rixty/KoboldAI-Client
1
1
Fork 0
mirror of https://github.com/KoboldAI/KoboldAI-Client.git synced 2025-06-05 21:59:24 +02:00
Code Issues Projects Releases Wiki Activity

Merge pull request #385 from one-some/accelerate-offloading

Browse Source 
Accelerate offloading
This commit is contained in:
henk717
2023-07-04 03:14:51 +02:00
committed by GitHub
parent b603690e4c ebe478458f
commit 94c62a4f90
12 changed files with 738 additions and 1941 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
aiserver.py
View File

@@ -1525,6 +1525,7 @@ def general_startup(override_args=None):
print(f"Allowed IPs: {allowed_ips}")
if args.cpu:
os.environ['CUDA_VISIBLE_DEVICES'] = "None"
koboldai_vars.use_colab_tpu = False
koboldai_vars.hascuda = False
koboldai_vars.usegpu = False
@@ -10811,7 +10812,7 @@ def run():
Session(app)
logger.init_ok("Flask", status="OK")
logger.init("Webserver", status="Starting")
patch_transformers()
patch_transformers(use_tpu=koboldai_vars.use_colab_tpu)
# Start Flask/SocketIO (Blocking, so this must be last method!)
port = args.port if "port" in args and args.port is not None else 5000
@@ -10908,7 +10909,7 @@ else:
logger.init("Flask", status="Starting")
Session(app)
logger.init_ok("Flask", status="OK")
patch_transformers()
patch_transformers(use_tpu=koboldai_vars.use_colab_tpu)
startup(command_line_backend)
koboldai_settings.port = args.port if "port" in args and args.port is not None else 5000
print("{0}\nServer started in WSGI mode!{1}".format(colors.GREEN, colors.END), flush=True)

