.f_b_proj", "model.layers.74.self_attn.fused_qkvg_proj", "model.layers.74.self_attn.g_proj", "model.layers.74.self_attn.in_proj_qkvgfab", "model.layers.74.self_attn.k_proj", "model.layers.74.self_attn.o_proj", "model.layers.74.self_attn.q_proj", "model.layers.74.self_attn.v_proj", "model.layers.75.self_attn.fused_qkv_a_proj", "model.layers.75.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.75.self_attn.fused_qkvg_proj", "model.layers.75.self_attn.g_proj", "model.layers.75.self_attn.in_proj_qkvgfab", "model.layers.75.self_attn.kv_a_proj_with_mqa", "model.layers.75.self_attn.kv_b_proj", "model.layers.75.self_attn.o_proj", "model.layers.75.self_attn.q_a_proj", "model.layers.75.self_attn.q_b_proj", "model.layers.76.self_attn.b_proj", "model.layers.76.self_attn.f_a_proj", "model.layers.76.self_attn.f_b_proj", "model.layers.76.self_attn.fused_qkvg_proj", "model.layers.76.self_attn.g_proj", "model.layers.76.self_attn.in_proj_qkvgfab", "model.layers.76.self_attn.k_proj", "model.layers.76.self_attn.o_proj", "model.layers.76.self_attn.q_proj", "model.layers.76.self_attn.v_proj", "model.layers.77.self_attn.b_proj", "model.layers.77.self_attn.f_a_proj", "model.layers.77.self_attn.f_b_proj", "model.layers.77.self_attn.fused_qkvg_proj", "model.layers.77.self_attn.g_proj", "model.layers.77.self_attn.in_proj_qkvgfab", "model.layers.77.self_attn.k_proj", "model.layers.77.self_attn.o_proj", "model.layers.77.self_attn.q_proj", "model.layers.77.self_attn.v_proj", "model.layers.78.self_attn.b_proj", "model.layers.78.self_attn.f_a_proj", "model.layers.78.self_attn.f_b_proj", "model.layers.78.self_attn.fused_qkvg_proj", "model.layers.78.self_attn.g_proj", "model.layers.78.self_attn.in_proj_qkvgfab", "model.layers.78.self_attn.k_proj", "model.layers.78.self_attn.o_proj", "model.layers.78.self_attn.q_proj", "model.layers.78.self_attn.v_proj", "model.layers.79.self_attn.fused_qkv_a_proj", "model.layers.79.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.79.self_attn.fused_qkvg_proj", "model.layers.79.self_attn.g_proj", "model.layers.79.self_attn.in_proj_qkvgfab", "model.layers.79.self_attn.kv_a_proj_with_mqa", "model.layers.79.self_attn.kv_b_proj", "model.layers.79.self_attn.o_proj", "model.layers.79.self_attn.q_a_proj", "model.layers.79.self_attn.q_b_proj", "model.layers.8.self_attn.b_proj", "model.layers.8.self_attn.f_a_proj", "model.layers.8.self_attn.f_b_proj", "model.layers.8.self_attn.fused_qkvg_proj", "model.layers.8.self_attn.g_proj", "model.layers.8.self_attn.in_proj_qkvgfab", "model.layers.8.self_attn.k_proj", "model.layers.8.self_attn.o_proj", "model.layers.8.self_attn.q_proj", "model.layers.8.self_attn.v_proj", "model.layers.80.self_attn.b_proj", "model.layers.80.self_attn.f_a_proj", "model.layers.80.self_attn.f_b_proj", "model.layers.80.self_attn.fused_qkvg_proj", "model.layers.80.self_attn.g_proj", "model.layers.80.self_attn.in_proj_qkvgfab", "model.layers.80.self_attn.k_proj", "model.layers.80.self_attn.o_proj", "model.layers.80.self_attn.q_proj", "model.layers.80.self_attn.v_proj", "model.layers.81.self_attn.b_proj", "model.layers.81.self_attn.f_a_proj", "model.layers.81.self_attn.f_b_proj", "model.layers.81.self_attn.fused_qkvg_proj", "model.layers.81.self_attn.g_proj", "model.layers.81.self_attn.in_proj_qkvgfab", "model.layers.81.self_attn.k_proj", "model.layers.81.self_attn.o_proj", "model.layers.81.self_attn.q_proj", "model.layers.81.self_attn.v_proj", "model.layers.82.self_attn.b_proj", "model.layers.82.self_attn.f_a_proj", "model.layers.82.self_attn.f_b_proj", "model.layers.82.self_attn.fused_qkvg_proj", "model.layers.82.self_attn.g_proj", "model.layers.82.self_attn.in_proj_qkvgfab", "model.layers.82.self_attn.k_proj", "model.layers.82.self_attn.o_proj", "model.layers.82.self_attn.q_proj", "model.layers.82.self_attn.v_proj", "model.layers.83.self_attn.fused_qkv_a_proj", "model.layers.83.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.83.self_attn.fused_qkvg_proj", "model.layers.83.self_attn.g_proj", "model.layers.83.self_attn.in_proj_qkvgfab", "model.layers.83.self_attn.kv_a_proj_with_mqa", "model.layers.83.self_attn.kv_b_proj", "model.layers.83.self_attn.o_proj", "model.layers.83.self_attn.q_a_proj", "model.layers.83.self_attn.q_b_proj", "model.layers.84.self_attn.b_proj", "model.layers.84.self_attn.f_a_proj", "model.layers.84.self_attn.f_b_proj", "model.layers.84.self_attn.fused_qkvg_proj", "model.layers.84.self_attn.g_proj", "model.layers.84.self_attn.in_proj_qkvgfab", "model.layers.84.self_attn.k_proj", "model.layers.84.self_attn.o_proj", "model.layers.84.self_attn.q_proj", "model.layers.84.self_attn.v_proj", "model.layers.85.self_attn.b_proj", "model.layers.85.self_attn.f_a_proj", "model.layers.85.self_attn.f_b_proj", "model.layers.85.self_attn.fused_qkvg_proj", "model.layers.85.self_attn.g_proj", "model.layers.85.self_attn.in_proj_qkvgfab", "model.layers.85.self_attn.k_proj", "model.layers.85.self_attn.o_proj", "model.layers.85.self_attn.q_proj", "model.layers.85.self_attn.v_proj", "model.layers.86.self_attn.b_proj", "model.layers.86.self_attn.f_a_proj", "model.layers.86.self_attn.f_b_proj", "model.layers.86.self_attn.fused_qkvg_proj", "model.layers.86.self_attn.g_proj", "model.layers.86.self_attn.in_proj_qkvgfab", "model.layers.86.self_attn.k_proj", "model.layers.86.self_attn.o_proj", "model.layers.86.self_attn.q_proj", "model.layers.86.self_attn.v_proj", "model.layers.87.self_attn.fused_qkv_a_proj", "model.layers.87.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.87.self_attn.fused_qkvg_proj", "model.layers.87.self_attn.g_proj", "model.layers.87.self_attn.in_proj_qkvgfab", "model.layers.87.self_attn.kv_a_proj_with_mqa", "model.layers.87.self_attn.kv_b_proj", "model.layers.87.self_attn.o_proj", "model.layers.87.self_attn.q_a_proj", "model.layers.87.self_attn.q_b_proj", "model.layers.88.self_attn.b_proj", "model.layers.88.self_attn.f_a_proj", "model.layers.88.self_attn.f_b_proj", "model.layers.88.self_attn.fused_qkvg_proj", "model.layers.88.self_attn.g_proj", "model.layers.88.self_attn.in_proj_qkvgfab", "model.layers.88.self_attn.k_proj", "model.layers.88.self_attn.o_proj", "model.layers.88.self_attn.q_proj", "model.layers.88.self_attn.v_proj", "model.layers.89.self_attn.b_proj", "model.layers.89.self_attn.f_a_proj", "model.layers.89.self_attn.f_b_proj", "model.layers.89.self_attn.fused_qkvg_proj", "model.layers.89.self_attn.g_proj", "model.layers.89.self_attn.in_proj_qkvgfab", "model.layers.89.self_attn.k_proj", "model.layers.89.self_attn.o_proj", "model.layers.89.self_attn.q_proj", "model.layers.89.self_attn.v_proj", "model.layers.9.self_attn.b_proj", "model.layers.9.self_attn.f_a_proj", "model.layers.9.self_attn.f_b_proj", "model.layers.9.self_attn.fused_qkvg_proj", "model.layers.9.self_attn.g_proj", "model.layers.9.self_attn.in_proj_qkvgfab", "model.layers.9.self_attn.k_proj", "model.layers.9.self_attn.o_proj", "model.layers.9.self_attn.q_proj", "model.layers.9.self_attn.v_proj", "model.layers.90.self_attn.b_proj", "model.layers.90.self_attn.f_a_proj", "model.layers.90.self_attn.f_b_proj", "model.layers.90.self_attn.fused_qkvg_proj", "model.layers.90.self_attn.g_proj", "model.layers.90.self_attn.in_proj_qkvgfab", "model.layers.90.self_attn.k_proj", "model.layers.90.self_attn.o_proj", "model.layers.90.self_attn.q_proj", "model.layers.90.self_attn.v_proj", "model.layers.91.self_attn.fused_qkv_a_proj", "model.layers.91.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.91.self_attn.fused_qkvg_proj", "model.layers.91.self_attn.g_proj", "model.layers.91.self_attn.in_proj_qkvgfab", "model.layers.91.self_attn.kv_a_proj_with_mqa", "model.layers.91.self_attn.kv_b_proj", "model.layers.91.self_attn.o_proj", "model.layers.91.self_attn.q_a_proj", "model.layers.91.self_attn.q_b_proj", "model.layers.92.self_attn.fused_qkv_a_proj", "model.layers.92.self_attn.fused_qkv_a_proj_with_mqa", "model.layers.92.self_attn.fused_qkvg_proj", "model.layers.92.self_attn.g_proj", "model.layers.92.self_attn.in_proj_qkvgfab", "model.layers.92.self_attn.kv_a_proj_with_mqa", "model.layers.92.self_attn.kv_b_proj", "model.layers.92.self_attn.o_proj", "model.layers.92.self_attn.q_a_proj", "model.layers.92.self_attn.q_b_proj" ] } }, "ignore": [ "*block_sparse_moe.gate*", "*mlp.gate*", "*block_sparse_moe.routed_expert_up_proj*", "*mlp.routed_expert_up_proj*", "*block_sparse_moe.routed_expert_down_proj*", "*mlp.routed_expert_down_proj*", "*block_sparse_moe.routed_expert_norm*", "*mlp.routed_expert_norm*", "*block_sparse_moe.shared_experts*", "*mlp.shared_experts*", "*self_attn.q_conv1d*", "*self_attn.k_conv1d*", "*self_attn.v_conv1d*", "*self_attn.o_norm*", "*self_attn.q_a_layernorm*", "*self_attn.kv_a_layernorm*", "*_res_proj*", "*_res_norm*", "*layernorm*", "language_model.model.layers.0.mlp.*", "layers.0.mlp.*", "model.layers.0.mlp.*", "language_model.lm_head", "vision_tower*", "mm_projector*" ], "producer": { "name": "modelopt", "version": "0.45.0" }, "quant_algo": "MIXED_PRECISION", "quant_method": "modelopt_mixed", "quantized_layers": { "language_model.model.layers.0.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.0.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.1.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.1.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.1.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.10.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.10.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.10.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.11.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.11.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.11.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.12.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.12.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.12.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.13.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.13.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.13.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.14.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.14.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.14.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.15.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.15.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.15.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.16.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.16.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.16.