mayank-mishra
commited on
Commit
•
afc3169
1
Parent(s):
2263de3
llama
Browse files- config.json +18 -30
- configuration_granite.py +0 -98
- generation_config.json +3 -4
- model-00001-of-00002.safetensors +2 -2
- model-00002-of-00002.safetensors +2 -2
- model.safetensors.index.json +514 -322
- modeling_granite.py +0 -1376
config.json
CHANGED
@@ -1,40 +1,28 @@
|
|
1 |
{
|
2 |
-
"activation_function": "swiglu",
|
3 |
-
"add_bias": true,
|
4 |
-
"apply_residual_connection_post_layernorm": false,
|
5 |
"architectures": [
|
6 |
-
"
|
7 |
],
|
8 |
-
"
|
9 |
-
"
|
10 |
-
"
|
11 |
-
"
|
12 |
-
"
|
13 |
-
|
14 |
-
"AutoModel": "modeling_granite.GraniteModel",
|
15 |
-
"AutoModelForCausalLM": "modeling_granite.GraniteForCausalLM"
|
16 |
-
},
|
17 |
-
"bos_token_id": 0,
|
18 |
-
"embd_pdrop": 0.1,
|
19 |
-
"eos_token_id": 0,
|
20 |
"initializer_range": 0.02,
|
21 |
-
"
|
22 |
-
"
|
23 |
-
"
|
24 |
-
"
|
25 |
-
"
|
26 |
-
"
|
27 |
-
"n_positions": 2048,
|
28 |
-
"normalization_function": "rmsnorm",
|
29 |
"num_key_value_heads": 32,
|
30 |
-
"
|
31 |
-
"
|
32 |
-
"
|
33 |
"rope_theta": 10000,
|
34 |
-
"
|
35 |
-
"scale_attn_weights": true,
|
36 |
"torch_dtype": "bfloat16",
|
37 |
-
"transformers_version": "4.
|
38 |
"use_cache": true,
|
39 |
"vocab_size": 49152
|
40 |
}
|
|
|
1 |
{
|
|
|
|
|
|
|
2 |
"architectures": [
|
3 |
+
"LlamaForCausalLM"
|
4 |
],
|
5 |
+
"attention_bias": true,
|
6 |
+
"attention_dropout": 0.1,
|
7 |
+
"bos_token_id": 1,
|
8 |
+
"eos_token_id": 2,
|
9 |
+
"hidden_act": "silu",
|
10 |
+
"hidden_size": 2560,
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
"initializer_range": 0.02,
|
12 |
+
"intermediate_size": 10240,
|
13 |
+
"max_position_embeddings": 2048,
|
14 |
+
"mlp_bias": true,
|
15 |
+
"model_type": "llama",
|
16 |
+
"num_attention_heads": 32,
|
17 |
+
"num_hidden_layers": 32,
|
|
|
|
|
18 |
"num_key_value_heads": 32,
|
19 |
+
"pretraining_tp": 1,
|
20 |
+
"rms_norm_eps": 1e-05,
|
21 |
+
"rope_scaling": null,
|
22 |
"rope_theta": 10000,
|
23 |
+
"tie_word_embeddings": true,
|
|
|
24 |
"torch_dtype": "bfloat16",
|
25 |
+
"transformers_version": "4.41.0.dev0",
|
26 |
"use_cache": true,
|
27 |
"vocab_size": 49152
|
28 |
}
|
configuration_granite.py
DELETED
@@ -1,98 +0,0 @@
|
|
1 |
-
from transformers import PretrainedConfig
|
2 |
-
|
3 |
-
|
4 |
-
class GraniteConfig(PretrainedConfig):
|
5 |
-
model_type = "granite"
|
6 |
-
|
7 |
-
keys_to_ignore_at_inference = ["past_key_values"]
|
8 |
-
attribute_map = {
|
9 |
-
"hidden_size": "n_embd",
|
10 |
-
"max_position_embeddings": "n_positions",
|
11 |
-
"num_attention_heads": "n_head",
|
12 |
-
"num_hidden_layers": "n_layer",
|
13 |
-
}
|
14 |
-
|
15 |
-
def __init__(
|
16 |
-
self,
|
17 |
-
vocab_size: int = 50257,
|
18 |
-
n_positions: int = 1024,
|
19 |
-
n_embd: int = 768,
|
20 |
-
n_layer: int = 12,
|
21 |
-
n_head: int = 12,
|
22 |
-
num_key_value_heads: int = None,
|
23 |
-
n_inner: int = None,
|
24 |
-
activation_function: str = "gelu_pytorch_tanh",
|
25 |
-
attention_head_type: str = "mqa",
|
26 |
-
resid_pdrop: float = 0.1,
|
27 |
-
embd_pdrop: float = 0.1,
|
28 |
-
attn_pdrop: float = 0.1,
|
29 |
-
normalization_function: str = "layernorm",
|
30 |
-
layer_norm_epsilon: float = 1e-5,
|
31 |
-
initializer_range: float = 0.02,
|
32 |
-
scale_attn_weights: bool = True,
|
33 |
-
attention_multiplier: float = None,
|
34 |
-
use_cache: bool = True,
|
35 |
-
bos_token_id: int = 50256,
|
36 |
-
eos_token_id: int = 50256,
|
37 |
-
pad_token_id: int = 50256,
|
38 |
-
attention_softmax_in_fp32: bool = True,
|
39 |
-
scale_attention_softmax_in_fp32: bool = True,
|
40 |
-
add_bias: bool = True,
|
41 |
-
position_embedding_type: str = "learned_absolute",
|
42 |
-
rope_theta: int = 10000,
|
43 |
-
**kwargs,
|
44 |
-
) -> None:
|
45 |
-
self.vocab_size = vocab_size
|
46 |
-
self.n_positions = n_positions
|
47 |
-
self.n_embd = n_embd
|
48 |
-
self.n_layer = n_layer
|
49 |
-
self.n_head = n_head
|
50 |
-
self.num_key_value_heads = num_key_value_heads
|
51 |
-
self.n_inner = 4 * n_embd if n_inner is None else n_inner
|
52 |
-
self.activation_function = activation_function
|
53 |
-
self.attention_head_type = attention_head_type
|
54 |
-
self.resid_pdrop = resid_pdrop
|
55 |
-
self.embd_pdrop = embd_pdrop
|
56 |
-
self.attn_pdrop = attn_pdrop
|
57 |
-
self.normalization_function = normalization_function
|
58 |
-
self.layer_norm_epsilon = layer_norm_epsilon
|
59 |
-
self.initializer_range = initializer_range
|
60 |
-
self.scale_attn_weights = scale_attn_weights
|
61 |
-
self.attention_multiplier = attention_multiplier
|
62 |
-
self.use_cache = use_cache
|
63 |
-
self.attention_softmax_in_fp32 = attention_softmax_in_fp32
|
64 |
-
self.scale_attention_softmax_in_fp32 = scale_attention_softmax_in_fp32
|
65 |
-
self.position_embedding_type = position_embedding_type
|
66 |
-
self.add_bias = add_bias
|
67 |
-
self.rope_theta = rope_theta
|
68 |
-
|
69 |
-
if self.attention_multiplier is not None:
|
70 |
-
assert self.scale_attn_weights
|
71 |
-
|
72 |
-
# for compatibility with some features
|
73 |
-
self.multi_query = attention_head_type == "mqa"
|
74 |
-
|
75 |
-
if attention_head_type == "mha":
|
76 |
-
if self.num_key_value_heads is None:
|
77 |
-
self.num_key_value_heads = self.n_head
|
78 |
-
|
79 |
-
assert (
|
80 |
-
self.n_head == self.num_key_value_heads
|
81 |
-
), "MultiHeadAttention should have same number of heads for query, keys and values"
|
82 |
-
elif attention_head_type == "mqa":
|
83 |
-
if self.num_key_value_heads is None:
|
84 |
-
self.num_key_value_heads = 1
|
85 |
-
|
86 |
-
assert self.num_key_value_heads == 1, "MultiQueryAttention should have 1 head for keys and values"
|
87 |
-
elif attention_head_type == "gqa":
|
88 |
-
assert (
|
89 |
-
self.num_key_value_heads is not None
|
90 |
-
), "`num_key_value_heads` needs to be specified with GroupedQueryAttention"
|
91 |
-
|
92 |
-
assert (
|
93 |
-
self.n_head % self.num_key_value_heads == 0
|
94 |
-
), "GroupedQueryAttention should have more than 1 head for keys and values"
|
95 |
-
else:
|
96 |
-
raise ValueError(f"unexpected attention_head_type ({attention_head_type})")
|
97 |
-
|
98 |
-
super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, pad_token_id=pad_token_id, **kwargs)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
generation_config.json
CHANGED
@@ -1,7 +1,6 @@
|
|
1 |
{
|
2 |
"_from_model_config": true,
|
3 |
-
"bos_token_id":
|
4 |
-
"eos_token_id":
|
5 |
-
"
|
6 |
-
"transformers_version": "4.38.1"
|
7 |
}
|
|
|
1 |
{
|
2 |
"_from_model_config": true,
|
3 |
+
"bos_token_id": 1,
|
4 |
+
"eos_token_id": 2,
|
5 |
+
"transformers_version": "4.41.0.dev0"
|
|
|
6 |
}
|
model-00001-of-00002.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:7127432f1ffff5b25875ce4e287c9dfa458655ad1b6183a2d274c061bf9e4b1b
|
3 |
+
size 4972021832
|
model-00002-of-00002.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:344a03bb26fe17bacac63b616c45397758aeb6dfb0f36b4495139dc4530b885b
|
3 |
+
size 1993043568
|
model.safetensors.index.json
CHANGED
@@ -3,327 +3,519 @@
|
|
3 |
"total_size": 6965007360
|
4 |
},
|
5 |
"weight_map": {
|
6 |
-
"
|
7 |
-
"
|
8 |
-
"
|
9 |
-
"
|
10 |
-
"
|
11 |
-
"
|
12 |
-
"
|
13 |
-
"
|
14 |
-
"
|
15 |
-
"
|
16 |
-
"
|
17 |
-
"
|
18 |
-
"
|
19 |
-
"
|
20 |
-
"
|
21 |
-
"
|
22 |
-
"
|
23 |
-
"
|
24 |
-
"
|
25 |
-
"
|
26 |
-
"
|
27 |
-
"
|
28 |
-
"
|
29 |
-
"
|
30 |
-
"
|
31 |
-
"
|
32 |
-
"
|
33 |
-
"
|
34 |
-
"
|
35 |
-
"
|
36 |
-
"
|
37 |
-
"
|
38 |
-
"
|
39 |
-
"
|
40 |
-
"
|
41 |
-
"
|
42 |
-
"
|
43 |
-
"
|
44 |
-
"
|
45 |
-
"
|
46 |
-
"
|
47 |
-
"
|
48 |
-
"
|
49 |
-
"
|
50 |
-
"
|
51 |
-
"
|
52 |
-
"
|
53 |
-
"
|
54 |
-
"
|
55 |
-
"
|
56 |
-
"
|
57 |
-
"
|
58 |
-
"
|
59 |
-
"
|
60 |
-
"
|
61 |
-
"
|
62 |
-
"
|
63 |
-
"
|
64 |
-
"
|
65 |
-
"
|
66 |
-
"
|
67 |
-
"
|
68 |
-
"
|
69 |
-
"
|
70 |
-
"
|
71 |
-
"
|
72 |
-
"
|
73 |
-
"
|
74 |
-
"
|
75 |
-
"
|
76 |
-
"
|
77 |
-
"
|
78 |
-
"
|
79 |
-
"
|
80 |
-
"
|
81 |
-
"
|
82 |
-
"
|
83 |
-
"
|
84 |
-
"
|
85 |
-
"
|
86 |
-
"
|
87 |
-
"
|
88 |
-
"
|
89 |
-
"
|
90 |
-
"
|
91 |
-
"
|
92 |
-
"
|
93 |
-
"
|
94 |
-
"
|
95 |
-
"
|
96 |
-
"
|
97 |
-
"
|
98 |
-
"
|
99 |
-
"
|
100 |
-
"
|
101 |
-
"
|
102 |
-
"
|
103 |
-
"
|
104 |
-
"
|
105 |
-
"
|
106 |
-
"
|
107 |
-
"
|
108 |
-
"
|
109 |
-
"
|
110 |
-
"
|
111 |
-
"
|
112 |
-
"
|
113 |
-
"
|
114 |
-
"
|
115 |
-
"
|
116 |
-
"
|
117 |
-
"
|
118 |
-
"
|
119 |
-
"
|
120 |
-
"
|
121 |
-
"
|
122 |
-
"
|
123 |
-
"
|
124 |
-
"
|
125 |
-
"
|
126 |
-
"
|
127 |
-
"
|
128 |
-
"
|
129 |
-
"
|
130 |
-
"
|
131 |
-
"
|
132 |
-
"
|
133 |
-
"
|
134 |
-
"
|
135 |
-
"
|
136 |
-
"
|
137 |
-
"
|
138 |
-
"
|
139 |
-
"
|
140 |
-
"
|
141 |
-
"
|
142 |
-
"
|
143 |
-
"
|
144 |
-
"
|
145 |
-
"
|
146 |
-
"
|
147 |
-
"
|
148 |
-
"
|
149 |
-
"
|
150 |
-
"
|
151 |
-
"
|
152 |
-
"
|
153 |
-
"
|
154 |
-
"
|
155 |
-
"
|
156 |
-
"
|
157 |
-
"
|
158 |
-
"
|
159 |
-
"
|
160 |
-
"
|
161 |
-
"
|
162 |
-
"
|
163 |
-
"
|
164 |
-
"
|
165 |
-
"
|
166 |
-
"
|
167 |
-
"
|
168 |
-
"
|
169 |
-
"
|
170 |
-
"
|
171 |
-
"
|
172 |
-
"
|
173 |
-
"
|
174 |
-
"
|
175 |
-
"
|
176 |
-
"
|
177 |
-
"
|
178 |
-
"
|
179 |
-
"
|
180 |
-
"
|
181 |
-
"
|
182 |
-
"
|
183 |
-
"
|
184 |
-
"
|
185 |
-
"
|
186 |
-
"
|
187 |
-
"
|
188 |
-
"
|
189 |
-
"
|
190 |
-
"
|
191 |
-
"
|
192 |
-
"
|
193 |
-
"
|
194 |
-
"
|
195 |
-
"
|
196 |
-
"
|
197 |
-
"
|
198 |
-
"
|
199 |
-
"
|
200 |
-
"
|
201 |
-
"
|
202 |
-
"
|
203 |
-
"
|
204 |
-
"
|
205 |
-
"
|
206 |
-
"
|
207 |
-
"
|
208 |
-
"
|
209 |
-
"
|
210 |
-
"
|
211 |
-
"
|
212 |
-
"
|
213 |
-
"
|
214 |
-
"
|
215 |
-
"
|
216 |
-
"
|
217 |
-
"
|
218 |
-
"
|
219 |
-
"
|
220 |
-
"
|
221 |
-
"
|
222 |
-
"
|
223 |
-
"
|
224 |
-
"
|
225 |
-
"
|
226 |
-
"
|
227 |
-
"
|
228 |
-
"
|
229 |
-
"
|
230 |
-
"
|
231 |
-
"
|
232 |
-
"
|
233 |
-
"
|
234 |
-
"
|
235 |
-
"
|
236 |
-
"
|
237 |
-
"
|
238 |
-
"
|
239 |
-
"
|
240 |
-
"
|
241 |
-
"
|
242 |
-
"
|
243 |
-
"
|
244 |
-
"
|
245 |
-
"
|
246 |
-
"
|
247 |
-
"
|
248 |
-
"
|
249 |
-
"
|
250 |
-
"
|
251 |
-
"
|
252 |
-
"
|
253 |
-
"
|
254 |
-
"
|
255 |
-
"
|
256 |
-
"
|
257 |
-
"
|
258 |
-
"
|
259 |
-
"
|
260 |
-
"
|
261 |
-
"
|
262 |
-
"
|
263 |
-
"
|
264 |
-
"
|
265 |
-
"
|
266 |
-
"
|
267 |
-
"
|
268 |
-
"
|
269 |
-
"
|
270 |
-
"
|
271 |
-
"
|
272 |
-
"
|
273 |
-
"
|
274 |
-
"
|
275 |
-
"
|
276 |
-
"
|
277 |
-
"
|
278 |
-
"
|
279 |
-
"
|
280 |
-
"
|
281 |
-
"
|
282 |
-
"
|
283 |
-
"
|
284 |
-
"
|
285 |
-
"
|
286 |
-
"
|
287 |
-
"
|
288 |
-
"
|
289 |
-
"
|
290 |
-
"
|
291 |
-
"
|
292 |
-
"
|
293 |
-
"
|
294 |
-
"
|
295 |
-
"
|
296 |
-
"
|
297 |
-
"
|
298 |
-
"
|
299 |
-
"
|
300 |
-
"
|
301 |
-
"
|
302 |
-
"
|
303 |
-
"
|
304 |
-
"
|
305 |
-
"
|
306 |
-
"
|
307 |
-
"
|
308 |
-
"
|
309 |
-
"
|
310 |
-
"
|
311 |
-
"
|
312 |
-
"
|
313 |
-
"
|
314 |
-
"
|
315 |
-
"
|
316 |
-
"
|
317 |
-
"
|
318 |
-
"
|
319 |
-
"
|
320 |
-
"
|
321 |
-
"
|
322 |
-
"
|
323 |
-
"
|
324 |
-
"
|
325 |
-
"
|
326 |
-
"
|
327 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
328 |
}
|
329 |
}
|
|
|
3 |
"total_size": 6965007360
|
4 |
},
|
5 |
"weight_map": {
|
6 |
+
"model.embed_tokens.weight": "model-00001-of-00002.safetensors",
|
7 |
+
"model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
8 |
+
"model.layers.0.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
9 |
+
"model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
10 |
+
"model.layers.0.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
11 |
+
"model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
12 |
+
"model.layers.0.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
13 |
+
"model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
14 |
+
"model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
15 |
+
"model.layers.0.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
16 |
+
"model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
17 |
+
"model.layers.0.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
18 |
+
"model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
19 |
+
"model.layers.0.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
20 |
+
"model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
21 |
+
"model.layers.0.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
22 |
+
"model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
23 |
+
"model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
24 |
+
"model.layers.1.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
25 |
+
"model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
26 |
+
"model.layers.1.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
27 |
+
"model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
28 |
+
"model.layers.1.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
29 |
+
"model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
30 |
+
"model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
31 |
+
"model.layers.1.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
32 |
+
"model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
33 |
+
"model.layers.1.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
34 |
+
"model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
35 |
+
"model.layers.1.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
36 |
+
"model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
37 |
+
"model.layers.1.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
38 |
+
"model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
39 |
+
"model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
40 |
+
"model.layers.10.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
41 |
+
"model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
42 |
+
"model.layers.10.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
43 |
+
"model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
44 |
+
"model.layers.10.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
45 |
+
"model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
46 |
+
"model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
47 |
+
"model.layers.10.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
48 |
+
"model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
49 |
+
"model.layers.10.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
50 |
+
"model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
51 |
+
"model.layers.10.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
52 |
+
"model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
53 |
+
"model.layers.10.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
54 |
+
"model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
55 |
+
"model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
56 |
+
"model.layers.11.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
57 |
+
"model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
58 |
+
"model.layers.11.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
59 |
+
"model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
60 |
+
"model.layers.11.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
61 |
+
"model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
62 |
+
"model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
63 |
+
"model.layers.11.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
64 |
+
"model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
65 |
+
"model.layers.11.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
66 |
+
"model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
67 |
+
"model.layers.11.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
68 |
+
"model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
69 |
+
"model.layers.11.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
70 |
+
"model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
71 |
+
"model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
72 |
+
"model.layers.12.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
73 |
+
"model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
74 |
+
"model.layers.12.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
75 |
+
"model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
76 |
+
"model.layers.12.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
77 |
+