955
breakmodel.py
View File

@@ -1,955 +0,0 @@
'''
This is a MODIFIED version of arrmansa's low VRAM patch.
https://github.com/arrmansa/Basic-UI-for-GPT-J-6B-with-low-vram/blob/main/GPT-J-6B-Low-Vram-UI.ipynb
The ORIGINAL version of the patch is released under the Apache License 2.0
Copyright 2021 arrmansa
Copyright 2021 finetuneanon
Copyright 2018, 2022 The Hugging Face team
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
'''
import torch
from torch import nn
import torch.cuda.comm
import copy
import gc
import os
import sys
import itertools
import bisect
import random
import utils
from typing import Dict, List, Optional, Union
from transformers.modeling_outputs import BaseModelOutputWithPast, BaseModelOutputWithPastAndCrossAttentions
from transformers.utils import logging
logger = logging.get_logger(__name__)
breakmodel = True
gpu_blocks = []
disk_blocks = 0
primary_device = 0 if torch.cuda.device_count() > 0 else "cpu"
from accelerate.hooks import attach_align_device_hook_on_blocks
from accelerate.utils import OffloadedWeightsLoader, check_device_map, extract_submodules_state_dict, offload_state_dict
from accelerate import dispatch_model
def dispatch_model_ex(
model: nn.Module,
device_map: Dict[str, Union[str, int, torch.device]],
main_device: Optional[torch.device] = None,
state_dict: Optional[Dict[str, torch.Tensor]] = None,
offload_dir: Union[str, os.PathLike] = None,
offload_buffers: bool = False,
**kwargs,
):
"""
This is a modified version of
https://github.com/huggingface/accelerate/blob/eeaba598f455fbd2c48661d7e816d3ff25ab050b/src/accelerate/big_modeling.py#L130
that still works when the main device is the CPU.
Dispatches a model according to a given device map. Layers of the model might be spread across GPUs, offloaded on
the CPU or even the disk.
Args:
model (`torch.nn.Module`):
The model to dispatch.
device_map (`Dict[str, Union[str, int, torch.device]]`):
A dictionary mapping module names in the models `state_dict` to the device they should go to. Note that
`"disk"` is accepted even if it's not a proper value for `torch.device`.
main_device (`str`, `int` or `torch.device`, *optional*):
The main execution device. Will default to the first device in the `device_map` different from `"cpu"` or
`"disk"`.
state_dict (`Dict[str, torch.Tensor]`, *optional*):
The state dict of the part of the model that will be kept on CPU.
offload_dir (`str` or `os.PathLike`):
The folder in which to offload the model weights (or where the model weights are already offloaded).
offload_buffers (`bool`, *optional*, defaults to `False`):
Whether or not to offload the buffers with the model parameters.
preload_module_classes (`List[str]`, *optional*):
A list of classes whose instances should load all their weights (even in the submodules) at the beginning
of the forward. This should only be used for classes that have submodules which are registered but not
called directly during the forward, for instance if a `dense` linear layer is registered, but at forward,
`dense.weight` and `dense.bias` are used in some operations instead of calling `dense` directly.
"""
if main_device != "cpu":
return dispatch_model(model, device_map, main_device, state_dict, offload_dir=offload_dir, offload_buffers=offload_buffers, **kwargs)
# Error early if the device map is incomplete.
check_device_map(model, device_map)
offload_devices = ["cpu", "disk"] if main_device != "cpu" else ["disk"]
if main_device is None:
main_device = [d for d in device_map.values() if d not in offload_devices][0]
cpu_modules = [name for name, device in device_map.items() if device == "cpu"] if main_device != "cpu" else []
if state_dict is None and len(cpu_modules) > 0:
state_dict = extract_submodules_state_dict(model.state_dict(), cpu_modules)
disk_modules = [name for name, device in device_map.items() if device == "disk"]
if offload_dir is None and len(disk_modules) > 0:
raise ValueError(
"We need an `offload_dir` to dispatch this model according to this `device_map`, the following submodules "
f"need to be offloaded: {', '.join(disk_modules)}."
)
if len(disk_modules) > 0 and (
not os.path.isdir(offload_dir) or not os.path.isfile(os.path.join(offload_dir, "index.json"))
):
disk_state_dict = extract_submodules_state_dict(model.state_dict(), disk_modules)
offload_state_dict(offload_dir, disk_state_dict)
execution_device = {
name: main_device if device in offload_devices else device for name, device in device_map.items()
}
offload = {name: device in offload_devices for name, device in device_map.items()}
save_folder = offload_dir if len(disk_modules) > 0 else None
if state_dict is not None or save_folder is not None:
weights_map = OffloadedWeightsLoader(state_dict=state_dict, save_folder=save_folder)
else:
weights_map = None
attach_align_device_hook_on_blocks(
model,
execution_device=execution_device,
offload=offload,
offload_buffers=offload_buffers,
weights_map=weights_map,
**kwargs,
)
model.hf_device_map = device_map
return model
# Copied from transformers.models.bart.modeling_bart._expand_mask
def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
"""
Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
"""
bsz, src_len = mask.size()
tgt_len = tgt_len if tgt_len is not None else src_len
expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
inverted_mask = 1.0 - expanded_mask
return inverted_mask.masked_fill(inverted_mask.bool(), torch.finfo(dtype).min)
def move_hidden_layers(transformer, h=None):
if h is None:
h = transformer.h
assert len(gpu_blocks) <= torch.cuda.device_count()
assert sum(gpu_blocks) <= len(h)
ram_blocks = len(h) - sum(gpu_blocks)
transformer.extrastorage = {}
torch.cuda.empty_cache()
able_to_pin_layers = True
for i in range(ram_blocks):
h[i].to("cpu")
transformer.extrastorage[i] = copy.deepcopy(h[i])
smalltensor = torch.tensor(0).to(primary_device)
for param1 in h[i].parameters():
param1.data = smalltensor
h[i].to(primary_device)
for param in transformer.extrastorage[i].parameters():
param.requires_grad = False
param.data = param.data.detach()
if able_to_pin_layers:
try:
param.data = param.data.pin_memory()
except:
able_to_pin_layers = False
print(f"WARNING: You only have enough shared GPU memory for {i} out of {ram_blocks} CPU layers. Expect suboptimal speed.", file=sys.stderr)
gc.collect()
torch.cuda.empty_cache()
if ram_blocks:
for param1,param2 in zip(h[0].parameters(),transformer.extrastorage[0].parameters()):
param1.data = param2.data.to(primary_device, non_blocking=False).detach()
for param1,param2 in zip(h[ram_blocks-1].parameters(),transformer.extrastorage[ram_blocks-1].parameters()):
param1.data = param2.data.to(primary_device, non_blocking=False).detach()
i = ram_blocks
for j in range(len(gpu_blocks)):
for _ in range(gpu_blocks[j]):
h[i].to(j)
i += 1
def new_forward_neo(
self,
input_ids=None,
past_key_values=None,
attention_mask=None,
token_type_ids=None,
position_ids=None,
head_mask=None,
inputs_embeds=None,
use_cache=None,
output_attentions=None,
output_hidden_states=None,
return_dict=None,
embs=None,
):
assert len(gpu_blocks) <= torch.cuda.device_count()
assert sum(gpu_blocks) <= len(self.h)
ram_blocks = len(self.h) - sum(gpu_blocks)
cumulative_gpu_blocks = tuple(itertools.accumulate(gpu_blocks))
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
)
use_cache = use_cache if use_cache is not None else self.config.use_cache
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
if input_ids is not None and inputs_embeds is not None:
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
elif input_ids is not None:
input_shape = input_ids.size()
input_ids = input_ids.view(-1, input_shape[-1])
batch_size = input_ids.shape[0]
elif inputs_embeds is not None:
input_shape = inputs_embeds.size()[:-1]
batch_size = inputs_embeds.shape[0]
else:
raise ValueError("You have to specify either input_ids or inputs_embeds")
device = input_ids.device if input_ids is not None else inputs_embeds.device
if token_type_ids is not None:
token_type_ids = token_type_ids.view(-1, input_shape[-1])
if position_ids is not None:
position_ids = position_ids.view(-1, input_shape[-1])
if past_key_values is None:
past_length = 0
past_key_values = tuple([None] * len(self.h))
else:
past_length = past_key_values[0][0].size(-2)
device = primary_device if breakmodel else input_ids.device if input_ids is not None else inputs_embeds.device
if position_ids is None:
position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
# Attention mask.
if attention_mask is not None:
assert batch_size > 0, "batch_size has to be defined and > 0"
attention_mask = attention_mask.view(batch_size, -1)
# We create a 3D attention mask from a 2D tensor mask.
# Sizes are [batch_size, 1, 1, to_seq_length]
# So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
# this attention mask is more simple than the triangular masking of causal attention
# used in OpenAI GPT, we just need to prepare the broadcast dimension here.
attention_mask = attention_mask[:, None, None, :]
# Since attention_mask is 1.0 for positions we want to attend and 0.0 for
# masked positions, this operation will create a tensor which is 0.0 for
# positions we want to attend and -10000.0 for masked positions.
# Since we are adding it to the raw scores before the softmax, this is
# effectively the same as removing these entirely.
attention_mask = attention_mask.to(dtype=self.dtype) # fp16 compatibility
attention_mask = (1.0 - attention_mask) * -10000.0
# Prepare head mask if needed
# 1.0 in head_mask indicate we keep the head
# attention_probs has shape bsz x num_heads x N x N
# head_mask has shape n_layer x batch x num_heads x N x N
head_mask = self.get_head_mask(head_mask, getattr(self.config, "num_layers", None) or self.config.n_layer)
if inputs_embeds is None:
if breakmodel:
input_ids = input_ids.to(primary_device)
inputs_embeds = self.wte(input_ids)
if embs is not None and not (use_cache is not None and use_cache and past_key_values is not None and len(past_key_values) > 0 and past_key_values[0] is not None):
offset = 0
for pos, emb in embs:
pos += offset
if len(emb.shape) == 2:
emb = emb.repeat(input_shape[0], 1, 1)
inputs_embeds[:, pos:pos+emb.shape[1]] = emb
offset += emb.shape[1]
if getattr(self, "wpe", None) is None:
hidden_states = inputs_embeds
else:
if breakmodel:
position_ids = position_ids.to(primary_device)
position_embeds = self.wpe(position_ids)
if breakmodel:
position_embeds = position_embeds.to(primary_device)
hidden_states = inputs_embeds + position_embeds
if token_type_ids is not None:
token_type_embeds = self.wte(token_type_ids)
hidden_states = hidden_states + token_type_embeds
hidden_states = self.drop(hidden_states)
output_shape = input_shape + (hidden_states.size(-1),)
presents = () if use_cache else None
all_self_attentions = () if output_attentions else None
all_hidden_states = () if output_hidden_states else None
if breakmodel and ram_blocks:
copystream = torch.cuda.Stream(device=primary_device, priority=-1)
for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
if breakmodel:
if i in range(ram_blocks):
index1 = (i+1)%ram_blocks
for param1,param2 in zip(self.h[index1].parameters(),self.h[(i-1)%ram_blocks].parameters()):
param1.data = param2.data
for param1,param2 in zip(self.h[index1].parameters(),self.extrastorage[index1].parameters()):
with torch.cuda.stream(copystream):
torch.cuda.comm.broadcast(param2.data,out = [param1.data])
if output_hidden_states:
all_hidden_states = all_hidden_states + (hidden_states.cpu(),)
if getattr(self.config, "gradient_checkpointing", False) and self.training:
if use_cache:
logger.warning(
"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
)
use_cache = False
def create_custom_forward(module):
def custom_forward(*inputs):
# None for past_key_value
return module(*inputs, use_cache, output_attentions)
return custom_forward
outputs = torch.utils.checkpoint.checkpoint(
create_custom_forward(block),
hidden_states,
None,
attention_mask,
head_mask[i],
)
else:
if breakmodel:
device = primary_device if i < ram_blocks else bisect.bisect_right(cumulative_gpu_blocks, i - ram_blocks)
outputs = block(
hidden_states.to(device) if breakmodel and hidden_states is not None else hidden_states,
layer_past=tuple(v.to(device) for v in layer_past if v is not None) if breakmodel and layer_past is not None and i >= ram_blocks and len(layer_past) and layer_past[0].device.index != device else layer_past,
attention_mask=attention_mask.to(device) if breakmodel and attention_mask is not None else attention_mask,
head_mask=head_mask[i].to(device) if breakmodel and head_mask[i] is not None else head_mask[i],
use_cache=use_cache,
output_attentions=output_attentions,
)
hidden_states = outputs[0]
if use_cache is True:
presents = presents + (outputs[1],)
if output_attentions:
all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)
if breakmodel:
if i in range(ram_blocks):
torch.cuda.synchronize()
torch.cuda.empty_cache()
if breakmodel:
if ram_blocks:
del copystream
torch.cuda.empty_cache()
hidden_states = hidden_states.to(primary_device)
hidden_states = self.ln_f(hidden_states)
if breakmodel:
hidden_states = hidden_states.to(primary_device)
hidden_states = hidden_states.view(*output_shape)
# Add last hidden state
if output_hidden_states:
all_hidden_states = all_hidden_states + (hidden_states,)
if not return_dict:
return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
return BaseModelOutputWithPast(
last_hidden_state=hidden_states,
past_key_values=presents,
hidden_states=all_hidden_states,
attentions=all_self_attentions,
)
def new_forward_xglm(
self,
input_ids=None,
attention_mask=None,
encoder_hidden_states=None,
encoder_attention_mask=None,
head_mask=None,
cross_attn_head_mask=None,
past_key_values=None,
inputs_embeds=None,
use_cache=None,
output_attentions=None,
output_hidden_states=None,
return_dict=None,
):
assert len(gpu_blocks) <= torch.cuda.device_count()
assert sum(gpu_blocks) <= len(self.layers)
ram_blocks = len(self.layers) - sum(gpu_blocks)
cumulative_gpu_blocks = tuple(itertools.accumulate(gpu_blocks))
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
)
use_cache = use_cache if use_cache is not None else self.config.use_cache
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
# retrieve input_ids and inputs_embeds
if input_ids is not None and inputs_embeds is not None:
raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
elif input_ids is not None:
input_shape = input_ids.size()
input_ids = input_ids.view(-1, input_shape[-1])
elif inputs_embeds is not None:
input_shape = inputs_embeds.size()[:-1]
else:
raise ValueError("You have to specify either input_ids or inputs_embeds")
# past_key_values_length
past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
if inputs_embeds is None:
if breakmodel:
input_ids = input_ids.to(primary_device)
inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale
attention_mask = self._prepare_decoder_attention_mask(
attention_mask, input_shape, inputs_embeds, past_key_values_length
)
# expand encoder attention mask
if encoder_hidden_states is not None and encoder_attention_mask is not None:
# [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
encoder_attention_mask = _expand_mask(encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1])
# embed positions
if breakmodel:
inputs_embeds = inputs_embeds.to(primary_device)
positions = self.embed_positions(input_ids, inputs_embeds, past_key_values_length)
if breakmodel:
positions = positions.to(primary_device)
hidden_states = inputs_embeds + positions
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
# decoder layers
all_hidden_states = () if output_hidden_states else None
all_self_attns = () if output_attentions else None
all_cross_attentions = () if (output_attentions and encoder_hidden_states is not None) else None
next_decoder_cache = () if use_cache else None
if breakmodel and ram_blocks:
copystream = torch.cuda.Stream(device=primary_device, priority=-1)
# check if head_mask/cross_attn_head_mask has a correct number of layers specified if desired
for attn_mask, mask_name in zip([head_mask, cross_attn_head_mask], ["head_mask", "cross_attn_head_mask"]):
if attn_mask is not None:
assert attn_mask.size()[0] == (
len(self.layers)
), f"The `{mask_name}` should be specified for {len(self.layers)} layers, but it is for {head_mask.size()[0]}."
for idx, decoder_layer in enumerate(self.layers):
i = idx
if breakmodel:
if i in range(ram_blocks):
index1 = (i+1)%ram_blocks
for param1,param2 in zip(self.layers[index1].parameters(),self.layers[(i-1)%ram_blocks].parameters()):
param1.data = param2.data
for param1,param2 in zip(self.layers[index1].parameters(),self.extrastorage[index1].parameters()):
with torch.cuda.stream(copystream):
torch.cuda.comm.broadcast(param2.data,out = [param1.data])
# add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
if output_hidden_states:
all_hidden_states += (hidden_states,)
dropout_probability = random.uniform(0, 1)
if self.training and (dropout_probability < self.layerdrop):
continue
past_key_value = past_key_values[idx] if past_key_values is not None else None
if self.gradient_checkpointing and self.training:
if use_cache:
logger.warning(
"`use_cache = True` is incompatible with gradient checkpointing`. Setting `use_cache = False`..."
)
use_cache = False
def create_custom_forward(module):
def custom_forward(*inputs):
# None for past_key_value
return module(*inputs, output_attentions, use_cache)
return custom_forward
layer_outputs = torch.utils.checkpoint.checkpoint(
create_custom_forward(decoder_layer),
hidden_states,
attention_mask,
encoder_hidden_states,
encoder_attention_mask,
head_mask[idx] if head_mask is not None else None,
cross_attn_head_mask[idx] if cross_attn_head_mask is not None else None,
None,
)
else:
if breakmodel:
device = primary_device if i < ram_blocks else bisect.bisect_right(cumulative_gpu_blocks, i - ram_blocks)
layer_outputs = decoder_layer(
hidden_states.to(device) if breakmodel and hidden_states is not None else hidden_states,
attention_mask=attention_mask.to(device) if breakmodel and attention_mask is not None else attention_mask,
encoder_hidden_states=encoder_hidden_states.to(device) if breakmodel and encoder_hidden_states is not None else encoder_hidden_states,
encoder_attention_mask=encoder_attention_mask.to(device) if breakmodel and encoder_attention_mask is not None else encoder_attention_mask,
layer_head_mask=((head_mask[idx].to(device) if breakmodel and head_mask[idx] is not None else head_mask[idx]) if head_mask is not None else None),
cross_attn_layer_head_mask=(
(cross_attn_head_mask[idx].to(device) if breakmodel and cross_attn_head_mask[idx] is not None else cross_attn_head_mask[idx]) if cross_attn_head_mask is not None else None
),
past_key_value=tuple(v.to(device) for v in past_key_value if v is not None) if breakmodel and past_key_value is not None and i >= ram_blocks and len(past_key_value) and past_key_value[0].device.index != device else past_key_value,
output_attentions=output_attentions,
use_cache=use_cache,
)
hidden_states = layer_outputs[0]
if use_cache:
next_decoder_cache += (layer_outputs[3 if output_attentions else 1],)
if output_attentions:
all_self_attns += (layer_outputs[1],)
if encoder_hidden_states is not None:
all_cross_attentions += (layer_outputs[2],)
if breakmodel:
if i in range(ram_blocks):
torch.cuda.synchronize()
torch.cuda.empty_cache()
if breakmodel:
if ram_blocks:
del copystream
torch.cuda.empty_cache()
hidden_states = hidden_states.to(primary_device)
hidden_states = self.layer_norm(hidden_states)
if breakmodel:
hidden_states = hidden_states.to(primary_device)
# add hidden states from the last decoder layer
if output_hidden_states:
all_hidden_states += (hidden_states,)
next_cache = next_decoder_cache if use_cache else None
if not return_dict:
return tuple(
v
for v in [hidden_states, next_cache, all_hidden_states, all_self_attns, all_cross_attentions]
if v is not None
)
return BaseModelOutputWithPastAndCrossAttentions(
last_hidden_state=hidden_states,
past_key_values=next_cache,
hidden_states=all_hidden_states,
attentions=all_self_attns,
cross_attentions=all_cross_attentions,
)
def new_forward_opt(
self,
input_ids=None,
attention_mask=None,
head_mask=None,
past_key_values=None,
inputs_embeds=None,
use_cache=None,
output_attentions=None,
output_hidden_states=None,
return_dict=None,
):
assert len(gpu_blocks) <= torch.cuda.device_count()
assert sum(gpu_blocks) <= len(self.layers)
ram_blocks = len(self.layers) - sum(gpu_blocks)
cumulative_gpu_blocks = tuple(itertools.accumulate(gpu_blocks))
output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
output_hidden_states = (
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
)
use_cache = use_cache if use_cache is not None else self.config.use_cache
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
# retrieve input_ids and inputs_embeds
if input_ids is not None and inputs_embeds is not None:
raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
elif input_ids is not None:
input_shape = input_ids.size()
input_ids = input_ids.view(-1, input_shape[-1])
elif inputs_embeds is not None:
input_shape = inputs_embeds.size()[:-1]
else:
raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0
if inputs_embeds is None:
if breakmodel:
input_ids = input_ids.to(primary_device)
inputs_embeds = self.embed_tokens(input_ids)
# embed positions
if breakmodel:
inputs_embeds = inputs_embeds.to(primary_device)
if attention_mask is None:
attention_mask = torch.ones(inputs_embeds.shape[:2], dtype=torch.bool, device=inputs_embeds.device)
positions = self.embed_positions(attention_mask)[:, past_key_values_length:, :]
if breakmodel:
positions = positions.to(primary_device)
attention_mask = self._prepare_decoder_attention_mask(
attention_mask, input_shape, inputs_embeds, past_key_values_length
)
if self.project_in is not None:
inputs_embeds = self.project_in(inputs_embeds)
hidden_states = inputs_embeds + positions
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
# decoder layers
all_hidden_states = () if output_hidden_states else None
all_self_attns = () if output_attentions else None
next_decoder_cache = () if use_cache else None
if breakmodel and ram_blocks:
copystream = torch.cuda.Stream(device=primary_device, priority=-1)
# check if head_mask has a correct number of layers specified if desired
for attn_mask, mask_name in zip([head_mask], ["head_mask"]):
if attn_mask is not None:
if attn_mask.size()[0] != (len(self.layers)):
raise ValueError(
f"The `{mask_name}` should be specified for {len(self.layers)} layers, but it is for"
f" {head_mask.size()[0]}."
)
for idx, decoder_layer in enumerate(self.layers):
i = idx
if breakmodel:
if i in range(ram_blocks):
index1 = (i+1)%ram_blocks
for param1,param2 in zip(self.layers[index1].parameters(),self.layers[(i-1)%ram_blocks].parameters()):
param1.data = param2.data
for param1,param2 in zip(self.layers[index1].parameters(),self.extrastorage[index1].parameters()):
with torch.cuda.stream(copystream):
torch.cuda.comm.broadcast(param2.data,out = [param1.data])
# add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
if output_hidden_states:
all_hidden_states += (hidden_states,)
dropout_probability = random.uniform(0, 1)
if self.training and (dropout_probability < self.layerdrop):
continue
past_key_value = past_key_values[idx] if past_key_values is not None else None
if self.gradient_checkpointing and self.training:
if use_cache:
logger.warning(
"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
)
use_cache = False
def create_custom_forward(module):
def custom_forward(*inputs):
# None for past_key_value
return module(*inputs, output_attentions, None)
return custom_forward
layer_outputs = torch.utils.checkpoint.checkpoint(
create_custom_forward(decoder_layer),
hidden_states,
attention_mask,
head_mask[idx] if head_mask is not None else None,
None,
)
else:
if breakmodel:
device = primary_device if i < ram_blocks else bisect.bisect_right(cumulative_gpu_blocks, i - ram_blocks)
layer_outputs = decoder_layer(
hidden_states.to(device) if breakmodel and hidden_states is not None else hidden_states,
attention_mask=attention_mask.to(device) if breakmodel and attention_mask is not None else attention_mask,
layer_head_mask=((head_mask[idx].to(device) if breakmodel and head_mask[idx] is not None else head_mask[idx]) if head_mask is not None else None),
past_key_value=tuple(v.to(device) for v in past_key_value if v is not None) if breakmodel and past_key_value is not None and i >= ram_blocks and len(past_key_value) and past_key_value[0].device.index != device else past_key_value,
output_attentions=output_attentions,
use_cache=use_cache,
)
hidden_states = layer_outputs[0]
if use_cache:
next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
if output_attentions:
all_self_attns += (layer_outputs[1],)
if breakmodel:
if i in range(ram_blocks):
torch.cuda.synchronize()
torch.cuda.empty_cache()
if breakmodel:
if ram_blocks:
del copystream
torch.cuda.empty_cache()
hidden_states = hidden_states.to(primary_device)
if self.project_out is not None:
hidden_states = self.project_out(hidden_states)
if breakmodel:
hidden_states = hidden_states.to(primary_device)
# add hidden states from the last decoder layer
if output_hidden_states:
all_hidden_states += (hidden_states,)
next_cache = next_decoder_cache if use_cache else None
if not return_dict:
return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
return BaseModelOutputWithPast(
last_hidden_state=hidden_states,
past_key_values=next_cache,
hidden_states=all_hidden_states,
attentions=all_self_attns,
)