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.17.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.17.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.17.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.18.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.18.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.18.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.19.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.19.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.19.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.2.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.2.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.2.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.20.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.20.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.20.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.21.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.21.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.21.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.22.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.22.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.22.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.23.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.23.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.23.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.24.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.24.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.24.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.25.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.25.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.25.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.26.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.26.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.26.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.27.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.27.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.27.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.28.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.28.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.28.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.29.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.29.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.29.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.3.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.3.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.3.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.30.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.30.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.30.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.31.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.31.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.31.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.32.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.32.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.32.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.33.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.33.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.33.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.34.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.34.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.34.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.35.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.35.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.35.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.36.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.36.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.36.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.37.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.37.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.37.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.38.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.38.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.38.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.39.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.39.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.39.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.4.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.4.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.4.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.40.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.40.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.40.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.41.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.41.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.41.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.42.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.42.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.42.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.43.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.43.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.43.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.44.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.44.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.44.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.45.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.45.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.45.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.46.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.46.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.46.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.47.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.47.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.47.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.48.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.48.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.48.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.49.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.49.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.49.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.5.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.5.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.5.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.50.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.50.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.50.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.51.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.51.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.51.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.52.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.52.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.52.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.53.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.53.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.53.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.54.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.54.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.54.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.55.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.55.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.55.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.56.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.56.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.56.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.57.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.57.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.57.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.58.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.58.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.58.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.59.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.59.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.59.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.6.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.6.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.6.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.60.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.60.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.60.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.61.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.61.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.61.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.62.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.62.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.62.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.63.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.63.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.63.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.64.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.64.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.64.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.65.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.65.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.65.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.66.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.66.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.66.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.67.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.67.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.67.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.68.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.68.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.68.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.69.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.69.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.69.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.7.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.7.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.7.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.70.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.70.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.70.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.71.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.71.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.71.