Upload folder using huggingface_hub

chat_template.jinja

@@ -0,0 +1,21 @@
+{%- set found_item = false -%}
+{%- for message in messages -%}
+    {%- if message['role'] == 'system' -%}
+        {%- set found_item = true -%}
+    {%- endif -%}
+{%- endfor -%}
+{%- if not found_item -%}
+{{'You are an AI programming assistant, utilizing the Deepseek Coder model, developed by Deepseek Company, and you only answer questions related to computer science. For politically sensitive questions, security and privacy issues, and other non-computer science questions, you will refuse to answer.\n'}}
+{%- endif %}
+{%- for message in messages %}
+    {%- if message['role'] == 'system' %}
+{{ message['content'] }}
+    {%- else %}
+        {%- if message['role'] == 'user' %}
+{{'### Instruction:\n' + message['content'] + '\n'}}
+        {%- else %}
+{{'### Response:\n' + message['content'] + '\n<|EOT|>\n'}}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{{'### Response:\n'}}

config.json

@@ -0,0 +1,29 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 100000,
+  "eos_token_id": 100001,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 11008,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 30,
+  "num_key_value_heads": 32,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.55.0",
+  "use_cache": true,
+  "vocab_size": 102400
+}

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6d794a34fe60fcf2c35f97a8a3dd9692798c52c5d8971be1284fac2d2de6f9d
+size 3978116712

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31280eb0444b8005810b03ed54678783f2a6c448e5e2f38c6fe0f839f99283f9
+size 838860896

model.safetensors.index.json.meta.json

@@ -0,0 +1 @@
+{"model.layers.0.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.0.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.0.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.0.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.0.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.0.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.0.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.1.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.1.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.1.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.1.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.1.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.1.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.1.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.2.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.2.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.2.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.2.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.2.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.2.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.2.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.3.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.3.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.3.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.3.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.3.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.3.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.3.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.4.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.4.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.4.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.4.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.4.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.4.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.4.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.5.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.5.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.5.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.5.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.5.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.5.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.5.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.6.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.6.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.6.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.6.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.6.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.6.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.6.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.7.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.7.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.7.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.7.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.7.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.7.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.7.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.8.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.8.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.8.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.8.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.8.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.8.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.8.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.9.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.9.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.9.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.9.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.9.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.9.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.9.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.10.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.10.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.10.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.10.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.10.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.10.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.10.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.11.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.11.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.11.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.11.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.11.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.11.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.11.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.12.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.12.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.12.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.12.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.12.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.12.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.12.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.13.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.13.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.13.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.13.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.13.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.13.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.13.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.14.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.14.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.14.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.14.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.14.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.14.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.14.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.15.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.15.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.15.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.15.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.15.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.15.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.15.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.16.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.16.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.16.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.16.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.16.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.16.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.16.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.17.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.17.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.17.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.17.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.17.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.17.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.17.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.18.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.18.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.18.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.18.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.18.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.18.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.18.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.19.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.19.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.19.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.19.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.19.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.19.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.19.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.20.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.20.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.20.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.20.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.20.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.20.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.20.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.21.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.21.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.21.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.21.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.21.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.21.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.21.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.22.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.22.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.22.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.22.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.22.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.22.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.22.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.23.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.23.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.23.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.23.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.23.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.23.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.23.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.24.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.24.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.24.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.24.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.24.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.24.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.24.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.25.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.25.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.25.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.25.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.25.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.25.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.25.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.26.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.26.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.26.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.26.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.26.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.26.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.26.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.27.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.27.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.27.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.27.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.27.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.27.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.27.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.28.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.28.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.28.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.28.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.28.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.28.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.28.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.29.self_attn.q_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.29.self_attn.k_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.29.self_attn.v_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.29.self_attn.o_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 4096], "scale.shape": [131072, 1], "awq_scale": null, "zero.shape": [131072, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.29.mlp.gate_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.29.mlp.up_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [11008, 4096], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}, "model.layers.29.mlp.down_proj": {"meta": {"nbits": 4, "group_size": 128, "shape": [4096, 11008], "scale.shape": [352256, 1], "awq_scale": null, "zero.shape": [352256, 1], "axis": 1, "packing": "4bit_u8", "method": "sinq_quantAux", "compute_dtype": "torch.float16", "unpack_view_dtype": "torch.uint8", "view_as_float": false}}}

special_tokens_map.json

@@ -0,0 +1,17 @@
+{
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<｜end▁of▁sentence｜>"
+}

tokenizer_config.json

@@ -0,0 +1,34 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "100000": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100001": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "extra_special_tokens": {},
+  "legacy": true,
+  "model_max_length": 16384,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizerFast",
+  "unk_token": null,
+  "use_default_system_prompt": false
+}