12
modeling/inference_model.py
View File

@@ -182,6 +182,16 @@ class InferenceModel:
setattr(self, parameter, parameters[parameter])
return
def get_auxilary_device(self) -> Union[str, int, torch.device]:
"""Get device auxilary tensors like inputs should be stored on."""
# NOTE: TPU isn't a torch device, so TPU stuff gets sent to CPU.
if utils.koboldai_vars.hascuda and utils.koboldai_vars.usegpu:
return utils.koboldai_vars.gpu_device
elif utils.koboldai_vars.hascuda:
return "cuda"
return "cpu"
def load(self, save_model: bool = False, initial_load: bool = False) -> None:
"""User-facing load function. Do not override this; try `_load()` instead."""
@@ -301,7 +311,7 @@ class InferenceModel:
)
start_time = time.time()
gen_in = gen_in.to(utils.get_auxilary_device())
gen_in = gen_in.to(self.get_auxilary_device())
logger.debug(
"core_generate: gen_in to device time {}s".format(time.time() - start_time)

314
modeling/inference_models/generic_hf_torch/class.py
View File

@@ -13,12 +13,6 @@ import modeling.lazy_loader as lazy_loader
import koboldai_settings
from logger import logger
try:
import breakmodel
except ModuleNotFoundError as e:
# Breakmodel is only expected to work on GPU
if not utils.koboldai_vars.use_colab_tpu:
raise e
from modeling.inference_models.hf_torch import HFTorchInferenceModel
@@ -26,10 +20,9 @@ model_backend_name = "Huggingface"
model_backend_type = "Huggingface" #This should be a generic name in case multiple model backends are compatible (think Hugging Face Custom and Basic Hugging Face)
class model_backend(HFTorchInferenceModel):
def _initialize_model(self):
return
def _load(self, save_model: bool, initial_load: bool) -> None:
utils.koboldai_vars.allowsp = True
@@ -41,9 +34,7 @@ class model_backend(HFTorchInferenceModel):
# utils.koboldai_vars.custmodpth = utils.koboldai_vars.model
if self.model_name == "NeoCustom":
self.model_name = os.path.basename(
os.path.normpath(self.path)
)
self.model_name = os.path.basename(os.path.normpath(self.path))
utils.koboldai_vars.model = self.model_name
# If we specify a model and it's in the root directory, we need to move
@@ -68,23 +59,31 @@ class model_backend(HFTorchInferenceModel):
# Also, lazy loader doesn't support GPT-2 models
self.lazy_load = False
logger.debug(
"lazy_load: {} hascuda: {} breakmodel: {} nobreakmode: {}".format(
self.lazy_load,
utils.koboldai_vars.hascuda,
self.breakmodel,
self.nobreakmodel,
)
)
# If we're using torch_lazy_loader, we need to get breakmodel config
# early so that it knows where to load the individual model tensors
logger.debug("lazy_load: {} hascuda: {} breakmodel: {} nobreakmode: {}".format(self.lazy_load, utils.koboldai_vars.hascuda, self.breakmodel, self.nobreakmodel))
if (
self.lazy_load
and utils.koboldai_vars.hascuda
and self.breakmodel
and not self.nobreakmodel
and utils.koboldai_vars.breakmodel
and not utils.koboldai_vars.nobreakmodel
):
logger.debug("loading breakmodel")
self.breakmodel_device_config(self.model_config)
if self.lazy_load:
# torch_lazy_loader.py and low_cpu_mem_usage can't be used at the same time
tf_kwargs.pop("low_cpu_mem_usage", None)
# If we're using lazy loader, we need to figure out what the model's hidden layers are called
with lazy_loader.use_lazy_load(
dematerialized_modules=True, use_accelerate_init_empty_weights=True
):
with lazy_loader.use_lazy_load(dematerialized_modules=True):
try:
metamodel = AutoModelForCausalLM.from_config(self.model_config)
utils.layers_module_names = utils.get_layers_module_names(metamodel)
@@ -95,143 +94,92 @@ class model_backend(HFTorchInferenceModel):
self.lazy_load = False
# Download model from Huggingface if it does not exist, otherwise load locally
with self._maybe_use_float16(), lazy_loader.use_lazy_load(
enable=self.lazy_load,
callback=self._get_lazy_load_callback(utils.num_layers(self.model_config))
if self.lazy_load
else None,
dematerialized_modules=True,
):
if self.lazy_load:
# torch_lazy_loader.py and low_cpu_mem_usage can't be used at the same time
tf_kwargs.pop("low_cpu_mem_usage", None)
if self.get_local_model_path():
# Model is stored locally, load it.
self.model = self._get_model(self.get_local_model_path(), tf_kwargs)
self.tokenizer = self._get_tokenizer(self.get_local_model_path())
else:
# Model not stored locally, we need to download it.
if self.get_local_model_path():
# Model is stored locally, load it.
self.model = self._get_model(self.get_local_model_path(), tf_kwargs)
self.tokenizer = self._get_tokenizer(self.get_local_model_path())
else:
# Model not stored locally, we need to download it.
# _rebuild_tensor patch for casting dtype and supporting LazyTensors
old_rebuild_tensor = torch._utils._rebuild_tensor
# _rebuild_tensor patch for casting dtype and supporting LazyTensors
old_rebuild_tensor = torch._utils._rebuild_tensor
def new_rebuild_tensor(
storage: Union[lazy_loader.LazyTensor, torch.Storage],
storage_offset,
shape,
stride,
):
if not isinstance(storage, lazy_loader.LazyTensor):
dtype = storage.dtype
else:
dtype = storage.storage_type.dtype
if not isinstance(dtype, torch.dtype):
dtype = storage.storage_type(0).dtype
if dtype is torch.float32 and len(shape) >= 2:
utils.koboldai_vars.fp32_model = True
return old_rebuild_tensor(storage, storage_offset, shape, stride)
def new_rebuild_tensor(
storage: Union[lazy_loader.LazyTensor, torch.Storage],
storage_offset,
shape,
stride,
):
if not isinstance(storage, lazy_loader.LazyTensor):
dtype = storage.dtype
else:
dtype = storage.storage_type.dtype
if not isinstance(dtype, torch.dtype):
dtype = storage.storage_type(0).dtype
if dtype is torch.float32 and len(shape) >= 2:
utils.koboldai_vars.fp32_model = True
return old_rebuild_tensor(storage, storage_offset, shape, stride)
torch._utils._rebuild_tensor = new_rebuild_tensor
self.model = self._get_model(self.model_name, tf_kwargs)
self.tokenizer = self._get_tokenizer(self.model_name)
torch._utils._rebuild_tensor = old_rebuild_tensor
torch._utils._rebuild_tensor = new_rebuild_tensor
self.model = self._get_model(self.model_name, tf_kwargs)
self.tokenizer = self._get_tokenizer(self.model_name)
torch._utils._rebuild_tensor = old_rebuild_tensor
if save_model:
self.tokenizer.save_pretrained(
self.get_local_model_path(ignore_existance=True)
)
if save_model:
self.tokenizer.save_pretrained(
self.get_local_model_path(ignore_existance=True)
if utils.koboldai_vars.fp32_model:
# Use save_pretrained to convert fp32 models to fp16,
# unless we are using disk cache because save_pretrained
# is not supported in that case
self.model = self.model.half()
self.model.save_pretrained(
self.get_local_model_path(ignore_existance=True),
max_shard_size="500MiB",
)
if utils.koboldai_vars.fp32_model and not breakmodel.disk_blocks:
# Use save_pretrained to convert fp32 models to fp16,
# unless we are using disk cache because save_pretrained
# is not supported in that case
self.model = self.model.half()
self.model.save_pretrained(
self.get_local_model_path(ignore_existance=True),
max_shard_size="500MiB",
)
else:
# For fp16 models, we can just copy the model files directly
import transformers.configuration_utils
import transformers.modeling_utils
import transformers.file_utils
import huggingface_hub
else:
# For fp16 models, we can just copy the model files directly
import transformers.configuration_utils
import transformers.modeling_utils
import transformers.file_utils
import huggingface_hub
# Save the config.json
shutil.move(
os.path.realpath(
huggingface_hub.hf_hub_download(
self.model_name,
transformers.configuration_utils.CONFIG_NAME,
revision=utils.koboldai_vars.revision,
cache_dir="cache",
local_files_only=True,
legacy_cache_layout=False,
)
),
os.path.join(
self.get_local_model_path(ignore_existance=True),
# Save the config.json
shutil.move(
os.path.realpath(
huggingface_hub.hf_hub_download(
self.model_name,
transformers.configuration_utils.CONFIG_NAME,
),
)
if utils.num_shards is None:
# Save the pytorch_model.bin or model.safetensors of an unsharded model
any_success = False
possible_checkpoint_names = [
transformers.modeling_utils.WEIGHTS_NAME,
"model.safetensors",
]
for possible_checkpoint_name in possible_checkpoint_names:
try:
shutil.move(
os.path.realpath(
huggingface_hub.hf_hub_download(
self.model_name,
possible_checkpoint_name,
revision=utils.koboldai_vars.revision,
cache_dir="cache",
local_files_only=True,
legacy_cache_layout=False,
)
),
os.path.join(
self.get_local_model_path(
ignore_existance=True
),
possible_checkpoint_name,
),
)
any_success = True
except Exception:
pass
if not any_success:
raise RuntimeError(f"Couldn't find any of {possible_checkpoint_names} in cache for {self.model_name} @ '{utils.koboldai_vars.revisison}'")
else:
# Handle saving sharded models
with open(utils.from_pretrained_index_filename) as f:
map_data = json.load(f)
filenames = set(map_data["weight_map"].values())
# Save the pytorch_model.bin.index.json of a sharded model
shutil.move(
os.path.realpath(utils.from_pretrained_index_filename),
os.path.join(
self.get_local_model_path(ignore_existance=True),
transformers.modeling_utils.WEIGHTS_INDEX_NAME,
),
revision=utils.koboldai_vars.revision,
cache_dir="cache",
local_files_only=True,
legacy_cache_layout=False,
)
# Then save the pytorch_model-#####-of-#####.bin files
for filename in filenames:
),
os.path.join(
self.get_local_model_path(ignore_existance=True),
transformers.configuration_utils.CONFIG_NAME,
),
)
if utils.num_shards is None:
# Save the pytorch_model.bin or model.safetensors of an unsharded model
any_success = False
possible_checkpoint_names = [
transformers.modeling_utils.WEIGHTS_NAME,
"model.safetensors",
]
for possible_checkpoint_name in possible_checkpoint_names:
try:
shutil.move(
os.path.realpath(
huggingface_hub.hf_hub_download(
self.model_name,
filename,
possible_checkpoint_name,
revision=utils.koboldai_vars.revision,
cache_dir="cache",
local_files_only=True,
@@ -242,38 +190,70 @@ class model_backend(HFTorchInferenceModel):
self.get_local_model_path(
ignore_existance=True
),
filename,
possible_checkpoint_name,
),
)
shutil.rmtree("cache/")
any_success = True
except Exception:
pass
if not any_success:
raise RuntimeError(
f"Couldn't find any of {possible_checkpoint_names} in cache for {self.model_name} @ '{utils.koboldai_vars.revisison}'"
)
else:
# Handle saving sharded models
with open(utils.from_pretrained_index_filename) as f:
map_data = json.load(f)
filenames = set(map_data["weight_map"].values())
# Save the pytorch_model.bin.index.json of a sharded model
shutil.move(
os.path.realpath(utils.from_pretrained_index_filename),
os.path.join(
self.get_local_model_path(ignore_existance=True),
transformers.modeling_utils.WEIGHTS_INDEX_NAME,
),
)
# Then save the pytorch_model-#####-of-#####.bin files
for filename in filenames:
shutil.move(
os.path.realpath(
huggingface_hub.hf_hub_download(
self.model_name,
filename,
revision=utils.koboldai_vars.revision,
cache_dir="cache",
local_files_only=True,
legacy_cache_layout=False,
)
),
os.path.join(
self.get_local_model_path(ignore_existance=True),
filename,
),
)
shutil.rmtree("cache/")
self.patch_embedding()
if utils.koboldai_vars.hascuda:
if self.usegpu or self.nobreakmodel:
# Use just VRAM
self.model = self.model.half().to(utils.koboldai_vars.gpu_device)
elif self.breakmodel:
# Use both RAM and VRAM (breakmodel)
if not self.lazy_load:
self.breakmodel_device_config(self.model.config)
self._move_to_devices()
elif breakmodel.disk_blocks > 0:
# Use disk
self._move_to_devices()
else:
# Use CPU
self.model = self.model.to("cpu").float()
elif breakmodel.disk_blocks > 0:
self._move_to_devices()
else:
self.model = self.model.to("cpu").float()
self.model.kai_model = self
utils.koboldai_vars.modeldim = self.get_hidden_size()
def _save_settings(self):
with open("settings/{}.generic_hf_torch.model_backend.settings".format(self.model_name.replace("/", "_")), "w") as f:
json.dump({"layers": self.layers if 'layers' in vars(self) else [], "disk_layers": self.disk_layers if 'disk_layers' in vars(self) else 0}, f, indent="")
with open(
"settings/{}.generic_hf_torch.model_backend.settings".format(
self.model_name.replace("/", "_")
),
"w",
) as f:
json.dump(
{
"layers": self.layers if "layers" in vars(self) else [],
"disk_layers": self.disk_layers
if "disk_layers" in vars(self)
else 0,
},
f,
indent="",
)