"model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
78 |
+
"model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
79 |
+
"model.layers.12.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
80 |
+
"model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
81 |
+
"model.layers.12.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
82 |
+
"model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
83 |
+
"model.layers.12.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
84 |
+
"model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
85 |
+
"model.layers.12.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
86 |
+
"model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
87 |
+
"model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
88 |
+
"model.layers.13.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
89 |
+
"model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
90 |
+
"model.layers.13.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
91 |
+
"model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
92 |
+
"model.layers.13.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
93 |
+
"model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
94 |
+
"model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
95 |
+
"model.layers.13.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
96 |
+
"model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
97 |
+
"model.layers.13.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
98 |
+
"model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
99 |
+
"model.layers.13.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
100 |
+
"model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
101 |
+
"model.layers.13.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
102 |
+
"model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
103 |
+
"model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
104 |
+
"model.layers.14.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
105 |
+
"model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
106 |
+
"model.layers.14.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
107 |
+
"model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
108 |
+
"model.layers.14.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
109 |
+
"model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
110 |
+
"model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
111 |
+
"model.layers.14.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
112 |
+
"model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
113 |
+
"model.layers.14.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
114 |
+
"model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
115 |
+
"model.layers.14.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
116 |
+
"model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
117 |
+
"model.layers.14.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
118 |
+
"model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
119 |
+
"model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
120 |
+
"model.layers.15.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
121 |
+
"model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
122 |
+
"model.layers.15.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
123 |
+
"model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
124 |
+
"model.layers.15.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
125 |
+
"model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
126 |
+
"model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
127 |
+
"model.layers.15.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
128 |
+
"model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
129 |
+
"model.layers.15.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
130 |
+
"model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
131 |
+
"model.layers.15.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
132 |
+
"model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
133 |
+
"model.layers.15.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
134 |
+
"model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
135 |
+
"model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
136 |
+
"model.layers.16.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
137 |
+
"model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
138 |
+
"model.layers.16.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
139 |
+
"model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
140 |
+
"model.layers.16.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
141 |
+
"model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
142 |
+
"model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
143 |
+
"model.layers.16.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
144 |
+
"model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
145 |
+
"model.layers.16.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
146 |
+
"model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
147 |
+
"model.layers.16.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
148 |
+
"model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
149 |
+
"model.layers.16.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
150 |
+
"model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
151 |
+
"model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
152 |
+
"model.layers.17.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
153 |
+
"model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
154 |
+
"model.layers.17.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
155 |
+
"model.layers.17.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
156 |
+
"model.layers.17.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
157 |
+
"model.layers.17.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
158 |
+
"model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
159 |
+
"model.layers.17.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
160 |
+
"model.layers.17.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
161 |
+
"model.layers.17.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
162 |
+
"model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
163 |
+
"model.layers.17.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
164 |
+
"model.layers.17.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
165 |
+
"model.layers.17.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
166 |
+
"model.layers.17.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
167 |
+
"model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
168 |
+
"model.layers.18.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
169 |
+
"model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
170 |
+
"model.layers.18.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
171 |
+
"model.layers.18.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
172 |
+
"model.layers.18.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
173 |
+
"model.layers.18.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
174 |
+
"model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
175 |
+
"model.layers.18.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
176 |
+
"model.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
177 |
+
"model.layers.18.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
178 |
+
"model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
179 |
+
"model.layers.18.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
180 |
+
"model.layers.18.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
181 |
+
"model.layers.18.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
182 |
+
"model.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
183 |
+
"model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
184 |
+
"model.layers.19.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
185 |
+
"model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
186 |
+
"model.layers.19.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
187 |
+
"model.layers.19.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
188 |
+
"model.layers.19.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
189 |
+
"model.layers.19.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
190 |
+
"model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
191 |
+
"model.layers.19.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
192 |
+
"model.layers.19.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
193 |
+
"model.layers.19.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
194 |
+
"model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
195 |
+
"model.layers.19.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
196 |
+
"model.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
197 |
+
"model.layers.19.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
198 |
+
"model.layers.19.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
199 |
+
"model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
200 |
+
"model.layers.2.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
201 |
+
"model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
202 |
+
"model.layers.2.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
203 |
+
"model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
204 |
+
"model.layers.2.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
205 |
+
"model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
206 |
+
"model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
207 |
+
"model.layers.2.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
208 |
+
"model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
209 |
+
"model.layers.2.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
210 |
+
"model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
211 |
+
"model.layers.2.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
212 |
+
"model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
213 |
+
"model.layers.2.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
214 |
+
"model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
215 |
+
"model.layers.20.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
216 |
+
"model.layers.20.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
217 |
+
"model.layers.20.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
218 |
+
"model.layers.20.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
219 |
+
"model.layers.20.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
220 |
+
"model.layers.20.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
221 |
+
"model.layers.20.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
222 |
+
"model.layers.20.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
223 |
+
"model.layers.20.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
224 |
+
"model.layers.20.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
225 |
+
"model.layers.20.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
226 |
+
"model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
227 |
+
"model.layers.20.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
228 |
+
"model.layers.20.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
229 |
+
"model.layers.20.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
230 |
+
"model.layers.20.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
231 |
+
"model.layers.21.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
232 |
+
"model.layers.21.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
233 |
+
"model.layers.21.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
234 |
+
"model.layers.21.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
235 |
+
"model.layers.21.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
236 |
+
"model.layers.21.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
237 |
+
"model.layers.21.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
238 |
+
"model.layers.21.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
239 |
+
"model.layers.21.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
240 |
+
"model.layers.21.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
241 |
+
"model.layers.21.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
242 |
+
"model.layers.21.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
243 |
+
"model.layers.21.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
244 |
+
"model.layers.21.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
245 |
+
"model.layers.21.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
246 |
+
"model.layers.21.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
247 |
+
"model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
248 |
+
"model.layers.22.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
249 |
+
"model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
250 |
+
"model.layers.22.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
251 |
+
"model.layers.22.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
252 |
+
"model.layers.22.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
253 |
+
"model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
254 |
+
"model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
255 |
+
"model.layers.22.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
256 |
+
"model.layers.22.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
257 |
+
"model.layers.22.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
258 |
+
"model.layers.22.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
259 |
+
"model.layers.22.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
260 |
+
"model.layers.22.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
261 |
+
"model.layers.22.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
262 |
+
"model.layers.22.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
263 |
+
"model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
264 |
+
"model.layers.23.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
265 |
+
"model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
266 |
+
"model.layers.23.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
267 |
+
"model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
268 |
+
"model.layers.23.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
269 |
+
"model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
270 |
+
"model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
271 |
+
"model.layers.23.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
272 |
+
"model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
273 |
+
"model.layers.23.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
274 |
+
"model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
275 |
+
"model.layers.23.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
276 |
+
"model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
277 |
+
"model.layers.23.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
278 |
+
"model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
279 |
+
"model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
280 |
+
"model.layers.24.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
281 |
+
"model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
282 |
+
"model.layers.24.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
283 |
+
"model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
284 |
+
"model.layers.24.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
285 |
+
"model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
286 |
+
"model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
287 |
+
"model.layers.24.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
288 |
+
"model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
289 |
+
"model.layers.24.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
290 |
+
"model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
291 |
+
"model.layers.24.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
292 |
+
"model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
293 |
+