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.72.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.72.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.72.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.73.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.73.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.73.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.74.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.74.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.74.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.75.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.75.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.75.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.76.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.76.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.76.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.77.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.77.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.77.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.78.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.78.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.78.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.79.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.79.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.79.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.8.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.8.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.8.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.80.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.80.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.80.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.81.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.81.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.81.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.82.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.82.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.82.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.83.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.83.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.83.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.84.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.84.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.84.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.85.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.85.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.85.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.86.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.86.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.86.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.87.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.87.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.87.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.88.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.88.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.88.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.89.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.89.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.89.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.9.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.9.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.9.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.90.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.90.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.90.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.91.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.91.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.91.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.92.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "language_model.model.layers.92.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "language_model.model.layers.92.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.0.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.1.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.1.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.1.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.10.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.10.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.10.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.11.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.11.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.11.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.11.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.12.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.12.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.12.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.13.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.13.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.13.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.14.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.14.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.14.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.15.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.15.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.15.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.15.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.16.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.16.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.16.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.17.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.17.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.17.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.18.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.18.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.18.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.19.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.19.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.19.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.19.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.2.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.2.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.2.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.20.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.20.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.20.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.21.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.21.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.21.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.22.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.22.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.22.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.23.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.23.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.23.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.23.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.24.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.24.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.24.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.25.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.25.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.25.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.26.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.26.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.26.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.27.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.27.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.27.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.27.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.28.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.28.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.28.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.29.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.29.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.29.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.3.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.3.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.3.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.3.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.30.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.30.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.30.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.31.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.31.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.31.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.31.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.32.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.32.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.32.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.33.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.33.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.33.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.34.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.34.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.34.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.35.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.35.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.35.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.35.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.36.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.36.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.36.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.37.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.37.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.37.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.38.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.38.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.38.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.39.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.39.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.39.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.39.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.4.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.4.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.4.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.40.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.40.