6
modeling/inference_models/hf.py
View File

@@ -157,7 +157,6 @@ class HFInferenceModel(InferenceModel):
def set_input_parameters(self, parameters):
if self.hf_torch and hasattr(self, "get_model_type") and self.get_model_type() != "gpt2":
import breakmodel
layer_count = self.model_config["n_layer"] if isinstance(self.model_config, dict) else self.model_config.num_layers if hasattr(self.model_config, "num_layers") else self.model_config.n_layer if hasattr(self.model_config, "n_layer") else self.model_config.num_hidden_layers if hasattr(self.model_config, 'num_hidden_layers') else None
if layer_count is not None and layer_count >= 0 and not self.nobreakmodel:
gpu_count = torch.cuda.device_count()
@@ -176,9 +175,8 @@ class HFInferenceModel(InferenceModel):
self.disk_layers = parameters['Disk_Layers'] if 'Disk_Layers' in parameters else 0
if isinstance(self.disk_layers, str):
self.disk_layers = int(self.disk_layers) if self.disk_layers.isnumeric() else 0
breakmodel.gpu_blocks = layers
breakmodel.disk_blocks = self.disk_layers
self.usegpu = self.cpu_layers == 0 and breakmodel.disk_blocks == 0 and sum(self.layers)-self.layers[0] == 0
print("TODO: Allow config")
# self.usegpu = self.cpu_layers == 0 and breakmodel.disk_blocks == 0 and sum(self.layers)-self.layers[0] == 0
self.model_type = self.get_model_type()
self.breakmodel = ((self.model_type != 'gpt2') or self.model_type in ("gpt_neo", "gptj", "xglm", "opt")) and not self.nobreakmodel
self.lazy_load = True

