Training in progress, epoch 1
Browse files- adapter_config.json +4 -4
- adapter_model.safetensors +1 -1
- tmpepi07lf9/_remote_module_non_scriptable.py +81 -0
- training_args.bin +1 -1
adapter_config.json
CHANGED
@@ -19,13 +19,13 @@
|
|
19 |
"rank_pattern": {},
|
20 |
"revision": null,
|
21 |
"target_modules": [
|
22 |
-
"
|
23 |
"o_proj",
|
24 |
"k_proj",
|
25 |
-
"
|
|
|
26 |
"down_proj",
|
27 |
-
"up_proj"
|
28 |
-
"v_proj"
|
29 |
],
|
30 |
"task_type": "CAUSAL_LM",
|
31 |
"use_rslora": false
|
|
|
19 |
"rank_pattern": {},
|
20 |
"revision": null,
|
21 |
"target_modules": [
|
22 |
+
"q_proj",
|
23 |
"o_proj",
|
24 |
"k_proj",
|
25 |
+
"gate_proj",
|
26 |
+
"v_proj",
|
27 |
"down_proj",
|
28 |
+
"up_proj"
|
|
|
29 |
],
|
30 |
"task_type": "CAUSAL_LM",
|
31 |
"use_rslora": false
|
adapter_model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 1803940752
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:438aee460402ec1fdef6f1376dc770011606bc8eca2a2d2b2ee763ab5a715a2d
|
3 |
size 1803940752
|
tmpepi07lf9/_remote_module_non_scriptable.py
ADDED
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import *
|
2 |
+
|
3 |
+
import torch
|
4 |
+
import torch.distributed.rpc as rpc
|
5 |
+
from torch import Tensor
|
6 |
+
from torch._jit_internal import Future
|
7 |
+
from torch.distributed.rpc import RRef
|
8 |
+
from typing import Tuple # pyre-ignore: unused import
|
9 |
+
|
10 |
+
|
11 |
+
module_interface_cls = None
|
12 |
+
|
13 |
+
|
14 |
+
def forward_async(self, *args, **kwargs):
|
15 |
+
args = (self.module_rref, self.device, self.is_device_map_set, *args)
|
16 |
+
kwargs = {**kwargs}
|
17 |
+
return rpc.rpc_async(
|
18 |
+
self.module_rref.owner(),
|
19 |
+
_remote_forward,
|
20 |
+
args,
|
21 |
+
kwargs,
|
22 |
+
)
|
23 |
+
|
24 |
+
|
25 |
+
def forward(self, *args, **kwargs):
|
26 |
+
args = (self.module_rref, self.device, self.is_device_map_set, *args)
|
27 |
+
kwargs = {**kwargs}
|
28 |
+
ret_fut = rpc.rpc_async(
|
29 |
+
self.module_rref.owner(),
|
30 |
+
_remote_forward,
|
31 |
+
args,
|
32 |
+
kwargs,
|
33 |
+
)
|
34 |
+
return ret_fut.wait()
|
35 |
+
|
36 |
+
|
37 |
+
_generated_methods = [
|
38 |
+
forward_async,
|
39 |
+
forward,
|
40 |
+
]
|
41 |
+
|
42 |
+
|
43 |
+
|
44 |
+
|
45 |
+
def _remote_forward(
|
46 |
+
module_rref: RRef[module_interface_cls], device: str, is_device_map_set: bool, *args, **kwargs):
|
47 |
+
module = module_rref.local_value()
|
48 |
+
device = torch.device(device)
|
49 |
+
|
50 |
+
if device.type != "cuda":
|
51 |
+
return module.forward(*args, **kwargs)
|
52 |
+
|
53 |
+
# If the module is on a cuda device,
|
54 |
+
# move any CPU tensor in args or kwargs to the same cuda device.
|
55 |
+
# Since torch script does not support generator expression,
|
56 |
+
# have to use concatenation instead of
|
57 |
+
# ``tuple(i.to(device) if isinstance(i, Tensor) else i for i in *args)``.
|
58 |
+
args = (*args,)
|
59 |
+
out_args: Tuple[()] = ()
|
60 |
+
for arg in args:
|
61 |
+
arg = (arg.to(device),) if isinstance(arg, Tensor) else (arg,)
|
62 |
+
out_args = out_args + arg
|
63 |
+
|
64 |
+
kwargs = {**kwargs}
|
65 |
+
for k, v in kwargs.items():
|
66 |
+
if isinstance(v, Tensor):
|
67 |
+
kwargs[k] = kwargs[k].to(device)
|
68 |
+
|
69 |
+
if is_device_map_set:
|
70 |
+
return module.forward(*out_args, **kwargs)
|
71 |
+
|
72 |
+
# If the device map is empty, then only CPU tensors are allowed to send over wire,
|
73 |
+
# so have to move any GPU tensor to CPU in the output.
|
74 |
+
# Since torch script does not support generator expression,
|
75 |
+
# have to use concatenation instead of
|
76 |
+
# ``tuple(i.cpu() if isinstance(i, Tensor) else i for i in module.forward(*out_args, **kwargs))``.
|
77 |
+
ret: Tuple[()] = ()
|
78 |
+
for i in module.forward(*out_args, **kwargs):
|
79 |
+
i = (i.cpu(),) if isinstance(i, Tensor) else (i,)
|
80 |
+
ret = ret + i
|
81 |
+
return ret
|
training_args.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 4664
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:dcc62bb1e6ac14e110070d8afe60cdbd0f23c0c49c4036bf2b417084e7372ebd
|
3 |
size 4664
|