"model.layers.24.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
294 |
+
"model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
295 |
+
"model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
296 |
+
"model.layers.25.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
297 |
+
"model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
298 |
+
"model.layers.25.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
299 |
+
"model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
300 |
+
"model.layers.25.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
301 |
+
"model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
302 |
+
"model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
303 |
+
"model.layers.25.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
304 |
+
"model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
305 |
+
"model.layers.25.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
306 |
+
"model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
307 |
+
"model.layers.25.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
308 |
+
"model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
309 |
+
"model.layers.25.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
310 |
+
"model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
311 |
+
"model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
312 |
+
"model.layers.26.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
313 |
+
"model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
314 |
+
"model.layers.26.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
315 |
+
"model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
316 |
+
"model.layers.26.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
317 |
+
"model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
318 |
+
"model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
319 |
+
"model.layers.26.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
320 |
+
"model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
321 |
+
"model.layers.26.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
322 |
+
"model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
323 |
+
"model.layers.26.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
324 |
+
"model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
325 |
+
"model.layers.26.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
326 |
+
"model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
327 |
+
"model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
328 |
+
"model.layers.27.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
329 |
+
"model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
330 |
+
"model.layers.27.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
331 |
+
"model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
332 |
+
"model.layers.27.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
333 |
+
"model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
334 |
+
"model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
335 |
+
"model.layers.27.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
336 |
+
"model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
337 |
+
"model.layers.27.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
338 |
+
"model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
339 |
+
"model.layers.27.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
340 |
+
"model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
341 |
+
"model.layers.27.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
342 |
+
"model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
343 |
+
"model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
344 |
+
"model.layers.28.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
345 |
+
"model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
346 |
+
"model.layers.28.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
347 |
+
"model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
348 |
+
"model.layers.28.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
349 |
+
"model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
350 |
+
"model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
351 |
+
"model.layers.28.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
352 |
+
"model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
353 |
+
"model.layers.28.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
354 |
+
"model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
355 |
+
"model.layers.28.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
356 |
+
"model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
357 |
+
"model.layers.28.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
358 |
+
"model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
359 |
+
"model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
360 |
+
"model.layers.29.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
361 |
+
"model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
362 |
+
"model.layers.29.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
363 |
+
"model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
364 |
+
"model.layers.29.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
365 |
+
"model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
366 |
+
"model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
367 |
+
"model.layers.29.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
368 |
+
"model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
369 |
+
"model.layers.29.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
370 |
+
"model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
371 |
+
"model.layers.29.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
372 |
+
"model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
373 |
+
"model.layers.29.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
374 |
+
"model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
375 |
+
"model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
376 |
+
"model.layers.3.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
377 |
+
"model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
378 |
+
"model.layers.3.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
379 |
+
"model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
380 |
+
"model.layers.3.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
381 |
+
"model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
382 |
+
"model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
383 |
+
"model.layers.3.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
384 |
+
"model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
385 |
+
"model.layers.3.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
386 |
+
"model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
387 |
+
"model.layers.3.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
388 |
+
"model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
389 |
+
"model.layers.3.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
390 |
+
"model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
391 |
+
"model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
392 |
+
"model.layers.30.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
393 |
+
"model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
394 |
+
"model.layers.30.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
395 |
+
"model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
396 |
+
"model.layers.30.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
397 |
+
"model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
398 |
+
"model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
399 |
+
"model.layers.30.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
400 |
+
"model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
401 |
+
"model.layers.30.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
402 |
+
"model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
403 |
+
"model.layers.30.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
404 |
+
"model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
405 |
+
"model.layers.30.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
406 |
+
"model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
407 |
+
"model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
408 |
+
"model.layers.31.mlp.down_proj.bias": "model-00002-of-00002.safetensors",
|
409 |
+
"model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
410 |
+
"model.layers.31.mlp.gate_proj.bias": "model-00002-of-00002.safetensors",
|
411 |
+
"model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
412 |
+
"model.layers.31.mlp.up_proj.bias": "model-00002-of-00002.safetensors",
|
413 |
+
"model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
414 |
+
"model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
415 |
+
"model.layers.31.self_attn.k_proj.bias": "model-00002-of-00002.safetensors",
|
416 |
+
"model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
417 |
+
"model.layers.31.self_attn.o_proj.bias": "model-00002-of-00002.safetensors",
|
418 |
+
"model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
419 |
+
"model.layers.31.self_attn.q_proj.bias": "model-00002-of-00002.safetensors",
|
420 |
+
"model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
421 |
+
"model.layers.31.self_attn.v_proj.bias": "model-00002-of-00002.safetensors",
|
422 |
+
"model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
423 |
+
"model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
424 |
+
"model.layers.4.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
425 |
+
"model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
426 |
+
"model.layers.4.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
427 |
+
"model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
428 |
+
"model.layers.4.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
429 |
+
"model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
430 |
+
"model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
431 |
+
"model.layers.4.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
432 |
+
"model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
433 |
+
"model.layers.4.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
434 |
+
"model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
435 |
+
"model.layers.4.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
436 |
+
"model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
437 |
+
"model.layers.4.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
438 |
+
"model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
439 |
+
"model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
440 |
+
"model.layers.5.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
441 |
+
"model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
442 |
+
"model.layers.5.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
443 |
+
"model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
444 |
+
"model.layers.5.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
445 |
+
"model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
446 |
+
"model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
447 |
+
"model.layers.5.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
448 |
+
"model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
449 |
+
"model.layers.5.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
450 |
+
"model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
451 |
+
"model.layers.5.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
452 |
+
"model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
453 |
+
"model.layers.5.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
454 |
+
"model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
455 |
+
"model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
456 |
+
"model.layers.6.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
457 |
+
"model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
458 |
+
"model.layers.6.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
459 |
+
"model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
460 |
+
"model.layers.6.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
461 |
+
"model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
462 |
+
"model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
463 |
+
"model.layers.6.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
464 |
+
"model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
465 |
+
"model.layers.6.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
466 |
+
"model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
467 |
+
"model.layers.6.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
468 |
+
"model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
469 |
+
"model.layers.6.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
470 |
+
"model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
471 |
+
"model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
472 |
+
"model.layers.7.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
473 |
+
"model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
474 |
+
"model.layers.7.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
475 |
+
"model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
476 |
+
"model.layers.7.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
477 |
+
"model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
478 |
+
"model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
479 |
+
"model.layers.7.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
480 |
+
"model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
481 |
+
"model.layers.7.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
482 |
+
"model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
483 |
+
"model.layers.7.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
484 |
+
"model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
485 |
+
"model.layers.7.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
486 |
+
"model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
487 |
+
"model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
488 |
+
"model.layers.8.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
489 |
+
"model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
490 |
+
"model.layers.8.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
491 |
+
"model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
492 |
+
"model.layers.8.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
493 |
+
"model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
494 |
+
"model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
495 |
+
"model.layers.8.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
496 |
+
"model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
497 |
+
"model.layers.8.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
498 |
+
"model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
499 |
+
"model.layers.8.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
500 |
+
"model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
501 |
+
"model.layers.8.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
502 |
+
"model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
503 |
+
"model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
504 |
+
"model.layers.9.mlp.down_proj.bias": "model-00001-of-00002.safetensors",
|
505 |
+
"model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
506 |
+
"model.layers.9.mlp.gate_proj.bias": "model-00001-of-00002.safetensors",
|
507 |
+
"model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
508 |
+
"model.layers.9.mlp.up_proj.bias": "model-00001-of-00002.safetensors",
|
509 |
+
"model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
510 |
+
"model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
511 |
+
"model.layers.9.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
512 |
+
"model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
513 |
+
"model.layers.9.self_attn.o_proj.bias": "model-00001-of-00002.safetensors",