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.40.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.41.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.41.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.41.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.42.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.42.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.42.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.43.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.43.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.43.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.43.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.44.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.44.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.44.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.45.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.45.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.45.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.46.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.46.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.46.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.47.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.47.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.47.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.47.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.48.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.48.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.48.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.49.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.49.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.49.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.5.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.5.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.5.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.50.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.50.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.50.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.51.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.51.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.51.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.51.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.52.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.52.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.52.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.53.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.53.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.53.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.54.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.54.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.54.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.55.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.55.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.55.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.55.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.56.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.56.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.56.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.57.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.57.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.57.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.58.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.58.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.58.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.59.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.59.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.59.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.59.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.6.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.6.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.6.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.60.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.60.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.60.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.61.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.61.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.61.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.62.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.62.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.62.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.63.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.63.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.63.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.63.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.64.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.64.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.64.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.65.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.65.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.65.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.66.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.66.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.66.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.67.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.67.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.67.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.67.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.68.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.68.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.68.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.69.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.69.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.69.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.7.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.7.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.7.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.7.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.70.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.70.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.70.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.71.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.71.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.71.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.71.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.72.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.72.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.72.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.73.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.73.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.73.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.74.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.74.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.74.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.75.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.75.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.75.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.75.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.76.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.76.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.76.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.77.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.77.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.77.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.78.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.78.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.78.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.79.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.79.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.79.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.79.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.8.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.8.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.8.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.80.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.80.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.80.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.81.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.81.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.81.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.82.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.82.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.82.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.83.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.83.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.83.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.83.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.84.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.84.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.84.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.85.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.85.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.85.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.86.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.86.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.86.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.87.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.87.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.87.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.87.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.88.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.88.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.88.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.89.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.89.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.89.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.9.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.9.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.9.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.90.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.90.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "layers.90.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "layers.91.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.91.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.91.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.91.