Upload model
Browse files- config.json +1 -2
- modelling_sparrow.py +39 -2
config.json
CHANGED
|
@@ -4,8 +4,7 @@
|
|
| 4 |
],
|
| 5 |
"attention_bias": false,
|
| 6 |
"auto_map": {
|
| 7 |
-
"AutoConfig": "
|
| 8 |
-
"AutoModelForCausalLM": "modelling_sparrow.SparrowModel"
|
| 9 |
},
|
| 10 |
"dropout": 0.0,
|
| 11 |
"flash_attn": true,
|
|
|
|
| 4 |
],
|
| 5 |
"attention_bias": false,
|
| 6 |
"auto_map": {
|
| 7 |
+
"AutoConfig": "modelling_sparrow.SparrowConfig"
|
|
|
|
| 8 |
},
|
| 9 |
"dropout": 0.0,
|
| 10 |
"flash_attn": true,
|
modelling_sparrow.py
CHANGED
|
@@ -1,12 +1,49 @@
|
|
| 1 |
import math
|
|
|
|
|
|
|
| 2 |
import torch
|
| 3 |
import torch.nn as nn
|
| 4 |
import torch.nn.functional as F
|
| 5 |
|
| 6 |
-
from transformers import PreTrainedModel
|
| 7 |
from transformers.modeling_outputs import CausalLMOutputWithPast
|
| 8 |
|
| 9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
## RoPE - from https://arxiv.org/pdf/2104.09864v5
|
| 12 |
def rotate_half(x):
|
|
|
|
| 1 |
import math
|
| 2 |
+
from typing import Optional
|
| 3 |
+
|
| 4 |
import torch
|
| 5 |
import torch.nn as nn
|
| 6 |
import torch.nn.functional as F
|
| 7 |
|
| 8 |
+
from transformers import PreTrainedModel, PretrainedConfig
|
| 9 |
from transformers.modeling_outputs import CausalLMOutputWithPast
|
| 10 |
|
| 11 |
+
class SparrowConfig(PretrainedConfig):
|
| 12 |
+
model_type = "sparrow"
|
| 13 |
+
|
| 14 |
+
def __init__(
|
| 15 |
+
self,
|
| 16 |
+
hidden_size: int = 512,
|
| 17 |
+
num_hidden_layers: int = 8,
|
| 18 |
+
num_attention_heads: int = 16,
|
| 19 |
+
num_key_value_heads: Optional[int] = None,
|
| 20 |
+
max_seq_len: int = 512,
|
| 21 |
+
attention_bias: bool = False,
|
| 22 |
+
flash_attn: bool = True,
|
| 23 |
+
vocab_size: int = 32000,
|
| 24 |
+
hidden_dim: Optional[int] = None,
|
| 25 |
+
intermediate_dim: int = 2048,
|
| 26 |
+
norm_eps: float = 1e-5,
|
| 27 |
+
mlp_bias: bool = False,
|
| 28 |
+
dropout: float = 0.0,
|
| 29 |
+
**kwargs,
|
| 30 |
+
):
|
| 31 |
+
super().__init__(**kwargs)
|
| 32 |
+
# attention args
|
| 33 |
+
self.hidden_size = hidden_size
|
| 34 |
+
self.num_hidden_layers = num_hidden_layers
|
| 35 |
+
self.num_attention_heads = num_attention_heads
|
| 36 |
+
self.num_key_value_heads = num_key_value_heads if num_key_value_heads is not None else num_attention_heads
|
| 37 |
+
self.max_seq_len = max_seq_len
|
| 38 |
+
self.attention_bias = attention_bias
|
| 39 |
+
self.flash_attn = flash_attn
|
| 40 |
+
# mlp args
|
| 41 |
+
self.vocab_size = vocab_size
|
| 42 |
+
self.hidden_dim = hidden_dim if hidden_dim is not None else hidden_size
|
| 43 |
+
self.intermediate_dim = intermediate_dim
|
| 44 |
+
self.norm_eps = norm_eps
|
| 45 |
+
self.mlp_bias = mlp_bias
|
| 46 |
+
self.dropout = dropout
|
| 47 |
|
| 48 |
## RoPE - from https://arxiv.org/pdf/2104.09864v5
|
| 49 |
def rotate_half(x):
|