|
514 |
+
"model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
515 |
+
"model.layers.9.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
516 |
+
"model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
517 |
+
"model.layers.9.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
518 |
+
"model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
519 |
+
"model.norm.weight": "model-00002-of-00002.safetensors"
|
520 |
}
|
521 |
}
|
modeling_granite.py
DELETED
@@ -1,1376 +0,0 @@
|
|
1 |
-
import math
|
2 |
-
import numbers
|
3 |
-
import warnings
|
4 |
-
from enum import Enum
|
5 |
-
from typing import Optional, Tuple, Union
|
6 |
-
|
7 |
-
import torch
|
8 |
-
import torch.nn as nn
|
9 |
-
import torch.nn.functional as F
|
10 |
-
from transformers import DynamicCache, PreTrainedModel
|
11 |
-
from transformers.activations import get_activation as get_base_activation
|
12 |
-
from transformers.modeling_outputs import BaseModelOutputWithPastAndCrossAttentions, CausalLMOutputWithCrossAttentions
|
13 |
-
from transformers.utils import is_flash_attn_2_available
|
14 |
-
|
15 |
-
from .configuration_granite import GraniteConfig
|
16 |
-
|
17 |
-
|
18 |
-
class PositionEmbeddingType(Enum):
|
19 |
-
learned_absolute = "learned_absolute"
|
20 |
-
alibi = "alibi"
|
21 |
-
rope = "rope"
|
22 |
-
|
23 |
-
|
24 |
-
class AttentionHeadType(Enum):
|
25 |
-
mha = "mha"
|
26 |
-
mqa = "mqa"
|
27 |
-
gqa = "gqa"
|
28 |
-
|
29 |
-
|
30 |
-
if is_flash_attn_2_available():
|
31 |
-
from flash_attn.bert_padding import IndexFirstAxis, pad_input, unpad_input
|
32 |
-
from flash_attn.flash_attn_interface import flash_attn_varlen_func
|
33 |
-
|
34 |
-
|
35 |
-
# Copied from transformers.models.llama.modeling_llama._get_unpad_data
|
36 |
-
def get_unpad_data(attention_mask: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
|
37 |
-
seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
|
38 |
-
indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
|
39 |
-
max_seqlen_in_batch = seqlens_in_batch.max().item()
|
40 |
-
cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
|
41 |
-
return indices, cu_seqlens, max_seqlen_in_batch
|
42 |
-
|
43 |
-
|
44 |
-
def repeat_key_value(x: torch.Tensor, num_heads: int, num_key_value_heads: int) -> torch.Tensor:
|
45 |
-
num_groups = num_heads // num_key_value_heads
|
46 |
-
|
47 |
-
# mha
|
48 |
-
if num_groups == 1:
|
49 |
-
return x
|
50 |
-
|
51 |
-
# mqa
|
52 |
-
if num_key_value_heads == 1:
|
53 |
-
return x.expand(-1, num_heads, -1, -1)
|
54 |
-
|
55 |
-
# gqa
|
56 |
-
return x.repeat_interleave(num_groups, dim=1)
|
57 |
-
|
58 |
-
|
59 |
-
##################################################
|
60 |
-
# activation functions
|
61 |
-
|
62 |
-
|
63 |
-
_GLU_BASE_MAPPING = {
|
64 |
-
"geglu": "gelu",
|
65 |
-
"miglu": "mish",
|
66 |
-
"mishglu": "mish",
|
67 |
-
"swiglu": "swish",
|
68 |
-
}
|
69 |
-
|
70 |
-
|
71 |
-
class GLUActivation(nn.Module):
|
72 |
-
def __init__(self, base_activation: nn.Module) -> None:
|
73 |
-
super().__init__()
|
74 |
-
self.base_activation = base_activation
|
75 |
-
|
76 |
-
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
77 |
-
x = x.chunk(2, dim=-1)
|
78 |
-
return x[0] * self.base_activation(x[1])
|
79 |
-
|
80 |
-
|
81 |
-
def is_glu(name: str) -> bool:
|
82 |
-
return name.endswith("glu")
|
83 |
-
|
84 |
-
|
85 |
-
def get_activation_function(name: str) -> nn.Module:
|
86 |
-
if is_glu(name):
|
87 |
-
# for glu and sigmoid_glu, we directly return the pytorch's GLU
|
88 |
-
if name in ["glu", "sigmoid_glu"]:
|
89 |
-
activation_function = nn.modules.GLU()
|
90 |
-
else:
|
91 |
-
if name in _GLU_BASE_MAPPING:
|
92 |
-
name = _GLU_BASE_MAPPING[name]
|
93 |
-
elif name.endswith("_glu"):
|
94 |
-
name = name.rstrip("_glu")
|
95 |
-
else:
|
96 |
-
raise ValueError("invalid activation function")
|
97 |
-
|
98 |
-
base_activation = get_base_activation(name)
|
99 |
-
activation_function = GLUActivation(base_activation)
|
100 |
-
else:
|
101 |
-
activation_function = get_base_activation(name)
|
102 |
-
|
103 |
-
return activation_function
|
104 |
-
|
105 |
-
|
106 |
-
##################################################
|
107 |
-
# normalization functions
|
108 |
-
|
109 |
-
|
110 |
-
class RMSNorm(nn.Module):
|
111 |
-
def __init__(self, normalized_shape: int, eps: float = 1e-6) -> None:
|
112 |
-
super().__init__()
|
113 |
-
|
114 |
-
self.weight = nn.Parameter(torch.ones(normalized_shape))
|
115 |
-
self.eps = eps
|
116 |
-
|
117 |
-
if isinstance(normalized_shape, numbers.Integral):
|
118 |
-
normalized_shape = (normalized_shape,)
|
119 |
-
self.normalized_shape = normalized_shape
|
120 |
-
|
121 |
-
def forward(self, input: torch.Tensor) -> torch.Tensor:
|
122 |
-
input_dtype = input.dtype
|
123 |
-
|
124 |
-
input = input.to(torch.float32)
|
125 |
-
variance = input.pow(2).mean(-1, keepdim=True)
|
126 |
-
input = input * torch.rsqrt(variance + self.eps)
|
127 |
-
|
128 |
-
return self.weight * input.to(input_dtype)
|
129 |
-
|
130 |
-
def extra_repr(self) -> str:
|
131 |
-
return f"{self.normalized_shape}, eps={self.eps}"
|
132 |
-
|
133 |
-
def reset_parameters(self) -> None:
|
134 |
-
nn.init.ones_(self.weight)
|
135 |
-
|
136 |
-
|
137 |
-
_NORMALIZATION_FUNCTIONS = {
|
138 |
-
"layernorm": nn.LayerNorm,
|
139 |
-
"rmsnorm": RMSNorm,
|
140 |
-
}
|
141 |
-
|
142 |
-
|
143 |
-
def get_normalization_function(name: str, normalized_shape: int, eps: float = 1e-5) -> nn.Module:
|
144 |
-
if name in _NORMALIZATION_FUNCTIONS:
|
145 |
-
return _NORMALIZATION_FUNCTIONS[name](normalized_shape, eps=eps)
|
146 |
-
|
147 |
-
raise ValueError(f"unexpected `normalization_function` {name}")
|
148 |
-
|
149 |
-
|
150 |
-
##################################################
|
151 |
-
# attention modules
|
152 |
-
|
153 |
-
|
154 |
-
class GraniteAttention(nn.Module):
|
155 |
-
def __init__(self, config: GraniteConfig, causal: bool, layer_idx: Optional[int] = None) -> None:
|
156 |
-
super().__init__()
|
157 |
-
|
158 |
-
self.causal = causal
|
159 |
-
self.hidden_size = config.n_embd
|
160 |
-
self.num_heads = config.n_head
|
161 |
-
self.num_key_value_heads = config.num_key_value_heads
|
162 |
-
self.add_bias = config.add_bias
|
163 |
-
|
164 |
-
assert (
|
165 |
-
self.hidden_size % self.num_heads == 0
|
166 |
-
), f"`hidden_size` ({self.hidden_size}) must be divisible by `num_heads` ({self.num_heads})"
|
167 |
-
|
168 |
-
self.head_dim = self.hidden_size // self.num_heads
|
169 |
-
self.attention_head_type = AttentionHeadType(config.attention_head_type)
|
170 |
-
|
171 |
-
self.position_embedding_type = PositionEmbeddingType(config.position_embedding_type)
|
172 |
-
self.scale_attn_weights = config.scale_attn_weights
|
173 |
-
self.attention_multiplier = config.attention_multiplier
|
174 |
-
|
175 |
-
self.layer_idx = layer_idx
|
176 |
-
self.attention_softmax_in_fp32 = config.attention_softmax_in_fp32
|
177 |
-
self.scale_attention_softmax_in_fp32 = (
|
178 |
-
config.scale_attention_softmax_in_fp32 and config.attention_softmax_in_fp32
|
179 |
-
)
|
180 |
-
|
181 |
-
if self.attention_head_type == AttentionHeadType.mha:
|
182 |
-
if self.num_key_value_heads is None:
|
183 |
-
self.num_key_value_heads = self.num_heads
|
184 |
-
|
185 |
-
assert (
|
186 |
-
self.num_heads == self.num_key_value_heads
|
187 |
-
), f"{self.__class__.__name__} should have same number of heads for query, keys and values"
|
188 |
-
elif self.attention_head_type == AttentionHeadType.gqa:
|
189 |
-
assert (
|
190 |
-
self.num_key_value_heads is not None
|
191 |
-
), "`num_key_value_heads` needs to be specified with GroupedQueryAttention"
|
192 |
-
|
193 |
-
assert self.num_heads % self.num_key_value_heads == 0, (
|
194 |
-
f"`num_heads` ({self.num_heads}) should be a multiple of `num_key_value_heads` "
|
195 |
-
f"({self.num_key_value_heads})"
|
196 |
-
)
|
197 |
-
elif self.attention_head_type == AttentionHeadType.mqa:
|
198 |
-
if self.num_key_value_heads is None:
|
199 |
-
self.num_key_value_heads = 1
|
200 |
-
|
201 |
-
assert self.num_key_value_heads == 1, f"{self.__class__.__name__} should have 1 head for keys and values"
|
202 |
-
else:
|
203 |
-
raise ValueError(f"unexpected attention_head_type ({self.attention_head_type})")
|
204 |
-
|
205 |
-
# note that the actual layout is different for the output and depends on whether we are using MHA, MQA or GQA
|
206 |
-
# (self.hidden_size + 2 * self.num_key_value_heads * self.head_dim) is just the actual number output features
|
207 |
-
self.c_attn = nn.Linear(
|
208 |
-
self.hidden_size, self.hidden_size + 2 * self.num_key_value_heads * self.head_dim, bias=self.add_bias
|
209 |
-
)
|
210 |
-
self.c_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=self.add_bias)
|
211 |
-
|
212 |
-
self.attn_pdrop = config.attn_pdrop
|
213 |
-
self.resid_pdrop = config.resid_pdrop
|
214 |
-
|
215 |
-
self.attn_dropout = nn.Identity() if self.attn_pdrop == 0 else nn.Dropout(self.attn_pdrop)
|
216 |
-
self.resid_dropout = nn.Identity() if self.resid_pdrop == 0 else nn.Dropout(self.resid_pdrop)
|
217 |
-
|
218 |
-
def _prepare_qkv_for_forward(self, hidden_states: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
|
219 |
-
# ==========================================================================================
|
220 |
-
# hidden_states -> (batch_size, query_length, num_heads * head_dim)
|
221 |
-
# ==========================================================================================
|
222 |
-
|
223 |
-
# the output of following is a tuple if using MQA with tensor parallel
|
224 |
-
hidden_states = self.c_attn(hidden_states)
|
225 |
-
|
226 |
-
# ==========================================================================================
|
227 |
-
# hidden_states -> (batch_size, query_length, [num_heads + num_key_value_heads * 2] * head_dim)
|
228 |
-
# ==========================================================================================
|
229 |
-
|
230 |
-
# for MHA, we can get away with doing just 1 transpose which is not true for GQA
|
231 |
-
if self.attention_head_type == AttentionHeadType.mha:
|
232 |
-
query, key, value = self._prepare_qkv_for_forward_mha(hidden_states)
|
233 |
-
elif self.attention_head_type == AttentionHeadType.gqa:
|
234 |
-
query, key, value = self._prepare_qkv_for_forward_gqa(hidden_states)
|
235 |
-
elif self.attention_head_type == AttentionHeadType.mqa:
|
236 |
-
query, key, value = self._prepare_qkv_for_forward_mqa(hidden_states)
|
237 |
-
else:
|
238 |
-
raise ValueError(f"unexpected attention_head_type ({self.attention_head_type})")
|
239 |
-
|
240 |
-
# ==========================================================================================
|
241 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
242 |
-
# key -> (batch_size, num_key_value_heads, query_length, head_dim)
|
243 |
-
# value -> (batch_size, num_key_value_heads, query_length, head_dim)
|
244 |
-
# ==========================================================================================
|
245 |
-
|
246 |
-
return query, key, value
|
247 |
-
|
248 |
-
def _prepare_qkv_for_forward_mha(
|
249 |
-
self, hidden_states: torch.Tensor
|
250 |
-
) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
|
251 |
-
batch_size, query_length = hidden_states.shape[:-1]
|
252 |
-
|
253 |
-
hidden_states = hidden_states.view(batch_size, query_length, self.num_heads, -1)
|
254 |
-
hidden_states = hidden_states.transpose(1, 2)
|
255 |
-
|
256 |
-
query, key, value = hidden_states.chunk(3, dim=-1)
|
257 |
-
|
258 |
-
return query, key, value
|
259 |
-
|
260 |
-
def _prepare_qkv_for_forward_gqa(
|
261 |
-
self, hidden_states: torch.Tensor
|
262 |
-
) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
|
263 |
-
batch_size, query_length = hidden_states.shape[:-1]
|
264 |
-
|
265 |
-
hidden_states = hidden_states.view(batch_size, query_length, self.num_key_value_heads, -1)
|
266 |
-
|
267 |
-
query, key, value = hidden_states.split(
|
268 |
-
((self.num_heads // self.num_key_value_heads) * self.head_dim, self.head_dim, self.head_dim), dim=-1
|
269 |
-
)
|
270 |
-
|
271 |
-
# this needs to be a reshape instead of view sadly
|
272 |
-
query = query.reshape(batch_size, query_length, -1, self.head_dim)
|
273 |
-
|
274 |
-
query = query.transpose(1, 2)
|
275 |
-
key = key.transpose(1, 2)
|
276 |
-
value = value.transpose(1, 2)
|
277 |
-
|
278 |
-
return query, key, value
|
279 |
-
|
280 |
-
def _prepare_qkv_for_forward_mqa(
|
281 |
-
self, hidden_states: torch.Tensor
|
282 |
-
) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
|
283 |
-
batch_size, query_length = hidden_states.shape[:-1]
|
284 |
-
|
285 |
-
query, key, value = hidden_states.split((self.hidden_size, self.head_dim, self.head_dim), dim=-1)
|
286 |
-
|
287 |
-
query = query.view(batch_size, query_length, self.num_heads, -1)
|
288 |
-
|
289 |
-
query = query.transpose(1, 2)
|
290 |
-
key = key.unsqueeze(1)
|
291 |
-
value = value.unsqueeze(1)
|
292 |
-
|
293 |
-
return query, key, value
|
294 |
-
|
295 |
-
def forward(
|
296 |
-
self,
|
297 |
-
hidden_states: torch.Tensor,
|
298 |
-
past_key_values: Optional[DynamicCache] = None,
|
299 |
-
attention_mask: Optional[torch.Tensor] = None,
|
300 |
-
rope_cos_sin: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
|
301 |
-
) -> torch.Tensor:
|
302 |
-
# ==========================================================================================
|
303 |
-
# hidden_states -> (batch_size, query_length, num_heads * head_dim)
|
304 |
-
# ==========================================================================================
|
305 |
-
|
306 |
-
query, key, value = self._prepare_qkv_for_forward(hidden_states)
|
307 |
-
|
308 |
-
# ==========================================================================================
|
309 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
310 |
-
# key -> (batch_size, num_key_value_heads, query_length, head_dim)
|
311 |
-
# value -> (batch_size, num_key_value_heads, query_length, head_dim)
|
312 |
-
# ==========================================================================================
|
313 |
-
|
314 |
-
if self.position_embedding_type == PositionEmbeddingType.rope:
|
315 |
-
query = apply_rotary_pos_emb(query, rope_cos_sin)
|
316 |
-
key = apply_rotary_pos_emb(key, rope_cos_sin)
|
317 |
-
|
318 |
-
if past_key_values is not None:
|
319 |
-
key, value = past_key_values.update(key, value, self.layer_idx)
|
320 |
-
|
321 |
-
# ==========================================================================================
|
322 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
323 |
-
# key -> (batch_size, num_key_value_heads, key_length, head_dim)
|
324 |
-
# value -> (batch_size, num_key_value_heads, key_length, head_dim)
|
325 |
-
# ==========================================================================================
|
326 |
-
|
327 |
-
key = key.transpose(-1, -2)
|
328 |
-
|
329 |
-
dtype = query.dtype
|
330 |
-
softmax_dtype = torch.float32 if self.attention_softmax_in_fp32 else dtype
|
331 |
-
|
332 |
-
if self.scale_attn_weights:
|
333 |
-
if self.attention_multiplier is None:
|
334 |
-
scale_factor = 1 / self.head_dim**0.5
|
335 |
-
else:
|
336 |
-
scale_factor = self.attention_multiplier
|
337 |
-
else:
|
338 |
-
scale_factor = 1
|
339 |
-
|
340 |
-
# ==========================================================================================
|
341 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
342 |
-
# key -> (batch_size, num_key_value_heads, head_dim, key_length)
|
343 |
-
# value -> (batch_size, num_key_value_heads, key_length, head_dim)
|
344 |
-
# ==========================================================================================
|
345 |
-
|
346 |
-
batch_size = query.shape[0]
|
347 |
-
query_length = query.shape[2]
|
348 |
-
key_length = key.shape[-1]
|
349 |
-
|
350 |
-
key = repeat_key_value(key, self.num_heads, self.num_key_value_heads)
|
351 |
-
value = repeat_key_value(value, self.num_heads, self.num_key_value_heads)
|
352 |
-
|
353 |
-
# Always copies
|
354 |
-
query = query.reshape(batch_size * self.num_heads, query_length, self.head_dim)
