27 lines
2.5 KiB
JSON
27 lines
2.5 KiB
JSON
|
{
|
||
|
"static_weights": {
|
||
|
"model.embed_tokens.weight": {"mtj": {"module": "embedding_shard/~/linear", "param": "w", "axis": 1, "transforms": ["transpose", "vocab_pad"]}},
|
||
|
"model.layer_norm.weight": {"mtj": {"module": "projection_shard/~/layer_norm", "param": "scale"}},
|
||
|
"model.layer_norm.bias": {"mtj": {"module": "projection_shard/~/layer_norm", "param": "offset"}},
|
||
|
"lm_head.weight": {"mtj": {"module": "projection_shard/~/linear", "param": "w", "axis": 2, "transforms": ["vocab_pad"]}}
|
||
|
},
|
||
|
"layer_weights": {
|
||
|
"model.layers.{layer}.self_attn.q_proj.weight": {"mtj": {"module": "layer_{layer}/~/linear", "param": "w", "axis": 2}},
|
||
|
"model.layers.{layer}.self_attn.q_proj.bias": {"mtj": {"module": "layer_{layer}/~/linear", "param": "b", "axis": 1}},
|
||
|
"model.layers.{layer}.self_attn.v_proj.weight": {"mtj": {"module": "layer_{layer}/~/linear_1", "param": "w", "axis": 2}},
|
||
|
"model.layers.{layer}.self_attn.v_proj.bias": {"mtj": {"module": "layer_{layer}/~/linear_1", "param": "b", "axis": 1}},
|
||
|
"model.layers.{layer}.self_attn.k_proj.weight": {"mtj": {"module": "layer_{layer}/~/linear_2", "param": "w", "axis": 2}},
|
||
|
"model.layers.{layer}.self_attn.k_proj.bias": {"mtj": {"module": "layer_{layer}/~/linear_2", "param": "b", "axis": 1}},
|
||
|
"model.layers.{layer}.self_attn.out_proj.weight": {"mtj": {"module": "layer_{layer}/~/linear_3", "param": "w", "axis": 1}},
|
||
|
"model.layers.{layer}.self_attn.out_proj.bias": {"mtj": {"module": "layer_{layer}/~/linear_3", "param": "b", "transforms": ["divide_by_shards"]}},
|
||
|
"model.layers.{layer}.fc1.weight": {"mtj": {"module": "layer_{layer}/~/linear_4", "param": "w", "axis": 2}},
|
||
|
"model.layers.{layer}.fc1.bias": {"mtj": {"module": "layer_{layer}/~/linear_4", "param": "b", "axis": 1}},
|
||
|
"model.layers.{layer}.fc2.weight": {"mtj": {"module": "layer_{layer}/~/linear_5", "param": "w", "axis": 1}},
|
||
|
"model.layers.{layer}.fc2.bias": {"mtj": {"module": "layer_{layer}/~/linear_5", "param": "b", "transforms": ["divide_by_shards"]}},
|
||
|
"model.layers.{layer}.self_attn_layer_norm.weight": {"mtj": {"module": "layer_{layer}/~/layer_norm", "param": "scale"}},
|
||
|
"model.layers.{layer}.self_attn_layer_norm.bias": {"mtj": {"module": "layer_{layer}/~/layer_norm", "param": "offset"}},
|
||
|
"model.layers.{layer}.final_layer_norm.weight": {"mtj": {"module": "layer_{layer}/~/layer_norm_1", "param": "scale"}},
|
||
|
"model.layers.{layer}.final_layer_norm.bias": {"mtj": {"module": "layer_{layer}/~/layer_norm_1", "param": "offset"}}
|
||
|
}
|
||
|
}
|