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.92.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.92.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "layers.92.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "layers.92.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.0.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.1.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.1.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.1.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.10.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.10.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.10.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.11.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.11.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.11.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.12.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.12.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.12.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.13.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.13.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.13.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.14.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.14.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.14.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.15.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.15.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.15.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.16.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.16.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.16.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.17.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.17.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.17.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.18.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.18.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.18.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.19.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.19.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.19.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.2.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.2.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.2.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.20.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.20.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.20.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.21.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.21.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.21.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.22.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.22.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.22.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.23.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.23.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.23.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.24.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.24.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.24.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.25.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.25.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.25.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.26.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.26.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.26.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.27.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.27.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.27.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.28.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.28.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.28.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.29.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.29.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.29.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.3.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.3.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.3.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.30.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.30.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.30.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.31.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.31.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.31.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.32.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.32.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.32.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.33.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.33.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.33.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.34.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.34.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.34.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.35.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.35.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.35.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.36.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.36.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.36.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.37.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.37.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.37.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.38.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.38.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.38.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.39.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.39.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.39.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.4.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.4.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.4.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.40.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.40.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.40.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.41.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.41.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.41.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.42.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.42.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.42.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.43.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.43.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.43.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.44.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.44.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.44.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.45.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.45.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.45.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.46.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.46.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.46.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.47.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.47.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.47.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.48.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.48.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.48.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.49.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.49.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.49.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.5.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.5.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.5.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.50.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.50.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.50.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.51.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.51.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.51.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.52.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.52.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.52.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.53.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.53.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.53.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.54.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.54.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.54.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.55.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.55.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.55.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.56.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.56.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.56.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.57.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.57.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.57.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.58.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.58.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.58.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.59.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.59.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.59.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.6.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.6.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.6.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.60.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.60.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.60.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.61.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.61.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.61.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.62.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.62.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.62.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.63.