|
355 |
-
# No copy when layer_past is provided.
|
356 |
-
key = key.reshape(batch_size * self.num_heads, self.head_dim, key_length)
|
357 |
-
|
358 |
-
# ==========================================================================================
|
359 |
-
# query -> (batch_size * num_heads, query_length, head_dim)
|
360 |
-
# key -> (batch_size * num_heads, head_dim, key_length)
|
361 |
-
# value -> (batch_size, num_heads, key_length, head_dim)
|
362 |
-
# ==========================================================================================
|
363 |
-
|
364 |
-
attn_weights = torch.empty(
|
365 |
-
(batch_size * self.num_heads, query_length, key_length), device=query.device, dtype=query.dtype
|
366 |
-
)
|
367 |
-
|
368 |
-
attn_weights = torch.baddbmm(attn_weights, query, key, beta=0, alpha=scale_factor).view(
|
369 |
-
batch_size, self.num_heads, query_length, key_length
|
370 |
-
)
|
371 |
-
|
372 |
-
# ==========================================================================================
|
373 |
-
# attn_weights -> (batch_size, num_heads, query_length, key_length)
|
374 |
-
# ==========================================================================================
|
375 |
-
|
376 |
-
attn_weights = attn_weights.to(softmax_dtype)
|
377 |
-
|
378 |
-
if attention_mask is not None:
|
379 |
-
attn_weights = attn_weights + attention_mask
|
380 |
-
|
381 |
-
attn_weights = F.softmax(attn_weights, dim=-1).to(dtype)
|
382 |
-
|
383 |
-
attn_weights = self.attn_dropout(attn_weights)
|
384 |
-
|
385 |
-
# ==========================================================================================
|
386 |
-
# value -> (batch_size, num_heads, key_length, head_dim)
|
387 |
-
# attn_weights -> (batch_size, num_heads, query_length, key_length)
|
388 |
-
# ==========================================================================================
|
389 |
-
|
390 |
-
attn_output = torch.matmul(attn_weights, value)
|
391 |
-
|
392 |
-
# ==========================================================================================
|
393 |
-
# attn_output -> (batch_size, num_heads, query_length, head_dim)
|
394 |
-
# ==========================================================================================
|
395 |
-
|
396 |
-
attn_output = attn_output.transpose(1, 2)
|
397 |
-
attn_output = attn_output.reshape(batch_size, -1, self.num_heads * self.head_dim)
|
398 |
-
|
399 |
-
# ==========================================================================================
|
400 |
-
# attn_output -> (batch_size, query_length, num_heads * head_dim)
|
401 |
-
# ==========================================================================================
|
402 |
-
|
403 |
-
attn_output = self.c_proj(attn_output)
|
404 |
-
attn_output = self.resid_dropout(attn_output)
|
405 |
-
|
406 |
-
return attn_output
|
407 |
-
|
408 |
-
|
409 |
-
class GraniteSDPA(GraniteAttention):
|
410 |
-
def forward(
|
411 |
-
self,
|
412 |
-
hidden_states: torch.Tensor,
|
413 |
-
past_key_values: Optional[DynamicCache] = None,
|
414 |
-
attention_mask: Optional[torch.Tensor] = None,
|
415 |
-
rope_cos_sin: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
|
416 |
-
) -> torch.Tensor:
|
417 |
-
# ==========================================================================================
|
418 |
-
# hidden_states -> (batch_size, query_length, num_heads * head_dim)
|
419 |
-
# ==========================================================================================
|
420 |
-
|
421 |
-
query, key, value = self._prepare_qkv_for_forward(hidden_states)
|
422 |
-
|
423 |
-
# ==========================================================================================
|
424 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
425 |
-
# key -> (batch_size, num_key_value_heads, query_length, head_dim)
|
426 |
-
# value -> (batch_size, num_key_value_heads, query_length, head_dim)
|
427 |
-
# ==========================================================================================
|
428 |
-
|
429 |
-
if self.position_embedding_type == PositionEmbeddingType.rope:
|
430 |
-
query = apply_rotary_pos_emb(query, rope_cos_sin)
|
431 |
-
key = apply_rotary_pos_emb(key, rope_cos_sin)
|
432 |
-
|
433 |
-
if past_key_values is not None:
|
434 |
-
key, value = past_key_values.update(key, value, self.layer_idx)
|
435 |
-
|
436 |
-
# ==========================================================================================
|
437 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
438 |
-
# key -> (batch_size, num_key_value_heads, key_length, head_dim)
|
439 |
-
# value -> (batch_size, num_key_value_heads, key_length, head_dim)
|
440 |
-
# ==========================================================================================
|
441 |
-
|
442 |
-
key = repeat_key_value(key, self.num_heads, self.num_key_value_heads)
|
443 |
-
value = repeat_key_value(value, self.num_heads, self.num_key_value_heads)
|
444 |
-
|
445 |
-
# ==========================================================================================
|
446 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
447 |
-
# key -> (batch_size, num_heads, key_length, head_dim)
|
448 |
-
# value -> (batch_size, num_heads, key_length, head_dim)
|
449 |
-
# ==========================================================================================
|
450 |
-
|
451 |
-
attn_output = F.scaled_dot_product_attention(
|
452 |
-
query,
|
453 |
-
key,
|
454 |
-
value,
|
455 |
-
attn_mask=attention_mask,
|
456 |
-
dropout_p=self.attn_pdrop if self.training else 0,
|
457 |
-
is_causal=self.causal if attention_mask is None else False,
|
458 |
-
scale=self.attention_multiplier if self.scale_attn_weights else 1,
|
459 |
-
)
|
460 |
-
|
461 |
-
# ==========================================================================================
|
462 |
-
# attn_output -> (batch_size, num_heads, query_length, head_dim)
|
463 |
-
# ==========================================================================================
|
464 |
-
|
465 |
-
batch_size = attn_output.shape[0]
|
466 |
-
attn_output = attn_output.transpose(1, 2)
|
467 |
-
attn_output = attn_output.reshape(batch_size, -1, self.num_heads * self.head_dim)
|
468 |
-
|
469 |
-
# ==========================================================================================
|
470 |
-
# attn_output -> (batch_size, query_length, num_heads * head_dim)
|
471 |
-
# ==========================================================================================
|
472 |
-
|
473 |
-
attn_output = self.c_proj(attn_output)
|
474 |
-
attn_output = self.resid_dropout(attn_output)
|
475 |
-
|
476 |
-
return attn_output
|
477 |
-
|
478 |
-
|
479 |
-
class GraniteFlashAttention2(GraniteAttention):
|
480 |
-
def forward(
|
481 |
-
self,
|
482 |
-
hidden_states: torch.Tensor,
|
483 |
-
past_key_values: Optional[DynamicCache] = None,
|
484 |
-
attention_mask: Optional[torch.Tensor] = None,
|
485 |
-
rope_cos_sin: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
|
486 |
-
) -> torch.Tensor:
|
487 |
-
# ==========================================================================================
|
488 |
-
# hidden_states -> (batch_size, query_length, num_heads * head_dim)
|
489 |
-
# ==========================================================================================
|
490 |
-
|
491 |
-
query, key, value = self._prepare_qkv_for_forward(hidden_states)
|
492 |
-
|
493 |
-
# ==========================================================================================
|
494 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
495 |
-
# key -> (batch_size, num_key_value_heads, query_length, head_dim)
|
496 |
-
# value -> (batch_size, num_key_value_heads, query_length, head_dim)
|
497 |
-
# ==========================================================================================
|
498 |
-
|
499 |
-
if self.position_embedding_type == PositionEmbeddingType.rope:
|
500 |
-
query = apply_rotary_pos_emb(query, rope_cos_sin)
|
501 |
-
key = apply_rotary_pos_emb(key, rope_cos_sin)
|
502 |
-
|
503 |
-
if past_key_values is not None:
|
504 |
-
key, value = past_key_values.update(key, value, self.layer_idx)
|
505 |
-
|
506 |
-
# ==========================================================================================
|
507 |
-
# query -> (batch_size, num_heads, query_length, head_dim)
|
508 |
-
# key -> (batch_size, num_key_value_heads, key_length, head_dim)
|
509 |
-
# value -> (batch_size, num_key_value_heads, key_length, head_dim)
|
510 |
-
# ==========================================================================================
|
511 |
-
|
512 |
-
# TODO avoid this extra transpose
|
513 |
-
query = query.transpose(1, 2)
|
514 |
-
if self.attention_head_type == AttentionHeadType.mqa:
|
515 |
-
key = key.squeeze(1).unsqueeze(2)
|
516 |
-
value = value.squeeze(1).unsqueeze(2)
|
517 |
-
else:
|
518 |
-
key = key.transpose(1, 2)
|
519 |
-
value = value.transpose(1, 2)
|
520 |
-
|
521 |
-
# ==========================================================================================
|
522 |
-
# query -> (batch_size, query_length, num_heads, head_dim)
|
523 |
-
# key -> (batch_size, key_length, num_heads, head_dim)
|
524 |
-
# value -> (batch_size, key_length, num_heads, head_dim)
|
525 |
-
# ==========================================================================================
|
526 |
-
|
527 |
-
batch_size, query_length = query.shape[:2]
|
528 |
-
key_length = key.shape[1]
|
529 |
-
indices_k, cu_seqlens_k, max_seqlen_k = get_unpad_data(attention_mask)
|
530 |
-
|
531 |
-
key = IndexFirstAxis.apply(
|
532 |
-
key.reshape(batch_size * key_length, self.num_key_value_heads, self.head_dim), indices_k
|
533 |
-
)
|
534 |
-
value = IndexFirstAxis.apply(
|
535 |
-
value.reshape(batch_size * key_length, self.num_key_value_heads, self.head_dim), indices_k
|
536 |
-
)
|
537 |
-
|
538 |
-
if query_length == key_length:
|
539 |
-
query = IndexFirstAxis.apply(
|
540 |
-
query.reshape(batch_size * key_length, self.num_heads, self.head_dim), indices_k
|
541 |
-
)
|
542 |
-
cu_seqlens_q = cu_seqlens_k
|
543 |
-
max_seqlen_q = max_seqlen_k
|
544 |
-
indices_q = indices_k
|
545 |
-
elif query_length == 1:
|
546 |
-
max_seqlen_q = 1
|
547 |
-
cu_seqlens_q = torch.arange(
|
548 |
-
batch_size + 1, dtype=torch.int32, device=query.device
|
549 |
-
) # There is a memcpy here, that is very bad.
|
550 |
-
indices_q = cu_seqlens_q[:-1]
|
551 |
-
query = query.squeeze(1)
|
552 |
-
else:
|
553 |
-
# The -q_len: slice assumes left padding.
|
554 |
-
attention_mask = attention_mask[:, -query_length:]
|
555 |
-
query, indices_q, cu_seqlens_q, max_seqlen_q = unpad_input(query, attention_mask)
|
556 |
-
|
557 |
-
# ==========================================================================================
|
558 |
-
# query -> (total_q, num_heads, head_dim)
|
559 |
-
# key -> (total_q, num_heads, head_dim)
|
560 |
-
# value -> (total_q, num_heads, head_dim)
|
561 |
-
# ==========================================================================================
|
562 |
-
|
563 |
-
attn_output = flash_attn_varlen_func(
|
564 |
-
query,
|
565 |
-
key,
|
566 |
-
value,
|
567 |
-
cu_seqlens_q=cu_seqlens_q,
|
568 |
-
cu_seqlens_k=cu_seqlens_k,
|
569 |
-
max_seqlen_q=max_seqlen_q,
|
570 |
-
max_seqlen_k=max_seqlen_k,
|
571 |
-
dropout_p=self.attn_pdrop if self.training else 0,
|
572 |
-
softmax_scale=self.attention_multiplier if self.scale_attn_weights else 1,
|
573 |
-
causal=self.causal,
|
574 |
-
)
|
575 |
-
|
576 |
-
# ==========================================================================================
|
577 |
-
# attn_output -> (total_q, num_heads, head_dim)
|
578 |
-
# ==========================================================================================
|
579 |
-
|
580 |
-
attn_output = pad_input(attn_output, indices_q, batch_size, query_length)
|
581 |
-
attn_output = attn_output.view(batch_size, query_length, -1)
|
582 |
-
|
583 |
-
# ==========================================================================================
|
584 |
-
# attn_output -> (batch_size, query_length, num_heads * head_dim)
|
585 |
-
# ==========================================================================================
|
586 |
-
|
587 |
-
attn_output = self.c_proj(attn_output)
|
588 |
-
attn_output = self.resid_dropout(attn_output)
|
589 |
-
|
590 |
-
return attn_output
|
591 |
-
|
592 |
-
|
593 |
-
_ATTENTION_MODULES = {
|
594 |
-
"eager": GraniteAttention,
|
595 |
-
"sdpa": GraniteSDPA,
|
596 |
-
"flash_attention_2": GraniteFlashAttention2,
|
597 |
-
}
|
598 |
-
|
599 |
-
|
600 |
-
def get_attention_module(
|
601 |
-
config: GraniteConfig, causal: bool, attention_implementation: str, layer_idx: int
|
602 |
-
) -> GraniteAttention:
|
603 |
-
if attention_implementation in _ATTENTION_MODULES:
|
604 |
-
return _ATTENTION_MODULES[attention_implementation](config, causal=causal, layer_idx=layer_idx)
|
605 |
-
raise ValueError(f"unexpected `attention_implementation` {attention_implementation}")
|
606 |
-
|
607 |
-
|
608 |
-
##################################################
|
609 |
-
# position embeddings
|
610 |
-
|
611 |
-
|
612 |
-
class Alibi(nn.Module):
|
613 |
-
def __init__(self, num_heads: int) -> None:
|
614 |
-
super().__init__()
|
615 |
-
self.num_heads = num_heads
|
616 |
-
|
617 |
-
self.reset_parameters()
|
618 |
-
|
619 |
-
def forward(
|
620 |
-
self, attention_mask: torch.Tensor, batch_size: int, key_length: int, device: torch.device, dtype: torch.dtype
|
621 |
-
) -> torch.Tensor:
|
622 |
-
"""
|
623 |
-
Link to paper: https://arxiv.org/abs/2108.12409 Alibi tensor is not causal as the original paper mentions, it
|
624 |
-
relies on a translation invariance of softmax for quick implementation: with l being a tensor, and a fixed value
|
625 |
-
`softmax(l+a) = softmax(l)`. Based on
|
626 |
-
https://github.com/ofirpress/attention_with_linear_biases/blob/a35aaca144e0eb6b789dfcb46784c4b8e31b7983/fairseq/models/transformer.py#L742
|
627 |
-
TODO @thomasw21 this doesn't work as nicely due to the masking strategy, and so masking varies slightly.