649
modeling/inference_models/hf_torch.py
View File

@@ -1,15 +1,13 @@
from __future__ import annotations
from dataclasses import dataclass
import gc
import os
import time
import bisect
import zipfile
import functools
import itertools
import traceback
import contextlib
from tqdm.auto import tqdm
from torch import nn
from typing import Dict, List, Optional, Union
import torch
@@ -39,19 +37,52 @@ from modeling.inference_model import (
use_core_manipulations,
)
try:
import breakmodel
import accelerate.utils
except ModuleNotFoundError as e:
if not utils.koboldai_vars.use_colab_tpu:
raise e
# When set to true, messages will appear in the console if samplers are not
# changing the scores. Keep in mind some samplers don't always change the
# scores for each token.
LOG_SAMPLER_NO_EFFECT = False
class BreakmodelConfig:
def __init__(self) -> None:
self.disk_blocks = 0
self.gpu_blocks = []
@property
def primary_device(self):
if utils.args.cpu:
return "cpu"
elif not sum(self.gpu_blocks):
# No blocks are on GPU
return "cpu"
elif torch.cuda.device_count() <= 0:
return "cpu"
for device_index, blocks in enumerate(self.gpu_blocks):
if blocks:
return device_index
return 0
def get_device_map(self, model: nn.Module) -> dict:
ram_blocks = len(utils.layers_module_names) - sum(self.gpu_blocks)
cumulative_gpu_blocks = tuple(itertools.accumulate(self.gpu_blocks))
device_map = {}
for name in utils.layers_module_names:
layer = int(name.rsplit(".", 1)[1])
device = (
("disk" if layer < self.disk_blocks else "cpu")
if layer < ram_blocks
else bisect.bisect_right(cumulative_gpu_blocks, layer - ram_blocks)
)
device_map[name] = device
for name in utils.get_missing_module_names(model, list(device_map.keys())):
device_map[name] = self.primary_device
return device_map
class HFTorchInferenceModel(HFInferenceModel):
def __init__(self) -> None:
super().__init__()
@@ -79,6 +110,29 @@ class HFTorchInferenceModel(HFInferenceModel):
post_token_probs=True,
)
self._old_stopping_criteria = None
self.breakmodel_config = BreakmodelConfig()
def set_input_parameters(self, parameters):
ret = super().set_input_parameters(parameters)
# Hook onto input param setting for setting breakmodel stuff
if self.breakmodel:
self.breakmodel_config.gpu_blocks = self.layers
self.breakmodel_config.disk_blocks = self.disk_layers
return ret
def get_auxilary_device(self) -> Union[str, int, torch.device]:
return self.breakmodel_config.primary_device
def _get_target_dtype(self) -> Union[torch.float16, torch.float32]:
if self.breakmodel_config.primary_device == "cpu":
return torch.float32
elif utils.args.cpu:
return torch.float32
elif not self.usegpu and not self.breakmodel:
return torch.float32
return torch.float16
def _apply_warpers(
self, scores: torch.Tensor, input_ids: torch.Tensor
@@ -125,19 +179,7 @@ class HFTorchInferenceModel(HFInferenceModel):
else:
return "Unknown"
def get_auxilary_device(self):
"""Get device auxilary tensors like inputs should be stored on."""
# NOTE: TPU isn't a torch device, so TPU stuff gets sent to CPU.
if utils.koboldai_vars.hascuda and self.usegpu:
return utils.koboldai_vars.gpu_device
elif utils.koboldai_vars.hascuda and self.breakmodel:
import breakmodel
return breakmodel.primary_device
return "cpu"
def _post_load(m_self) -> None:
if not utils.koboldai_vars.model_type:
utils.koboldai_vars.model_type = m_self.get_model_type()
@@ -236,9 +278,11 @@ class HFTorchInferenceModel(HFInferenceModel):
gen_in = torch.tensor(prompt_tokens, dtype=torch.long)[None]
else:
gen_in = prompt_tokens
device = self.get_auxilary_device()
gen_in = gen_in.to(device)
if not self.usegpu and not self.breakmodel:
gen_in = gen_in.to("cpu")
else:
device = self.get_auxilary_device()
gen_in = gen_in.to(device)
additional_bad_words_ids = [self.tokenizer.encode("\n")] if single_line else []
@@ -254,8 +298,7 @@ class HFTorchInferenceModel(HFInferenceModel):
len(prompt_tokens) + max_new, utils.koboldai_vars.max_length
),
repetition_penalty=1.0,
bad_words_ids=self.badwordsids
+ additional_bad_words_ids,
bad_words_ids=self.badwordsids + additional_bad_words_ids,
use_cache=True,
num_return_sequences=batch_count,
)
@@ -275,6 +318,9 @@ class HFTorchInferenceModel(HFInferenceModel):
tf_kwargs["revision"] = utils.koboldai_vars.revision
tf_kwargs["cache_dir"] = "cache"
if self.lazy_load:
tf_kwargs.pop("low_cpu_mem_usage", None)
# If we have model hints for legacy model, use them rather than fall back.
try:
if self.model_name == "GPT2Custom":
@@ -286,7 +332,41 @@ class HFTorchInferenceModel(HFInferenceModel):
# Try to determine model type from either AutoModel or falling back to legacy
try:
return AutoModelForCausalLM.from_pretrained(location, **tf_kwargs)
if self.lazy_load:
with lazy_loader.use_lazy_load(dematerialized_modules=True):
metamodel = AutoModelForCausalLM.from_config(self.model_config)
if utils.args.cpu:
cpu_map = {name: "cpu" for name in utils.layers_module_names}
for name in utils.get_missing_module_names(
metamodel, list(cpu_map.keys())
):
cpu_map[name] = "cpu"
tf_kwargs["device_map"] = cpu_map
else:
tf_kwargs["device_map"] = self.breakmodel_config.get_device_map(
metamodel
)
with lazy_loader.use_lazy_load(
enable=self.lazy_load,
# DO NOT DEMATERIALIZE MODULES / INIT WEIGHTS EMPTY!!! IT WILL EXPLODE!!!!!!!
dematerialized_modules=False,
):
model = AutoModelForCausalLM.from_pretrained(
location,
offload_folder="accelerate-disk-cache",
torch_dtype=self._get_target_dtype(),
**tf_kwargs,
)
if not self.lazy_load:
# We need to move the model to the desired device
if (not self.usegpu) or torch.cuda.device_count() <= 0:
model = model.to("cpu")
else:
model = model.to("cuda")
return model
except Exception as e:
traceback_string = traceback.format_exc().lower()
@@ -325,49 +405,6 @@ class HFTorchInferenceModel(HFInferenceModel):
return True
def _move_to_devices(self) -> None:
for key, value in self.model.state_dict().items():
target_dtype = (
torch.float32 if breakmodel.primary_device == "cpu" else torch.float16
)
if value.dtype is not target_dtype:
accelerate.utils.set_module_tensor_to_device(
self.model,
tensor_name=key,
device=torch.device(value.device),
value=value,
dtype=target_dtype,
)
disk_blocks = breakmodel.disk_blocks
gpu_blocks = breakmodel.gpu_blocks
ram_blocks = len(utils.layers_module_names) - sum(gpu_blocks)
cumulative_gpu_blocks = tuple(itertools.accumulate(gpu_blocks))
device_map = {}
for name in utils.layers_module_names:
layer = int(name.rsplit(".", 1)[1])
device = (
("disk" if layer < disk_blocks else "cpu")
if layer < ram_blocks
else bisect.bisect_right(cumulative_gpu_blocks, layer - ram_blocks)
)
device_map[name] = device
for name in utils.get_missing_module_names(self.model, list(device_map.keys())):
device_map[name] = breakmodel.primary_device
breakmodel.dispatch_model_ex(
self.model,
device_map,
main_device=breakmodel.primary_device,
offload_buffers=True,
offload_dir="accelerate-disk-cache",
)
gc.collect()
return
# Function to patch transformers to use our soft prompt
def patch_embedding(self) -> None:
if getattr(Embedding, "_koboldai_patch_causallm_model", None):
@@ -409,401 +446,20 @@ class HFTorchInferenceModel(HFInferenceModel):
Embedding.__call__ = new_embedding_call
Embedding._koboldai_patch_causallm_model = self.model
def _get_lazy_load_callback(self, n_layers: int, convert_to_float16: bool = True):
if not self.lazy_load:
return
disk_blocks = breakmodel.disk_blocks
gpu_blocks = breakmodel.gpu_blocks
ram_blocks = ram_blocks = n_layers - sum(gpu_blocks)
cumulative_gpu_blocks = tuple(itertools.accumulate(gpu_blocks))
def lazy_load_callback(
model_dict: Dict[str, Union[lazy_loader.LazyTensor, torch.Tensor]],
f,
is_safetensors: bool = False,
**_,
):
if lazy_load_callback.nested:
return
lazy_load_callback.nested = True
device_map: Dict[str, Union[str, int]] = {}
@functools.lru_cache(maxsize=None)
def get_original_key(key) -> Optional[str]:
key_candidates = [
original_key
for original_key in utils.module_names
if original_key.endswith(key)
]
if not key_candidates:
logger.debug(f"!!! No key candidates for {key}")
return None
return max(key_candidates, key=len)
for key, value in model_dict.items():
original_key = get_original_key(key)
if not original_key:
continue
if isinstance(value, lazy_loader.LazyTensor) and not any(
original_key.startswith(n) for n in utils.layers_module_names
):
device_map[key] = (
utils.koboldai_vars.gpu_device
if utils.koboldai_vars.hascuda and self.usegpu
else "cpu"
if not utils.koboldai_vars.hascuda
or not self.breakmodel
else breakmodel.primary_device
)
else:
layer = int(
max(
(
n
for n in utils.layers_module_names
if original_key.startswith(n)
),
key=len,
).rsplit(".", 1)[1]
)
device = (
utils.koboldai_vars.gpu_device
if utils.koboldai_vars.hascuda and self.usegpu
else "disk"
if layer < disk_blocks and layer < ram_blocks
else "cpu"
if not utils.koboldai_vars.hascuda
or not self.breakmodel
else "shared"
if layer < ram_blocks
else bisect.bisect_right(
cumulative_gpu_blocks, layer - ram_blocks
)
)
device_map[key] = device
if utils.num_shards is None or utils.current_shard == 0:
utils.offload_index = {}
if os.path.isdir("accelerate-disk-cache"):
# Delete all of the files in the disk cache folder without deleting the folder itself to allow people to create symbolic links for this folder
# (the folder doesn't contain any subfolders so os.remove will do just fine)
for filename in os.listdir("accelerate-disk-cache"):
try:
os.remove(os.path.join("accelerate-disk-cache", filename))
except OSError:
pass
os.makedirs("accelerate-disk-cache", exist_ok=True)
if utils.num_shards is not None:
num_tensors = len(
utils.get_sharded_checkpoint_num_tensors(
utils.from_pretrained_model_name,
utils.from_pretrained_index_filename,
is_safetensors=is_safetensors,
**utils.from_pretrained_kwargs,
)
)
else:
num_tensors = len(device_map)
print(flush=True)
utils.koboldai_vars.status_message = "Loading model"
utils.koboldai_vars.total_layers = num_tensors
utils.koboldai_vars.loaded_layers = 0
utils.bar = tqdm(
total=num_tensors,
desc="Loading model tensors",
file=utils.UIProgressBarFile(),
position=1
)
if not is_safetensors:
# Torch lazyload
with zipfile.ZipFile(f, "r") as z:
try:
last_storage_key = None
zipfolder = os.path.basename(os.path.normpath(f)).split(".")[0]
f = None
current_offset = 0
able_to_pin_layers = True
if utils.num_shards is not None:
utils.current_shard += 1
for key in sorted(
device_map.keys(),
key=lambda k: (
model_dict[k].key,
model_dict[k].seek_offset,
),
):
storage_key = model_dict[key].key
if (
storage_key != last_storage_key
or model_dict[key].seek_offset < current_offset
