diff --git a/deepseek-r1-1.5b-ternary/config.json b/deepseek-r1-1.5b-ternary/config.json new file mode 100644 index 0000000000000000000000000000000000000000..d037a166be9fbde3fdb6fb7af74fb57426acb345 --- /dev/null +++ b/deepseek-r1-1.5b-ternary/config.json @@ -0,0 +1,12 @@ +{ + "hidden_size": 1536, + "intermediate_size": 8960, + "num_attention_heads": 12, + "num_key_value_heads": 2, + "num_hidden_layers": 28, + "vocab_size": 151936, + "head_dim": 128, + "rope_theta": 1000000.0, + "rms_norm_eps": 1e-06, + "alpha": 0.7 +} \ No newline at end of file diff --git a/deepseek-r1-1.5b-ternary/model_layers_0_self_attn_q_proj_weight.scales b/deepseek-r1-1.5b-ternary/model_layers_0_self_attn_q_proj_weight.scales new file mode 100644 index 0000000000000000000000000000000000000000..4b8a9beffdb022bb3afc844be31eac52e0e0f0e1 Binary files /dev/null and b/deepseek-r1-1.5b-ternary/model_layers_0_self_attn_q_proj_weight.scales differ diff --git a/deepseek-r1-1.5b-ternary/model_layers_0_self_attn_v_proj_weight.neg b/deepseek-r1-1.5b-ternary/model_layers_0_self_attn_v_proj_weight.neg new file mode 100644 index 0000000000000000000000000000000000000000..cc6d9f0c437993026e57d81934f949dda7f34e03 Binary files /dev/null and b/deepseek-r1-1.5b-ternary/model_layers_0_self_attn_v_proj_weight.neg differ diff --git a/deepseek-r1-1.5b-ternary/model_layers_10_self_attn_k_proj_weight.scales b/deepseek-r1-1.5b-ternary/model_layers_10_self_attn_k_proj_weight.scales new file mode 100644 index 0000000000000000000000000000000000000000..266a87adcc65a24fcf8e5a867f17f4e1b614b595 Binary files /dev/null and b/deepseek-r1-1.5b-ternary/model_layers_10_self_attn_k_proj_weight.scales differ diff --git a/deepseek-r1-1.5b-ternary/model_layers_11_self_attn_k_proj_bias.fp16 b/deepseek-r1-1.5b-ternary/model_layers_11_self_attn_k_proj_bias.fp16 new file mode 100644 index 0000000000000000000000000000000000000000..062db56c99f1e0f7bf63d93fa5f28fe511ff6724 Binary files /dev/null and b/deepseek-r1-1.5b-ternary/model_layers_11_self_attn_k_proj_bias.fp16 differ diff --git a/deepseek-r1-1.5b-ternary/model_layers_12_self_attn_k_proj_weight.scales b/deepseek-r1-1.5b-ternary/model_layers_12_self_attn_k_proj_weight.scales new file mode 100644 index 0000000000000000000000000000000000000000..43a1914b38aee6fc766bbd2ff2af3910b63a749a Binary files /dev/null and b/deepseek-r1-1.5b-ternary/model_layers_12_self_attn_k_proj_weight.scales differ diff --git a/deepseek-r1-1.5b-ternary/model_layers_12_self_attn_o_proj_weight.scales b/deepseek-r1-1.5b-ternary/model_layers_12_self_attn_o_proj_weight.scales new file mode 100644 index 0000000000000000000000000000000000000000..af059b6c0702bef1c63c7e44ae768e3f63dd2101 Binary files /dev/null and b/deepseek-r1-1.5b-ternary/model_layers_12_self_attn_o_proj_weight.scales differ diff --git a/deepseek-r1-1.5b-ternary/model_layers_13_input_layernorm_weight.fp16 b/deepseek-r1-1.5b-ternary/model_layers_13_input_layernorm_weight.fp16 new file mode 100644 index 0000000000000000000000000000000000000000..40b686b83607f6f85048ff11f9cbd63cf1b152cc Binary files /dev/null and b/deepseek-r1-1.5b-ternary/model_layers_13_input_layernorm_weight.fp16 differ diff --git a/deepseek-r1-1.5b-ternary/model_layers_13_self_attn_k_proj_weight.scales b/deepseek-r1-1.5b-ternary/model_layers_13_self_attn_k_proj_weight.scales new file mode 100644 index 0000000000000000000000000000000000000000..40e1cbc0d9908cd18d078ba4e97c21a4a8775bff --- /dev/null +++ b/deepseek-r1-1.5b-ternary/model_layers_13_self_attn_k_proj_weight.scales @@ -0,0 +1,8 @@ +s<<a=C=%J=6[= +=3_=L6X=`=W="?=i=Ni=i=_h=M^=?p=v=(]=n=7k=Z7y=pq=yp=`=]s=P= +]=(wd=^k=]=5Vb=ji=߅Q=l=4L^=H=[ +M==<=aW=9Y=ѠN=}g=Y,I=f='q=J=\=!h=H= ==g=݉=}=J===ɋ=l|=Z9==|={ ==P<ł=2=Dx=p=V>=h=K<+=u=de==.?= ==~==;1==r==h=