RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w8a8

1{2  "_name_or_path": "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",3  "architectures": [4    "LlamaForCausalLM"5  ],6  "attention_bias": false,7  "attention_dropout": 0.0,8  "bos_token_id": 128000,9  "eos_token_id": [10    128001,11    128008,12    12800913  ],14  "head_dim": 128,15  "hidden_act": "silu",16  "hidden_size": 8192,17  "initializer_range": 0.02,18  "intermediate_size": 28672,19  "max_position_embeddings": 131072,20  "mlp_bias": false,21  "model_type": "llama",22  "num_attention_heads": 64,23  "num_hidden_layers": 80,24  "num_key_value_heads": 8,25  "pretraining_tp": 1,26  "quantization_config": {27    "config_groups": {28      "group_0": {29        "input_activations": {30          "actorder": null,31          "block_structure": null,32          "dynamic": true,33          "group_size": null,34          "num_bits": 8,35          "observer": null,36          "observer_kwargs": {},37          "strategy": "token",38          "symmetric": true,39          "type": "int"40        },41        "output_activations": null,42        "targets": [43          "Linear"44        ],45        "weights": {46          "actorder": null,47          "block_structure": null,48          "dynamic": false,49          "group_size": null,50          "num_bits": 8,51          "observer": "mse",52          "observer_kwargs": {},53          "strategy": "channel",54          "symmetric": true,55          "type": "int"56        }57      }58    },59    "format": "int-quantized",60    "global_compression_ratio": 1.5343121209820911,61    "ignore": [62      "lm_head"63    ],64    "kv_cache_scheme": null,65    "quant_method": "compressed-tensors",66    "quantization_status": "compressed"67  },68  "rms_norm_eps": 1e-05,69  "rope_scaling": {70    "factor": 8.0,71    "high_freq_factor": 4.0,72    "low_freq_factor": 1.0,73    "original_max_position_embeddings": 8192,74    "rope_type": "llama3"75  },76  "rope_theta": 500000.0,77  "tie_word_embeddings": false,78  "torch_dtype": "bfloat16",79  "transformers_version": "4.48.0",80  "use_cache": true,81  "vocab_size": 12825682}