):
last_storage_key = storage_key
if isinstance(f, zipfile.ZipExtFile):
f.close()
ziproot = z.namelist()[0].split("/")[0]
f = z.open(f"{ziproot}/data/{storage_key}")
current_offset = 0
if current_offset != model_dict[key].seek_offset:
f.read(model_dict[key].seek_offset - current_offset)
current_offset = model_dict[key].seek_offset
device = device_map[key]
size = functools.reduce(
lambda x, y: x * y, model_dict[key].shape, 1
)
dtype = model_dict[key].dtype
nbytes = (
size
if dtype is torch.bool
else size
* (
(
torch.finfo
if dtype.is_floating_point
else torch.iinfo
)(dtype).bits
>> 3
)
)
# print(f"Transferring <{key}> to {f'({device.upper()})' if isinstance(device, str) else '[device ' + str(device) + ']'} ... ", end="", flush=True)
#logger.debug(f"Transferring <{key}> to {f'({device.upper()})' if isinstance(device, str) else '[device ' + str(device) + ']'} ... ")
model_dict[key] = model_dict[key].materialize(
f, map_location="cpu"
)
if model_dict[key].dtype is torch.float32:
utils.koboldai_vars.fp32_model = True
if (
convert_to_float16
and breakmodel.primary_device != "cpu"
and utils.koboldai_vars.hascuda
and (
self.breakmodel
or self.usegpu
)
and model_dict[key].dtype is torch.float32
):
model_dict[key] = model_dict[key].to(torch.float16)
if breakmodel.primary_device == "cpu" or (
not self.usegpu
and not self.breakmodel
and model_dict[key].dtype is torch.float16
):
model_dict[key] = model_dict[key].to(torch.float32)
if device == "shared":
model_dict[key] = model_dict[key].to("cpu").detach_()
if able_to_pin_layers:
try:
model_dict[key] = model_dict[key].pin_memory()
except:
able_to_pin_layers = False
elif device == "disk":
accelerate.utils.offload_weight(
model_dict[key],
get_original_key(key),
"accelerate-disk-cache",
index=utils.offload_index,
)
model_dict[key] = model_dict[key].to("meta")
else:
model_dict[key] = model_dict[key].to(device)
# print("OK", flush=True)
current_offset += nbytes
utils.bar.update(1)
utils.koboldai_vars.loaded_layers += 1
finally:
if (
utils.num_shards is None
or utils.current_shard >= utils.num_shards
):
if utils.offload_index:
for name, tensor in utils.named_buffers:
dtype = tensor.dtype
if (
convert_to_float16
and breakmodel.primary_device != "cpu"
and utils.koboldai_vars.hascuda
and (
self.breakmodel
or self.usegpu
)
):
dtype = torch.float16
if breakmodel.primary_device == "cpu" or (
not self.usegpu
and not self.breakmodel
):
dtype = torch.float32
if (
name in model_dict
and model_dict[name].dtype is not dtype
):
model_dict[name] = model_dict[name].to(dtype)
if tensor.dtype is not dtype:
tensor = tensor.to(dtype)
if name not in utils.offload_index:
accelerate.utils.offload_weight(
tensor,
name,
"accelerate-disk-cache",
index=utils.offload_index,
)
accelerate.utils.save_offload_index(
utils.offload_index, "accelerate-disk-cache"
)
utils.bar.close()
utils.bar = None
utils.koboldai_vars.status_message = ""
lazy_load_callback.nested = False
if isinstance(f, zipfile.ZipExtFile):
f.close()
else:
# Loading with safetensors
try:
able_to_pin_layers = True
if utils.num_shards is not None:
utils.current_shard += 1
for key in sorted(
device_map.keys(),
key=lambda k: model_dict[k].key,
):
storage_key = model_dict[key].key
device = device_map[key]
# print(f"Transferring <{key}> to {f'({device.upper()})' if isinstance(device, str) else '[device ' + str(device) + ']'} ... ", end="", flush=True)
model_dict[key] = model_dict[key].materialize(
f, map_location="cpu"
)
if model_dict[key].dtype is torch.float32:
utils.koboldai_vars.fp32_model = True
if (
convert_to_float16
and breakmodel.primary_device != "cpu"
and utils.koboldai_vars.hascuda
and (
self.breakmodel
or self.usegpu
)
and model_dict[key].dtype is torch.float32
):
model_dict[key] = model_dict[key].to(torch.float16)
if breakmodel.primary_device == "cpu" or (
not self.usegpu
and not self.breakmodel
and model_dict[key].dtype is torch.float16
):
model_dict[key] = model_dict[key].to(torch.float32)
if device == "shared":
model_dict[key] = model_dict[key].to("cpu").detach_()
if able_to_pin_layers:
try:
model_dict[key] = model_dict[key].pin_memory()
except:
able_to_pin_layers = False
elif device == "disk":
accelerate.utils.offload_weight(
model_dict[key],
get_original_key(key),
"accelerate-disk-cache",
index=utils.offload_index,
)
model_dict[key] = model_dict[key].to("meta")
else:
model_dict[key] = model_dict[key].to(device)
utils.bar.update(1)
utils.koboldai_vars.loaded_layers += 1
finally:
if (
utils.num_shards is None
or utils.current_shard >= utils.num_shards
):
if utils.offload_index:
for name, tensor in utils.named_buffers:
dtype = tensor.dtype
if (
convert_to_float16
and breakmodel.primary_device != "cpu"
and utils.koboldai_vars.hascuda
and (
self.breakmodel
or self.usegpu
)
):
dtype = torch.float16
if breakmodel.primary_device == "cpu" or (
not self.usegpu
and not self.breakmodel
):
dtype = torch.float32
if (
name in model_dict
and model_dict[name].dtype is not dtype
):
model_dict[name] = model_dict[name].to(dtype)
if tensor.dtype is not dtype:
tensor = tensor.to(dtype)
if name not in utils.offload_index:
accelerate.utils.offload_weight(
tensor,
name,
"accelerate-disk-cache",
index=utils.offload_index,
)
accelerate.utils.save_offload_index(
utils.offload_index, "accelerate-disk-cache"
)
utils.bar.close()
utils.bar = None
utils.koboldai_vars.status_message = ""
lazy_load_callback.nested = False
lazy_load_callback.nested = False
return lazy_load_callback
@contextlib.contextmanager
def _maybe_use_float16(self, always_use: bool = False):
if always_use or (
utils.koboldai_vars.hascuda
and self.low_mem
and (self.usegpu or self.breakmodel)
):
original_dtype = torch.get_default_dtype()
torch.set_default_dtype(torch.float16)
yield True
torch.set_default_dtype(original_dtype)
else:
yield False
def breakmodel_device_list(self, n_layers, primary=None, selected=None):
# TODO: Find a better place for this or rework this
device_count = torch.cuda.device_count()
if device_count < 2:
primary = None
logger.debug("n_layers: {}".format(n_layers))
logger.debug("gpu blocks: {}".format(breakmodel.gpu_blocks))
gpu_blocks = breakmodel.gpu_blocks + (
device_count - len(breakmodel.gpu_blocks)
logger.debug("gpu blocks: {}".format(self.breakmodel_config.gpu_blocks))
gpu_blocks = self.breakmodel_config.gpu_blocks + (
device_count - len(self.breakmodel_config.gpu_blocks)
) * [0]
print(f"{Colors.YELLOW} DEVICE ID | LAYERS | DEVICE NAME{Colors.END}")
for i in range(device_count):
name = torch.cuda.get_device_name(i)
if len(name) > 47:
@@ -813,72 +469,83 @@ class HFTorchInferenceModel(HFInferenceModel):
print(
f"{row_color}{Colors.YELLOW + '->' + row_color if i == selected else ' '} {'(primary)' if i == primary else ' '*9} {i:3} {sep_color}|{row_color} {gpu_blocks[i]:3} {sep_color}|{row_color} {name}{Colors.END}"
)
row_color = Colors.END
sep_color = Colors.YELLOW
print(
f"{row_color}{Colors.YELLOW + '->' + row_color if -1 == selected else ' '} {' '*9} N/A {sep_color}|{row_color} {breakmodel.disk_blocks:3} {sep_color}|{row_color} (Disk cache){Colors.END}"
f"{row_color}{Colors.YELLOW + '->' + row_color if -1 == selected else ' '} {' '*9} N/A {sep_color}|{row_color} {self.breakmodel_config.disk_blocks:3} {sep_color}|{row_color} (Disk cache){Colors.END}"
)
print(
f"{row_color} {' '*9} N/A {sep_color}|{row_color} {n_layers:3} {sep_color}|{row_color} (CPU){Colors.END}"
)
def breakmodel_device_config(self, config):
# TODO: Find a better place for this or rework this
global breakmodel, generator
import breakmodel
n_layers = utils.num_layers(config)
logger.debug("gpu blocks before modification: {}".format(breakmodel.gpu_blocks))
logger.debug(
"gpu blocks before modification: {}".format(
self.breakmodel_config.gpu_blocks
)
)
if utils.args.cpu:
breakmodel.gpu_blocks = [0] * n_layers
self.breakmodel_config.gpu_blocks = [0] * n_layers
return
elif breakmodel.gpu_blocks == []:
elif self.breakmodel_config.gpu_blocks == []:
logger.info("Breakmodel not specified, assuming GPU 0")
breakmodel.gpu_blocks = [n_layers]
self.breakmodel_config.gpu_blocks = [n_layers]
n_layers = 0
else:
s = n_layers
for i in range(len(breakmodel.gpu_blocks)):
if breakmodel.gpu_blocks[i] <= -1:
breakmodel.gpu_blocks[i] = s
for i in range(len(self.breakmodel_config.gpu_blocks)):
if self.breakmodel_config.gpu_blocks[i] <= -1:
self.breakmodel_config.gpu_blocks[i] = s
break
else:
s -= breakmodel.gpu_blocks[i]
assert sum(breakmodel.gpu_blocks) <= n_layers
n_layers -= sum(breakmodel.gpu_blocks)
if breakmodel.disk_blocks is not None:
assert breakmodel.disk_blocks <= n_layers
n_layers -= breakmodel.disk_blocks
s -= self.breakmodel_config.gpu_blocks[i]
assert sum(self.breakmodel_config.gpu_blocks) <= n_layers
n_layers -= sum(self.breakmodel_config.gpu_blocks)
if self.breakmodel_config.disk_blocks is not None:
assert self.breakmodel_config.disk_blocks <= n_layers
n_layers -= self.breakmodel_config.disk_blocks
logger.init_ok("Final device configuration:", status="Info")
self.breakmodel_device_list(n_layers, primary=breakmodel.primary_device)
with open("settings/{}.breakmodel".format(self.model_name.replace("/", "_")), "w") as file:
file.write("{}\n{}".format(",".join(map(str, breakmodel.gpu_blocks)), breakmodel.disk_blocks))
self.breakmodel_device_list(
n_layers, primary=self.breakmodel_config.primary_device
)
with open(
"settings/{}.breakmodel".format(self.model_name.replace("/", "_")), "w"
) as file:
file.write(
"{}\n{}".format(
",".join(map(str, self.breakmodel_config.gpu_blocks)),
self.breakmodel_config.disk_blocks,
)
)
# If all layers are on the same device, use the old GPU generation mode
while len(breakmodel.gpu_blocks) and breakmodel.gpu_blocks[-1] == 0:
breakmodel.gpu_blocks.pop()
while (
len(self.breakmodel_config.gpu_blocks)
and self.breakmodel_config.gpu_blocks[-1] == 0
):
self.breakmodel_config.gpu_blocks.pop()
self.breakmodel = True
if len(breakmodel.gpu_blocks) and breakmodel.gpu_blocks[-1] in (
if len(self.breakmodel_config.gpu_blocks) and self.breakmodel_config.gpu_blocks[
-1
] in (
-1,
utils.num_layers(config),
):
logger.debug("All layers on same GPU. Breakmodel disabled")
self.breakmodel = False
self.usegpu = True
utils.koboldai_vars.gpu_device = len(breakmodel.gpu_blocks) - 1
utils.koboldai_vars.gpu_device = len(self.breakmodel_config.gpu_blocks) - 1
return
if not breakmodel.gpu_blocks:
if not self.breakmodel_config.gpu_blocks:
logger.warning("Nothing assigned to a GPU, reverting to CPU only mode")
import breakmodel
breakmodel.primary_device = "cpu"
self.breakmodel = False
self.usegpu = False
return