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.63.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.63.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.64.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.64.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.64.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.65.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.65.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.65.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.66.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.66.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.66.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.67.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.67.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.67.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.68.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.68.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.68.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.69.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.69.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.69.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.7.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.7.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.7.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.70.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.70.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.70.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.71.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.71.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.71.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.72.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.72.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.72.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.73.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.73.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.73.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.74.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.74.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.74.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.75.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.75.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.75.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.76.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.76.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.76.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.77.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.77.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.77.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.78.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.78.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.78.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.79.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.79.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.79.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.8.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.8.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.8.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.80.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.80.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.80.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.81.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.81.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.81.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.82.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.82.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.82.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.83.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.83.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.83.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.84.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.84.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.84.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.85.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.85.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.85.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.86.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.86.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.86.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.87.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.87.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.87.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.88.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.88.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.88.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.89.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.89.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.89.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.9.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.9.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.9.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.90.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.90.self_attn.b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.f_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.f_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.k_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.q_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.90.self_attn.v_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.91.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.91.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.91.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.block_sparse_moe.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.92.mlp.experts": { "group_size": 16, "quant_algo": "NVFP4" }, "model.layers.92.self_attn.fused_qkv_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.fused_qkv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.fused_qkvg_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.g_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.in_proj_qkvgfab": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.kv_a_proj_with_mqa": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.kv_b_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.o_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.q_a_proj": { "quant_algo": "FP8_PB_WO" }, "model.layers.92.self_attn.q_b_proj": { "quant_algo": "FP8_PB_WO" } } }, "text_config": { "_name_or_path": "", "activation_situ_beta": 4.0, "activation_situ_linear_beta": 25.0, "add_cross_attention": false, "architectures": [ "KimiLinearForCausalLM" ], "attn_res_block_size": 12, "auto_map": { "AutoConfig": "configuration_kimi_k3.KimiLinearConfig", "AutoModel": "modeling_kimi_linear.KimiLinearModel", "AutoModelForCausalLM": "modeling_kimi_linear.KimiLinearForCausalLM" }, "bad_words_ids": null, "begin_suppress_tokens": null, "bos_token_id": 163584, "chunk_size_feed_forward": 0, "cross_attention_hidden_size": null, "decoder_start_token_id": null, "diversity_penalty": 0.0, "do_sample": false, "dtype": "bfloat16", "early_stopping": false, "encoder_no_repeat_ngram_size": 0, "eos_token_id": 163586, "exponential_decay_length_penalty": null, "finetuning_task": null, "first_k_dense_replace": 1, "forced_bos_token_id": null, "forced_eos_token_id": null, "hidden_act": "situ", "hidden_size": 7168, "id2label": { "0": "LABEL_0", "1": "LABEL_1" }, "initializer_range": 0.02, "intermediate_size": 33792, "is_decoder": false, "is_encoder_decoder": false, "kv_lora_rank": 512, "label2id": { "LABEL_0": 0, "LABEL_1": 1 }, "latent_moe_use_norm": true, "length_penalty": 1.0, "linear_attn_config": { "full_attn_layers": [ 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 64, 68, 72, 76, 80, 84, 88, 92, 93 ], "gate_lower_bound": -5.0, "head_dim": 128, "kda_layers": [ 1, 2, 3, 5, 6, 7, 9, 10, 11, 13, 14, 15, 17, 18, 19, 21, 22, 23, 25, 26, 27, 29, 30, 31, 33, 34, 35, 37, 38, 39, 41, 42, 43, 45, 46, 47, 49, 50, 51, 53, 54, 55, 57, 58, 59, 61, 62, 63, 65, 66, 67, 69, 70, 71, 73, 74, 75, 77, 78, 79, 81, 82, 83, 85, 86, 87, 89, 90, 91 ], "num_heads": 96, "short_conv_kernel_size": 4, "use_full_rank_gate": true }, "max_length": 20, "max_position_embeddings": 1048576, "min_length": 0, "mla_use_nope": true, "mla_use_output_gate": true, "model_type": "kimi_linear", "moe_intermediate_size": 3072, "moe_layer_freq": 1, "moe_renormalize": true, "moe_router_activation_func": "sigmoid", "no_repeat_ngram_size": 0, "num_attention_heads": 96, "num_beam_groups": 1, "num_beams": 1, "num_expert_group": 1, "num_experts": 896, "num_experts_per_token": 16, "num_hidden_layers": 93, "num_key_value_heads": 96, "num_nextn_predict_layers": 0, "num_return_sequences": 1, "num_shared_experts": 2, "output_attentions": false, "output_hidden_states": false, "output_scores": false, "pad_token_id": 163839, "prefix": null, "problem_type": null, "pruned_heads": {}, "q_lora_rank": 1536, "qk_nope_head_dim": 128, "qk_rope_head_dim": 64, "remove_invalid_values": false, "repetition_penalty": 1.0, "return_dict": true, "return_dict_in_generate": false, "rms_norm_eps": 1e-05, "routed_expert_hidden_size": 3584, "routed_scaling_factor": 1.0, "sep_token_id": null, "suppress_tokens": null, "task_specific_params": null, "temperature": 1.0, "tf_legacy_loss": false, "tie_encoder_decoder": false, "tie_word_embeddings": false, "tokenizer_class": null, "top_k": 50, "top_p": 1.0, "topk_group": 1, "topk_method": "noaux_tc", "torchscript": false, "transformers_version": "4.56.2", "typical_p": 1.0, "use_bfloat16": false, "use_cache": true, "use_grouped_topk": true, "v_head_dim": 128, "vocab_size": 163840 }, "tie_word_embeddings": false, "vision_config": { "_attn_implementation": "flash_attention_2", "activation_func": "gelu_pytorch_tanh", "attn_bias": false, "init_pos_emb_height": 64, "init_pos_emb_time": 4, "init_pos_emb_width": 64, "linear_bias": false, "merge_kernel_size": [ 2, 2 ], "merge_type": "sd2_tpool", "mlp_type": "mlp2", "mm_hidden_size": 1024, "mm_projector_type": "patchmergerv2", "norm_type": "rmsnorm", "patch_embed_proj_bias": false, "patch_size": 14, "pos_emb_interpolation_mode": "bilinear", "pos_emb_type": "divided_fixed", "projector_hidden_act": "gelu", "projector_ln_eps": 1e-05, "qkv_hidden_size": 1536, "text_hidden_size": 7168, "vt_hidden_size": 1024, "vt_intermediate_size": 4096, "vt_num_attention_heads": 12, "vt_num_hidden_layers": 27 } }