|
628 |
-
|
629 |
-
Args:
|
630 |
-
attention_mask (torch.Tensor): attention_mask tensor of shape (`batch_size`, `key_length`)
|
631 |
-
num_heads (int): `num_heads` for the model
|
632 |
-
batch_size (int): `batch_size`
|
633 |
-
key_length (int): `key_length`
|
634 |
-
device (torch.device): device for the tensors
|
635 |
-
dtype (torch.dtype): dtype to use for the tensors
|
636 |
-
|
637 |
-
Returns:
|
638 |
-
torch.Tensor: alibi tensor of shape (`batch_size`, `num_heads`, `key_length`)
|
639 |
-
"""
|
640 |
-
|
641 |
-
# Note: alibi will added to the attention bias that will be applied to the query, key product of attention
|
642 |
-
# => therefore alibi will have to be of shape (batch_size, num_heads, query_length, key_length)
|
643 |
-
# => here we set (batch_size=1, num_heads=num_heads, query_length=1, key_length=max_length)
|
644 |
-
# => the query_length dimension will then be broadcasted correctly
|
645 |
-
# This is more or less identical to T5's relative position bias:
|
646 |
-
# https://github.com/huggingface/transformers/blob/f681437203baa7671de3174b0fa583c349d9d5e1/src/transformers/models/t5/modeling_t5.py#L527
|
647 |
-
if attention_mask is None:
|
648 |
-
arange_tensor = (
|
649 |
-
torch.arange(key_length, device=device).unsqueeze(0).unsqueeze(0).expand(batch_size, -1, -1)
|
650 |
-
)
|
651 |
-
else:
|
652 |
-
arange_tensor = (attention_mask.cumsum(dim=-1) - 1).masked_fill_(attention_mask == 0, 0).unsqueeze(1)
|
653 |
-
|
654 |
-
alibi = self.slopes.unsqueeze(1) * arange_tensor
|
655 |
-
return alibi.to(dtype)
|
656 |
-
|
657 |
-
def reset_parameters(self) -> None:
|
658 |
-
closest_power_of_2 = 2 ** math.floor(math.log2(self.num_heads))
|
659 |
-
base = torch.tensor(2 ** (-(2 ** -(math.log2(closest_power_of_2) - 3))), dtype=torch.float32)
|
660 |
-
powers = torch.arange(1, 1 + closest_power_of_2, dtype=torch.int32)
|
661 |
-
slopes = torch.pow(base, powers)
|
662 |
-
|
663 |
-
if closest_power_of_2 != self.num_heads:
|
664 |
-
extra_base = torch.tensor(2 ** (-(2 ** -(math.log2(2 * closest_power_of_2) - 3))), dtype=torch.float32)
|
665 |
-
num_remaining_heads = min(closest_power_of_2, self.num_heads - closest_power_of_2)
|
666 |
-
extra_powers = torch.arange(1, 1 + 2 * num_remaining_heads, 2, dtype=torch.int32)
|
667 |
-
slopes = torch.cat([slopes, torch.pow(extra_base, extra_powers)], dim=0)
|
668 |
-
|
669 |
-
self.register_buffer("slopes", slopes, persistent=False)
|
670 |
-
|
671 |
-
|
672 |
-
class RoPE(nn.Module):
|
673 |
-
def __init__(
|
674 |
-
self,
|
675 |
-
head_dim: int,
|
676 |
-
max_position_embeddings: int = 2048,
|
677 |
-
base: int = 10000,
|
678 |
-
) -> None:
|
679 |
-
super().__init__()
|
680 |
-
|
681 |
-
self.head_dim = head_dim
|
682 |
-
self.max_position_embeddings = max_position_embeddings
|
683 |
-
self.base = base
|
684 |
-
self.mscale = 1
|
685 |
-
|
686 |
-
self.reset_parameters()
|
687 |
-
|
688 |
-
def forward(self, seq_len: int, dtype: torch.dtype, device: torch.device) -> Tuple[torch.Tensor, torch.Tensor]:
|
689 |
-
if seq_len > self.max_seq_len_cached:
|
690 |
-
self._set_cos_sin_cache(seq_len=seq_len, device=device, dtype=dtype)
|
691 |
-
|
692 |
-
cos = self.cos_cached[:seq_len].to(dtype)
|
693 |
-
sin = self.sin_cached[:seq_len].to(dtype)
|
694 |
-
|
695 |
-
return cos, sin
|
696 |
-
|
697 |
-
def reset_parameters(self) -> None:
|
698 |
-
inv_freq = 1.0 / (self.base ** (torch.arange(0, self.head_dim, 2).float() / self.head_dim))
|
699 |
-
self.register_buffer("inv_freq", inv_freq, persistent=False)
|
700 |
-
|
701 |
-
# Build here to make `torch.jit.trace` work.
|
702 |
-
self._set_cos_sin_cache(
|
703 |
-
seq_len=self.max_position_embeddings, device=self.inv_freq.device, dtype=torch.get_default_dtype()
|
704 |
-
)
|
705 |
-
|
706 |
-
@torch.no_grad()
|
707 |
-
def _set_cos_sin_cache(self, seq_len: int, device: torch.device, dtype: torch.dtype) -> None:
|
708 |
-
self.max_seq_len_cached = seq_len
|
709 |
-
t = torch.arange(self.max_seq_len_cached, device=device, dtype=self.inv_freq.dtype)
|
710 |
-
|
711 |
-
freqs = torch.outer(t, self.inv_freq)
|
712 |
-
# Different from paper, but it uses a different permutation in order to obtain the same calculation
|
713 |
-
emb = torch.cat((freqs, freqs), dim=-1)
|
714 |
-
|
715 |
-
self.register_buffer("cos_cached", (emb.cos() * self.mscale).to(dtype), persistent=False)
|
716 |
-
self.register_buffer("sin_cached", (emb.sin() * self.mscale).to(dtype), persistent=False)
|
717 |
-
|
718 |
-
|
719 |
-
def apply_rotary_pos_emb(x: torch.Tensor, cos_sin: Tuple[torch.Tensor, torch.Tensor]) -> torch.Tensor:
|
720 |
-
cos, sin = cos_sin
|
721 |
-
x = (x * cos) + (_rotate_half(x) * sin)
|
722 |
-
return x
|
723 |
-
|
724 |
-
|
725 |
-
def _rotate_half(x: torch.Tensor) -> torch.Tensor:
|
726 |
-
x1, x2 = torch.chunk(x, 2, dim=-1)
|
727 |
-
return torch.cat((-x2, x1), dim=-1)
|
728 |
-
|
729 |
-
|
730 |
-
##################################################
|
731 |
-
# MLP
|
732 |
-
|
733 |
-
|
734 |
-
class GraniteMLP(nn.Module):
|
735 |
-
def __init__(self, config: GraniteConfig) -> None:
|
736 |
-
super().__init__()
|
737 |
-
|
738 |
-
hidden_size = config.n_embd
|
739 |
-
intermediate_size = config.n_inner
|
740 |
-
activation_function = config.activation_function
|
741 |
-
add_bias = config.add_bias
|
742 |
-
residual_dropout = config.resid_pdrop
|
743 |
-
|
744 |
-
self.c_fc = nn.Linear(
|
745 |
-
hidden_size,
|
746 |
-
2 * intermediate_size if is_glu(activation_function) else intermediate_size,
|
747 |
-
bias=add_bias,
|
748 |
-
)
|
749 |
-
self.act = get_activation_function(activation_function)
|
750 |
-
self.c_proj = nn.Linear(intermediate_size, hidden_size, bias=add_bias)
|
751 |
-
self.dropout = nn.Identity() if residual_dropout == 0 else nn.Dropout(residual_dropout)
|
752 |
-
|
753 |
-
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
|
754 |
-
hidden_states = self.c_fc(hidden_states)
|
755 |
-
hidden_states = self.act(hidden_states)
|
756 |
-
hidden_states = self.c_proj(hidden_states)
|
757 |
-
hidden_states = self.dropout(hidden_states)
|
758 |
-
return hidden_states
|
759 |
-
|
760 |
-
|
761 |
-
##################################################
|
762 |
-
# transformer layer
|
763 |
-
|
764 |
-
|
765 |
-
class GraniteBlock(nn.Module):
|
766 |
-
def __init__(
|
767 |
-
self,
|
768 |
-
config: GraniteConfig,
|
769 |
-
attention_implementation: str,
|
770 |
-
layer_idx: Optional[int] = None,
|
771 |
-
) -> None:
|
772 |
-
super().__init__()
|
773 |
-
|
774 |
-
hidden_size = config.hidden_size
|
775 |
-
self.inner_dim = config.n_inner
|
776 |
-
self.layer_idx = layer_idx
|
777 |
-
|
778 |
-
self.ln_1 = get_normalization_function(
|
779 |
-
config.normalization_function,
|
780 |
-
hidden_size,
|
781 |
-
eps=config.layer_norm_epsilon,
|
782 |
-
)
|
783 |
-
self.attn = get_attention_module(config, True, attention_implementation, layer_idx)
|
784 |
-
self.ln_2 = get_normalization_function(
|
785 |
-
config.normalization_function,
|
786 |
-
hidden_size,
|
787 |
-
eps=config.layer_norm_epsilon,
|
788 |
-
)
|
789 |
-
self.mlp = GraniteMLP(config)
|
790 |
-
|
791 |
-
def forward(
|
792 |
-
self,
|
793 |
-
hidden_states: torch.Tensor,
|
794 |
-
past_key_values: Optional[DynamicCache] = None,
|
795 |
-
attention_mask: Optional[torch.Tensor] = None,
|
796 |
-
rope_cos_sin: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
|
797 |
-
) -> torch.Tensor:
|
798 |
-
residual = hidden_states
|
799 |
-
hidden_states = self.ln_1(hidden_states)
|
800 |
-
|
801 |
-
attn_output = self.attn(
|
802 |
-
hidden_states,
|
803 |
-
past_key_values=past_key_values,
|
804 |
-
attention_mask=attention_mask,
|
805 |
-
rope_cos_sin=rope_cos_sin,
|
806 |
-
)
|
807 |
-
|
808 |
-
# residual connection
|
809 |
-
hidden_states = attn_output + residual
|
810 |
-
|
811 |
-
residual = hidden_states
|
812 |
-
hidden_states = self.ln_2(hidden_states)
|
813 |
-
|
814 |
-
feed_forward_hidden_states = self.mlp(hidden_states)
|
815 |
-
|
816 |
-
# residual connection
|
817 |
-
hidden_states = residual + feed_forward_hidden_states
|
818 |
-
|
819 |
-
return hidden_states
|
820 |
-
|
821 |
-
|
822 |
-
##################################################
|
823 |
-
# model classes
|
824 |
-
|
825 |
-
|
826 |
-
class GranitePreTrainedModel(PreTrainedModel):
|
827 |
-
"""
|
828 |
-
An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
|
829 |
-
models.
|
830 |
-
"""
|
831 |
-
|
832 |
-
config_class = GraniteConfig
|
833 |
-
base_model_prefix = "transformer"
|
834 |
-
causal = True
|
835 |
-
_no_split_modules = ["GraniteBlock"]
|
836 |
-
_skip_keys_device_placement = "past_key_values"
|
837 |
-
_supports_sdpa = True
|
838 |
-
_supports_flash_attn_2 = True
|
839 |
-
|
840 |
-
def __init__(self, config: GraniteConfig, *inputs, **kwargs):
|
841 |
-
super().__init__(config, *inputs, **kwargs)
|
842 |
-
|
843 |
-
self.attention_implementation = self.config._attn_implementation
|
844 |
-
self._use_eager_attention = self.attention_implementation == "eager"
|
845 |
-
self._use_sdpa = self.attention_implementation == "sdpa"
|
846 |
-
self._use_flash_attention_2 = self.attention_implementation == "flash_attention_2"
|
847 |
-
|
848 |
-
self.initializer_range = config.initializer_range
|
849 |
-
|
850 |
-
def _init_weights(self, module: nn.Module) -> None:
|
851 |
-
if isinstance(module, (nn.LayerNorm, RMSNorm, Alibi, RoPE)):
|
852 |
-
module.reset_parameters()
|
853 |
-
elif isinstance(module, nn.Linear):
|
854 |
-
nn.init.normal_(module.weight, mean=0, std=self.initializer_range)
|
855 |
-
if module.bias is not None:
|
856 |
-
module.bias.zero_()
|
857 |
-
elif isinstance(module, nn.Embedding):
|
858 |
-
nn.init.normal_(module.weight, mean=0, std=self.initializer_range)
|
859 |
-
if module.padding_idx is not None:
|
860 |
-
module.weight[module.padding_idx].zero_()
|
861 |
-
|
862 |
-
|
863 |
-
class GraniteModel(GranitePreTrainedModel):
|
864 |
-
_keys_to_ignore_on_load_missing = ["attn.masked_bias"]
|
865 |
-
mask_value = None
|
866 |
-
|
867 |
-
def __init__(self, config: GraniteConfig, **kwargs) -> None:
|
868 |
-
super().__init__(config, **kwargs)
|
869 |
-
|
870 |
-
self.attention_head_type = AttentionHeadType(config.attention_head_type)
|
871 |
-
self.embed_dim = config.hidden_size
|
872 |
-
self.num_heads = config.num_attention_heads
|
873 |
-
self.num_key_value_heads = config.num_key_value_heads
|
874 |
-
|
875 |
-
assert (
|
876 |
-
self.embed_dim % self.num_heads == 0
|
877 |
-
), f"`embed_dim` ({self.embed_dim}) must be divisible by `num_heads` ({self.num_heads})"
|
878 |
-
|
879 |
-
self.head_dim = self.embed_dim // self.num_heads
|
880 |
-
|
881 |
-
self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
|
882 |
-
|
883 |
-
self.drop = nn.Identity() if config.embd_pdrop == 0 else nn.Dropout(config.embd_pdrop)
|
884 |
-
self.h = nn.ModuleList(
|
885 |
-
[GraniteBlock(config, self.attention_implementation, layer_idx=i) for i in range(config.num_hidden_layers)]
|
886 |
-
)
|
887 |
-
self.ln_f = get_normalization_function(
|
888 |
-
config.normalization_function,
|
889 |
-
self.embed_dim,
|
890 |
-
eps=config.layer_norm_epsilon,
|
891 |
-
)
|
892 |
-
|
893 |
-
self.position_embedding_type = PositionEmbeddingType(config.position_embedding_type)
|
894 |
-
|
895 |
-
if self.position_embedding_type == PositionEmbeddingType.learned_absolute:
|
896 |
-
self.wpe = nn.Embedding(config.n_positions, self.embed_dim)
|
897 |
-
elif self.position_embedding_type == PositionEmbeddingType.alibi:
|
898 |
-
assert not self._use_flash_attention_2, "alibi is not implemented with FlashAttention"
|
899 |
-
|
900 |
-
self.alibi = Alibi(self.num_heads)
|
901 |
-
elif self.position_embedding_type == PositionEmbeddingType.rope:
|
902 |
-
self.rope = RoPE(self.head_dim, max_position_embeddings=config.n_positions, base=config.rope_theta)
|
903 |
-
else:
|
904 |
-
raise NotImplementedError()
|
905 |
-
|
906 |
-
# Initialize weights and apply final processing
|
907 |
-
self.post_init()
|
908 |
-
|
909 |
-
def get_input_embeddings(self) -> nn.Embedding:
|
910 |
-
return self.wte
|
911 |
-
|
912 |
-
def set_input_embeddings(self, new_embeddings: nn.Embedding) -> None:
|
913 |
-
self.wte = new_embeddings
|
914 |
-
|
915 |
-
def forward(
|
916 |
-
self,
|
917 |
-
input_ids: Optional[torch.Tensor] = None,
|
918 |
-
past_key_values: Optional[DynamicCache] = None,
|
919 |
-
attention_mask: Optional[torch.Tensor] = None,
|
920 |
-
token_type_ids: Optional[torch.Tensor] = None,
|
921 |
-
position_ids: Optional[torch.Tensor] = None,
|
922 |
-
inputs_embeds: Optional[torch.Tensor] = None,
|
923 |
-
use_cache: Optional[bool] = None,
|
924 |
-
output_hidden_states: Optional[bool] = None,
|
925 |
-
return_dict: Optional[bool] = None,
|
926 |
-
) -> Union[Tuple, BaseModelOutputWithPastAndCrossAttentions]:
|
927 |
-
(
|
928 |
-
output_hidden_states,
|
929 |
-
use_cache,
|
930 |
-
return_dict,
|
931 |
-
input_shape,
|
932 |
-
hidden_states,
|
933 |
-
attention_mask,
|
934 |
-
position_ids,
|
935 |
-
rope_cos_sin,
|
936 |
-
past_key_values,
|
937 |
-
) = self._prepare_a_bunch_of_stuff(
|
938 |
-
input_ids=input_ids,
|
939 |
-
past_key_values=past_key_values,
|
940 |
-
attention_mask=attention_mask,
|
941 |
-
token_type_ids=token_type_ids,
|
942 |
-
position_ids=position_ids,
|
943 |
-
inputs_embeds=inputs_embeds,
|
944 |
-
use_cache=use_cache,
|
945 |
-
output_hidden_states=output_hidden_states,
|
946 |
-
return_dict=return_dict,
|
947 |
-
)
|
948 |
-
|
949 |
-
# ==========================================================================================
|
950 |
-
# flash:
|
951 |
-
# attention_mask -> (batch_size, key_length)
|
952 |
-
# else:
|
953 |
-
# attention_mask -> (batch_size, 1, query_length, key_length)
|
954 |
-
# ==========================================================================================
|
955 |
-
|
956 |
-
output_shape = input_shape + (hidden_states.size(-1),)
|
957 |
-
|
958 |
-
past_key_values = DynamicCache() if use_cache and past_key_values is None else past_key_values
|
959 |
-
all_hidden_states = () if output_hidden_states else None
|
960 |
-
for block in self.h:
|
961 |
-
if output_hidden_states:
|
962 |
-
all_hidden_states += (hidden_states,)
|
963 |
-
|
964 |
-
hidden_states = block(
|
965 |
-
hidden_states,
|
966 |
-
past_key_values=past_key_values,
|
967 |
-
attention_mask=attention_mask,
|
968 |
-
rope_cos_sin=rope_cos_sin,
|
969 |
-
)
|
970 |
-
|
971 |
-
hidden_states = self.ln_f(hidden_states)
|
972 |
-
|
973 |
-
hidden_states = hidden_states.view(output_shape)
|
974 |
-
# Add last hidden state
|
975 |
-
if output_hidden_states:
|
976 |
-
all_hidden_states += (hidden_states,)
|
977 |
-
|
978 |
-
if not return_dict:
|
979 |
-
return tuple(v for v in [hidden_states, past_key_values, all_hidden_states] if v is not None)
|
980 |
-
|
981 |
-
return BaseModelOutputWithPastAndCrossAttentions(
|
982 |
-
last_hidden_state=hidden_states,
|
983 |
-
past_key_values=past_key_values,
|
984 |
-
hidden_states=all_hidden_states,
|
985 |
-
)
|
986 |
-
|
987 |
-
def _get_position_ids(
|
988 |
-
self, attention_mask: torch.Tensor, past_length: int, query_length: int, key_length: int, device: torch.device
|
989 |
-
) -> torch.Tensor:
|
990 |
-
if attention_mask is not None and len(attention_mask.shape) == 2:
|
991 |
-
# create position_ids on the fly for batch generation
|
992 |
-
position_ids = attention_mask.long().cumsum(-1) - 1
|
993 |
-
position_ids.masked_fill_(attention_mask == 0, 0)
|
994 |
-
if past_length > 0:
|
995 |
-
position_ids = position_ids[:, past_length:key_length:]
|
996 |
-
else:
|
997 |
-
position_ids = torch.arange(past_length, key_length, dtype=torch.long, device=device)
|
998 |
-
position_ids = position_ids.unsqueeze(0).view(-1, query_length)
|
999 |
-
|
1000 |
-
return position_ids
|
1001 |
-
|
1002 |
-
def _get_alibi_bias(
|
1003 |
-
self,
|
1004 |
-
attention_mask: torch.Tensor,
|
1005 |
-
batch_size: int,
|
1006 |
-
query_length: int,
|
1007 |
-
key_length: int,
|
1008 |
-
device: torch.device,
|
1009 |
-
dtype: torch.dtype,
|
1010 |
-
) -> torch.Tensor:
|
1011 |
-
if self.position_embedding_type != PositionEmbeddingType.alibi:
|
1012 |
-
return None
|
1013 |
-
|
1014 |
-
alibi_bias = self.alibi(attention_mask, batch_size, key_length, device, dtype)
|
1015 |
-
|
1016 |
-
# ==========================================================================================
|
1017 |
-
# alibi_bias -> (batch_size, num_heads, key_length)
|
1018 |
-
# ==========================================================================================
|
1019 |
-
|
1020 |
-
alibi_bias = alibi_bias.unsqueeze(2)
|
1021 |
-
if query_length != 1:
|
1022 |
-
alibi_bias = alibi_bias.expand(-1, -1, query_length, -1)
|
1023 |
-
|
1024 |
-
# ==========================================================================================
|
1025 |
-
# alibi_bias -> (batch_size, num_heads, query_length, key_length)
|
1026 |
-
# ==========================================================================================
|
1027 |
-
|
1028 |
-
return alibi_bias
|
1029 |
-
|
1030 |
-
def _get_rope_cos_sin(
|
1031 |
-
self, key_length: int, position_ids: torch.Tensor, dtype: torch.dtype, device: torch.device
|
1032 |
-
) -> Optional[Tuple[torch.Tensor, torch.Tensor]]:
|
1033 |
-
if self.position_embedding_type == PositionEmbeddingType.rope:
|
1034 |
-
cos, sin = self.rope(key_length, dtype=dtype, device=device)
|
1035 |
-
cos = cos[position_ids].unsqueeze(1)
|
1036 |
-
sin = sin[position_ids].unsqueeze(1)
|
1037 |
-
return cos, sin
|
1038 |
-
|
1039 |
-
def _prepare_causal_attention_mask(
|
1040 |
-
self, attention_mask: torch.Tensor, batch_size: int, query_length: int, key_length: int, device: torch.device
|
1041 |
-
) -> torch.Tensor:
|
1042 |
-
past_length = key_length - query_length
|
1043 |
-
|
1044 |
-
# ==========================================================================================
|
1045 |
-
# attention_mask -> (batch_size, key_length)
|
1046 |
-
# ==========================================================================================
|
1047 |
-
|
1048 |
-
if query_length > 1:
|
1049 |
-
# (query_length, key_length)
|
1050 |
-
causal_mask = torch.empty((query_length, key_length), dtype=torch.bool, device=device)
|
1051 |
-
causal_mask[:, past_length:] = torch.tril(
|
1052 |
-
torch.ones(query_length, query_length, dtype=torch.bool, device=device)
|
1053 |
-
)
|
1054 |
-
|
1055 |
-
if past_length > 0:
|
1056 |
-
causal_mask[:, :past_length] = True
|
1057 |
-
|
1058 |
-
# (query_length, key_length) -> (1, query_length, key_length)
|
1059 |
-
causal_mask = causal_mask.unsqueeze(0)
|
1060 |
-
|
1061 |
-
if attention_mask is None:
|
1062 |
-
# (1, query_length, key_length) -> (batch_size, query_length, key_length)
|
1063 |
-
causal_mask = causal_mask.expand(batch_size, -1, -1)
|
1064 |
-
else:
|
1065 |
-
# (1, query_length, key_length) & (batch_size, 1, key_length) -> (batch_size, query_length, key_length)
|
1066 |
-
causal_mask = causal_mask & attention_mask.unsqueeze(1).to(torch.bool)
|
1067 |
-
else:
|
1068 |
-
if attention_mask is None:
|
1069 |
-
# (batch_size, query_length, key_length)
|
1070 |
-
causal_mask = torch.ones(batch_size, query_length, key_length, dtype=torch.bool, device=device)
|
1071 |
-
else:
|
1072 |
-
# (batch_size, query_length, key_length)
|
1073 |
-
causal_mask = attention_mask.unsqueeze(1).to(dtype=torch.bool, device=device)
|
1074 |
-
|
1075 |
-
# ==========================================================================================
|
1076 |
-
# attention_mask -> (batch_size, query_length, key_length)
|
1077 |
-
# ==========================================================================================
|
1078 |
-
|
1079 |
-
causal_mask = causal_mask.unsqueeze(1)
|
1080 |
-
|
1081 |
-
# ==========================================================================================
|
1082 |
-
# attention_mask -> (batch_size, 1, query_length, key_length)
|
1083 |
-
# ==========================================================================================
|
1084 |
-
|
1085 |
-
return causal_mask
|
1086 |
-
|
1087 |
-
def _get_initial_hidden_state(
|
1088 |
-
self,
|
1089 |
-
input_ids: torch.Tensor,
|
1090 |
-
inputs_embeds: torch.Tensor,
|
1091 |
-
position_ids: torch.Tensor,
|
1092 |
-
token_type_ids: torch.Tensor,
|
1093 |
-
) -> torch.Tensor:
|
1094 |
-
if inputs_embeds is None:
|
1095 |
-
inputs_embeds = self.wte(input_ids)
|
1096 |
-
|
1097 |
-
if self.position_embedding_type == PositionEmbeddingType.learned_absolute:
|
1098 |
-
inputs_embeds = inputs_embeds + self.wpe(position_ids)
|
1099 |
-
|
1100 |
-
if token_type_ids is not None:
|
1101 |
-
inputs_embeds = inputs_embeds + self.wte(token_type_ids)
|
1102 |
-
|
1103 |
-
inputs_embeds = self.drop(inputs_embeds)
|
1104 |
-
|
1105 |
-
return inputs_embeds
|
1106 |
-
|
1107 |
-
def _prepare_a_bunch_of_stuff(
|
1108 |
-
self,
|
1109 |
-
input_ids: torch.Tensor,
|
1110 |
-
past_key_values: DynamicCache,
|
1111 |
-
attention_mask: torch.Tensor,
|
1112 |
-
token_type_ids: torch.Tensor,
|
1113 |
-
position_ids: torch.Tensor,
|
1114 |
-
inputs_embeds: torch.Tensor,
|
1115 |
-
use_cache: bool,
|
1116 |
-
output_hidden_states: bool,
|
1117 |
-
return_dict: bool,
|
1118 |
-
) -> Tuple[
|
1119 |
-
bool,
|
1120 |
-
bool,
|
1121 |
-
bool,
|
1122 |
-
torch.Size,
|
1123 |
-
torch.Tensor,
|
1124 |
-
torch.Tensor,
|
1125 |
-
torch.Tensor,
|
1126 |
-
Optional[Tuple[torch.Tensor, torch.Tensor]],
|
1127 |
-
DynamicCache,
|
1128 |
-
]:
|
1129 |
-
output_hidden_states = (
|
1130 |
-
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
1131 |
-
)
|
1132 |
-
|
1133 |
-
use_cache = self.config.use_cache if use_cache is None else use_cache
|
1134 |
-
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
1135 |
-
|
1136 |
-
if input_ids is not None and inputs_embeds is not None:
|
1137 |
-
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
|
1138 |
-
elif input_ids is not None:
|
1139 |
-
input_shape = input_ids.size()
|
1140 |
-
elif inputs_embeds is not None:
|
1141 |
-
# TODO special handling for padding free transformer needed here if we support inputs_embeds argument
|
1142 |
-
input_shape = inputs_embeds.size()[:-1]
|
1143 |
-
else:
|
1144 |
-
raise ValueError("You have to specify either input_ids or inputs_embeds")
|
1145 |
-
|
1146 |
-
batch_size = input_shape[0]
|
1147 |
-
device = input_ids.device if input_ids is not None else inputs_embeds.device
|
1148 |
-
|
1149 |
-
if self.position_embedding_type == PositionEmbeddingType.alibi:
|
1150 |
-
if position_ids is not None:
|
1151 |
-
warnings.warn("`position_ids` have no functionality with Alibi.", FutureWarning)
|
1152 |
-
|
1153 |
-
if token_type_ids is not None:
|
1154 |
-
token_type_ids = token_type_ids.view(-1, input_shape[-1])
|
1155 |
-
|
1156 |
-
# ==========================================================================================
|
1157 |
-
# input_ids -> (batch_size, query_length)
|
1158 |
-
# attention_mask -> None or (batch_size, key_length)
|
1159 |
-
# position_ids -> None or (batch_size, key_length)
|
1160 |
-
# ==========================================================================================
|
1161 |
-
|
1162 |
-
past_length = 0 if past_key_values is None else past_key_values.get_seq_length()
|
1163 |
-
query_length = input_shape[-1]
|
1164 |
-
key_length = past_length + query_length
|
1165 |
-
|
1166 |
-
if position_ids is None:
|
1167 |
-
position_ids = self._get_position_ids(attention_mask, past_length, query_length, key_length, device)
|
1168 |
-
|
1169 |
-
# ==========================================================================================
|
1170 |
-
# input_ids -> (batch_size, query_length)
|
1171 |
-
# attention_mask -> None or (batch_size, key_length)
|
1172 |
-
# position_ids -> (batch_size, query_length)
|
1173 |
-
# ==========================================================================================
|
1174 |
-
|
1175 |
-
hidden_states = self._get_initial_hidden_state(input_ids, inputs_embeds, position_ids, token_type_ids)
|
1176 |
-
|
1177 |
-
# ==========================================================================================
|
1178 |
-
# hidden_states -> (batch_size, query_length, num_heads * head_dim)
|
1179 |
-
# ==========================================================================================
|
1180 |
-
|
1181 |
-
alibi_bias = self._get_alibi_bias(
|
1182 |
-
attention_mask, batch_size, query_length, key_length, device, hidden_states.dtype
|
1183 |
-
)
|
1184 |
-
|
1185 |
-
# ==========================================================================================
|
1186 |
-
# alibi_bias -> (batch_size, num_heads, query_length, key_length)
|
1187 |
-
# ==========================================================================================
|
1188 |
-
|
1189 |
-
rope_cos_sin = self._get_rope_cos_sin(
|
1190 |
-
key_length, position_ids, dtype=hidden_states.dtype, device=hidden_states.device
|
1191 |
-
)
|
1192 |
-
|
1193 |
-
# ==========================================================================================
|
1194 |
-
# rope_cos_sin -> 2 * (key_length, head_dim)
|
1195 |
-
# ==========================================================================================
|
1196 |
-
|
1197 |
-
# prepare causal mask only if not using flash attention
|
1198 |
-
if self._use_flash_attention_2:
|
1199 |
-
if attention_mask is None:
|
1200 |
-
attention_mask = torch.ones_like(input_ids)
|
1201 |
-
elif self._use_sdpa:
|
1202 |
-
# we use the causal/non-causal argument of SDPA for attention in this case
|
1203 |
-
if attention_mask is not None:
|
1204 |
-
attention_mask = self._prepare_causal_attention_mask(
|
1205 |
-
attention_mask, batch_size, query_length, key_length, device
|
1206 |
-
)
|
1207 |
-
|
1208 |
-
attention_mask = torch.where(
|
1209 |
-
attention_mask,
|
1210 |
-
~attention_mask if alibi_bias is None else alibi_bias,
|
1211 |
-
self._get_mask_value(attention_mask.device, hidden_states.dtype),
|
1212 |
-
)
|
1213 |
-
else:
|
1214 |
-
attention_mask = self._prepare_causal_attention_mask(
|
1215 |
-
attention_mask, batch_size, query_length, key_length, device
|
1216 |
-
)
|
1217 |
-
|
1218 |
-
attention_mask = torch.where(
|
1219 |
-
attention_mask,
|
1220 |
-
~attention_mask if alibi_bias is None else alibi_bias,
|
1221 |
-
self._get_mask_value(attention_mask.device, hidden_states.dtype),
|
1222 |
-
)
|
1223 |
-
|
1224 |
-
return (
|
1225 |
-
output_hidden_states,
|
1226 |
-
use_cache,
|
1227 |
-
return_dict,
|
1228 |
-
input_shape,
|
1229 |
-
hidden_states,
|
1230 |
-
attention_mask,
|
1231 |
-
position_ids,
|
1232 |
-
rope_cos_sin,
|
1233 |
-
past_key_values,
|
1234 |
-
)
|
1235 |
-
|
1236 |
-
def _get_mask_value(self, device: torch.device, dtype: torch.dtype) -> torch.Tensor:
|
1237 |
-
# torch.where expects a tensor. We use a cache to avoid recreating it every time.
|
1238 |
-
if self.mask_value is None or self.mask_value.dtype != dtype or self.mask_value.device != device:
|
1239 |
-
self.mask_value = torch.full([], torch.finfo(torch.float16).min, dtype=dtype, device=device)
|
1240 |
-
return self.mask_value
|
1241 |
-
|
1242 |
-
|
1243 |
-
class GraniteForCausalLM(GranitePreTrainedModel):
|
1244 |
-
_keys_to_ignore_on_load_missing = ["lm_head.weight"]
|
1245 |
-
|
1246 |
-
def __init__(self, config: GraniteConfig, **kwargs) -> None:
|
1247 |
-
super().__init__(config, **kwargs)
|
1248 |
-
self.transformer = GraniteModel(config, **kwargs)
|
1249 |
-
self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
|
1250 |
-
|
1251 |
-
# Initialize weights and apply final processing
|
1252 |
-
self.post_init()
|
1253 |
-
|
1254 |
-
def get_input_embeddings(self) -> nn.Embedding:
|
1255 |
-
return self.transformer.wte
|
1256 |
-
|
1257 |
-
def set_input_embeddings(self, value: nn.Embedding) -> None:
|
1258 |
-
self.transformer.wte = value
|
1259 |
-
|
1260 |
-
def get_output_embeddings(self) -> nn.Linear:
|
1261 |
-
return self.lm_head
|
1262 |
-
|
1263 |
-
def set_output_embeddings(self, new_embeddings: nn.Linear) -> None:
|
1264 |
-
self.lm_head = new_embeddings
|
1265 |
-
|
1266 |
-
# FIXME typing
|
1267 |
-
def prepare_inputs_for_generation(
|
1268 |
-
self,
|
1269 |
-
input_ids: torch.Tensor,
|
1270 |
-
past_key_values: Optional[DynamicCache] = None,
|
1271 |
-
inputs_embeds: Optional[torch.Tensor] = None,
|
1272 |
-
**kwargs,
|
1273 |
-
) -> dict:
|
1274 |
-
token_type_ids = kwargs.get("token_type_ids", None)
|
1275 |
-
# Omit tokens covered by past_key_values
|
1276 |
-
if past_key_values:
|
1277 |
-
past_length = past_key_values.get_seq_length()
|
1278 |
-
|
1279 |
-
# Some generation methods already pass only the last input ID
|
1280 |
-
if input_ids.shape[1] > past_length:
|
1281 |
-
remove_prefix_length = past_length
|
1282 |
-
else:
|
1283 |
-
# Default to old behavior: keep only final ID
|
1284 |
-
remove_prefix_length = input_ids.shape[1] - 1
|
1285 |
-
|
1286 |
-
input_ids = input_ids[:, remove_prefix_length:]
|
1287 |
-
if token_type_ids is not None:
|
1288 |
-
token_type_ids = token_type_ids[:, -input_ids.shape[1] :]
|
1289 |
-
|
1290 |
-
attention_mask = kwargs.get("attention_mask", None)
|
1291 |
-
position_ids = kwargs.get("position_ids", None)
|
1292 |
-
|
1293 |
-
if attention_mask is not None and position_ids is None:
|
1294 |
-
# create position_ids on the fly for batch generation
|
1295 |
-
position_ids = attention_mask.long().cumsum(-1) - 1
|
1296 |
-
position_ids.masked_fill_(attention_mask == 0, 0)
|
1297 |
-
if past_key_values:
|
1298 |
-
position_ids = position_ids[:, -input_ids.shape[1] :]
|
1299 |
-
else:
|
1300 |
-
position_ids = None
|
1301 |
-
|
1302 |
-
# if `inputs_embeds` are passed, we only want to use them in the 1st generation step
|
1303 |
-
if inputs_embeds is not None and past_key_values is None:
|
1304 |
-
model_inputs = {"inputs_embeds": inputs_embeds}
|
1305 |
-
else:
|
1306 |
-
model_inputs = {"input_ids": input_ids}
|
1307 |
-
|
1308 |
-
model_inputs.update(
|
1309 |
-
{
|
1310 |
-
"past_key_values": past_key_values,
|
1311 |
-
"use_cache": kwargs.get("use_cache"),
|
1312 |
-
"position_ids": position_ids,
|
1313 |
-
"attention_mask": attention_mask,
|
1314 |
-
"token_type_ids": token_type_ids,
|
1315 |
-
}
|
1316 |
-
)
|
1317 |
-
return model_inputs
|
1318 |
-
|
1319 |
-
def forward(
|
1320 |
-
self,
|
1321 |
-
input_ids: Optional[Union[torch.Tensor]] = None,
|
1322 |
-
past_key_values: Optional[DynamicCache] = None,
|
1323 |
-
attention_mask: Optional[torch.Tensor] = None,
|
1324 |
-
token_type_ids: Optional[Union[torch.Tensor]] = None,
|
1325 |
-
position_ids: Optional[Union[torch.Tensor]] = None,
|
1326 |
-
inputs_embeds: Optional[Union[torch.Tensor]] = None,
|
1327 |
-
labels: Optional[Union[torch.Tensor]] = None,
|
1328 |
-
use_cache: Optional[bool] = None,
|
1329 |
-
output_attentions: Optional[bool] = None,
|
1330 |
-
output_hidden_states: Optional[bool] = None,
|
1331 |
-
return_dict: Optional[bool] = None,
|
1332 |
-
) -> Union[Tuple, CausalLMOutputWithCrossAttentions]:
|
1333 |
-
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
1334 |
-
|
1335 |
-
# ==========================================================================================
|
1336 |
-
# input_ids -> (batch_size, query_length)
|
1337 |
-
# attention_mask -> None or (batch_size, key_length)
|
1338 |
-
# position_ids -> None or (batch_size, key_length)
|
1339 |
-
# ==========================================================================================
|
1340 |
-
|
1341 |
-
transformer_outputs = self.transformer(
|
1342 |
-
input_ids,
|
1343 |
-
past_key_values=past_key_values,
|
1344 |
-
attention_mask=attention_mask,
|
1345 |
-
token_type_ids=token_type_ids,
|
1346 |
-
position_ids=position_ids,
|
1347 |
-
inputs_embeds=inputs_embeds,
|
1348 |
-
use_cache=use_cache,
|
1349 |
-
output_hidden_states=output_hidden_states,
|
1350 |
-
return_dict=return_dict,
|
1351 |
-
)
|
1352 |
-
hidden_states = transformer_outputs[0]
|
1353 |
-
|
1354 |
-
lm_logits = self.lm_head(hidden_states)
|
1355 |
-
|
1356 |
-
loss = None
|
1357 |
-
# Shift so that tokens < n predict n
|
1358 |
-
if labels is not None:
|
1359 |
-
shift_logits = lm_logits[..., :-1, :].contiguous()
|
1360 |
-
shift_labels = labels[..., 1:].contiguous().to(shift_logits.device)
|
1361 |
-
|
1362 |
-
# Flatten the tokens
|
1363 |
-
loss_fct = nn.CrossEntropyLoss()
|
1364 |
-
loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
|
1365 |
-
|
1366 |
-
if not return_dict:
|
1367 |
-
output = (lm_logits,) + transformer_outputs[1:]
|
1368 |
-
return ((loss,) + output) if loss is not None else output
|
1369 |
-
|
1370 |
-
return CausalLMOutputWithCrossAttentions(
|
1371 |
-
loss=loss,
|
1372 |
-
logits=lm_logits,
|
1373 |
-
past_key_values=transformer_outputs.past_key_values,
|
1374 |
-
hidden_states=transformer_outputs.hidden_states,
|
1375 |
-
attentions=transformer_outputs.attentions,
|
1376 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|