312
modeling/lazy_loader.py
View File

@@ -47,6 +47,7 @@ POSSIBILITY OF SUCH DAMAGE.
import contextlib
from functools import reduce
import itertools
import time
import zipfile
import pickle
import torch
@@ -54,35 +55,57 @@ import numpy as np
import collections
import _codecs
import os
from typing import Any, Callable, Dict, Optional, Tuple, Type
import accelerate
from torch.nn import Module
from typing import Any, Callable, Dict, Optional, Tuple, Type, Union
from torch.storage import UntypedStorage
# Safetensors is a dependency for the local version, TPU/Colab doesn't
# support it yet.
try:
import safetensors
HAS_SAFETENSORS = True
except ModuleNotFoundError:
HAS_SAFETENSORS = False
import utils
from logger import logger
# Storage of zipfile handles for each shard
torch_checkpoint_file_handles = {}
_EXTRA_STATE_KEY_SUFFIX = "_extra_state"
class CheckpointChunkCache:
"""Storage for common checkpoint weight files to speed up loading. In order
for this to be effective at all, weights must be loaded in ascending order
of (key, seek_offset).
"""
# There is considerable room for improvement here; we could peek into the
# state dict and preload the N most frequent weight files or something, but
# this first implementation is on par with the speed of whatever the
# previous callback did.
STORAGE_TYPE_MAP = {
torch.float64: torch.DoubleStorage,
torch.float32: torch.FloatStorage,
torch.float16: torch.HalfStorage,
torch.int64: torch.LongStorage,
torch.int32: torch.IntStorage,
torch.int16: torch.ShortStorage,
torch.int8: torch.CharStorage,
torch.uint8: torch.ByteStorage,
torch.bool: torch.BoolStorage,
torch.bfloat16: torch.BFloat16Storage,
}
file_name = None
key = None
handle = None
hit_data = {"hits": 0, "misses": 0}
@classmethod
def clear(cls, unload_model: bool = False) -> None:
if unload_model:
cls.hit_data["hits"] = 0
cls.hit_data["misses"] = 0
if cls.handle:
cls.handle.close()
cls.file_name = None
cls.key = None
cls.handle = None
class LazyTensor:
@@ -111,6 +134,7 @@ class TorchLazyTensor(LazyTensor):
self.stride = stride
self.requires_grad = requires_grad
self.backward_hooks = backward_hooks
self.file_name = None
def __view(self, f: Callable):
return f"{type(self).__name__}(storage_type={f(self.storage_type)}, key={f(self.key)}, location={f(self.location)}, dtype={f(self.dtype)}, seek_offset={f(self.seek_offset)}, shape={f(self.shape)}, stride={f(self.stride)}, requires_grad={f(self.requires_grad)}, backward_hooks={f(self.backward_hooks)})"
@@ -120,12 +144,42 @@ class TorchLazyTensor(LazyTensor):
def materialize(
self,
checkpoint: Union[zipfile.ZipFile, zipfile.ZipExtFile],
map_location=None,
no_grad=True,
filename="pytorch_model.bin",
) -> torch.Tensor:
filename = os.path.basename(os.path.normpath(filename)).split(".")[0]
checkpoint = torch_checkpoint_file_handles[self.file_name]
filename = os.path.basename(os.path.normpath(self.file_name)).split(".")[0]
# Often we are using the same weight file to store multiple tensors, so
# let's cache the file handle to maintain a seek position and other
# fast stuff.
if (
CheckpointChunkCache.file_name != filename
or CheckpointChunkCache.key != self.key
or not CheckpointChunkCache.handle
):
# Cache miss. Assuming weights are loaded in order of
# (key, seek_offset), this means we need to invalidate the cache.
# print("!", end="", flush=True)
CheckpointChunkCache.hit_data["misses"] += 1
CheckpointChunkCache.clear()
CheckpointChunkCache.file_name = filename
CheckpointChunkCache.key = self.key
try:
CheckpointChunkCache.handle = checkpoint.open(
f"archive/data/{self.key}", "r"
)
except KeyError:
CheckpointChunkCache.handle = checkpoint.open(
f"{filename}/data/{self.key}", "r"
)
else:
# Cache hit. Hip hip hooray! :^)
# print(".", end="", flush=True)
CheckpointChunkCache.hit_data["hits"] += 1
size = reduce(lambda x, y: x * y, self.shape, 1)
dtype = self.dtype
nbytes = (
@@ -133,27 +187,24 @@ class TorchLazyTensor(LazyTensor):
if dtype is torch.bool
else size
* (
(torch.finfo if dtype.is_floating_point else torch.iinfo)(dtype).bits
(torch.finfo if self.dtype.is_floating_point else torch.iinfo)(
self.dtype
).bits
>> 3
)
)
if isinstance(checkpoint, zipfile.ZipFile):
try:
f = checkpoint.open(f"archive/data/{self.key}", "r")
except:
f = checkpoint.open(f"{filename}/data/{self.key}", "r")
f.read(self.seek_offset)
else:
f = checkpoint
try:
storage = STORAGE_TYPE_MAP[dtype].from_buffer(f.read(nbytes), "little")
finally:
if isinstance(checkpoint, zipfile.ZipFile):
f.close()
assert isinstance(checkpoint, zipfile.ZipFile)
CheckpointChunkCache.handle.seek(self.seek_offset, os.SEEK_SET)
storage = UntypedStorage.from_buffer(
CheckpointChunkCache.handle.read(nbytes), "little", dtype=self.dtype
)
storage = torch.serialization._get_restore_location(map_location)(
storage, self.location
)
tensor = torch.tensor([], dtype=storage.dtype, device=storage.device)
tensor = torch.tensor([], dtype=self.dtype, device=storage.device)
tensor.set_(storage, 0, self.shape, self.stride)
tensor.requires_grad = not no_grad and self.requires_grad
tensor._backward_hooks = self.backward_hooks
@@ -237,6 +288,7 @@ class _LazyUnpickler(RestrictedUnpickler):
lazy_loaded_storages: Dict[str, LazyTensor]
def __init__(self, *args, **kwargs):
# print(args, kwargs)
self.lazy_loaded_storages = {}
return super().__init__(*args, **kwargs)
@@ -271,117 +323,6 @@ def _rebuild_tensor(lazy_storage: LazyTensor, storage_offset, shape, stride):
return lazy_storage
# Modified version of https://github.com/pytorch/pytorch/blob/v1.11.0-rc4/torch/nn/modules/module.py#L1346-L1438
def _load_from_state_dict(
self,
state_dict,
prefix,
local_metadata,
strict,
missing_keys,
unexpected_keys,
error_msgs,
):
for hook in self._load_state_dict_pre_hooks.values():
hook(
state_dict,
prefix,
local_metadata,
strict,
missing_keys,
unexpected_keys,
error_msgs,
)
persistent_buffers = {
k: v
for k, v in self._buffers.items()
if k not in self._non_persistent_buffers_set
}
local_name_params = itertools.chain(
self._parameters.items(), persistent_buffers.items()
)
local_state = {k: v for k, v in local_name_params if v is not None}
for name, param in local_state.items():
key = prefix + name
if key in state_dict:
input_param = state_dict[key]
if not torch.overrides.is_tensor_like(input_param):
error_msgs.append(
'While copying the parameter named "{}", '
"expected torch.Tensor or Tensor-like object from checkpoint but "
"received {}".format(key, type(input_param))
)
continue
# This is used to avoid copying uninitialized parameters into
# non-lazy modules, since they dont have the hook to do the checks
# in such case, it will error when accessing the .shape attribute.
is_param_lazy = torch.nn.parameter.is_lazy(param)
# Backward compatibility: loading 1-dim tensor from 0.3.* to version 0.4+
if (
not is_param_lazy
and len(param.shape) == 0
and len(input_param.shape) == 1
):
input_param = input_param[0]
if not is_param_lazy and input_param.shape != param.shape:
# local shape should match the one in checkpoint
error_msgs.append(
"size mismatch for {}: copying a param with shape {} from checkpoint, "
"the shape in current model is {}.".format(
key, input_param.shape, param.shape
)
)
continue
try:
with torch.no_grad():
# param.copy_(input_param)
new_param = torch.nn.Parameter(
input_param, requires_grad=param.requires_grad
) # This line is new
if name in self._parameters: # This line is new
self._parameters[name] = new_param # This line is new
if name in persistent_buffers: # This line is new
self._buffers[name] = new_param # This line is new
except Exception as ex:
error_msgs.append(
'While copying the parameter named "{}", '
"whose dimensions in the model are {} and "
"whose dimensions in the checkpoint are {}, "
"an exception occurred : {}.".format(
key, param.size(), input_param.size(), ex.args
)
)
elif strict:
missing_keys.append(key)
extra_state_key = prefix + _EXTRA_STATE_KEY_SUFFIX
if (
hasattr(Module, "set_extra_state")
and getattr(self.__class__, "set_extra_state", Module.set_extra_state)
is not Module.set_extra_state
): # if getattr(self.__class__, "set_extra_state", Module.set_extra_state) is not Module.set_extra_state:
if extra_state_key in state_dict:
self.set_extra_state(state_dict[extra_state_key])
elif strict:
missing_keys.append(extra_state_key)
elif strict and (extra_state_key in state_dict):
unexpected_keys.append(extra_state_key)
if strict:
for key in state_dict.keys():
if key.startswith(prefix) and key != extra_state_key:
input_name = key[len(prefix) :]
input_name = input_name.split(".", 1)[
0
] # get the name of param/buffer/child
if input_name not in self._modules and input_name not in local_state:
unexpected_keys.append(key)
def safetensors_load_tensor_independently(
checkpoint_file: str, tensor_key: str, device: Any
) -> torch.Tensor:
@@ -412,15 +353,18 @@ def patch_safetensors(callback):
tensors = {}
with safetensors.safe_open(
checkpoint_file, framework="pt", device=intermediary_device,
checkpoint_file,
framework="pt",
device=intermediary_device,
) as f:
for key in f.keys():
tensors[key] = None
for key in tensors.keys():
tensors[key] = SafetensorsLazyTensor(
checkpoint_file=checkpoint_file, key=key, location=intermediary_device,
checkpoint_file=checkpoint_file,
key=key,
location=intermediary_device,
)
if callback is not None:
@@ -462,13 +406,14 @@ def use_lazy_load(
enable=True,
callback: Optional[Callable] = None,
dematerialized_modules=False,
use_accelerate_init_empty_weights=False,
):
if not enable:
with use_custom_unpickler(RestrictedUnpickler):
yield False
return
begin_time = time.time()
try:
old_rebuild_tensor = torch._utils._rebuild_tensor
torch._utils._rebuild_tensor = _rebuild_tensor
@@ -477,22 +422,30 @@ def use_lazy_load(
old_torch_load = torch.load
def torch_load(f, map_location=None, pickle_module=pickle, **pickle_load_args):
retval = old_torch_load(
model_dict = old_torch_load(
f=f,
map_location=map_location,
pickle_module=pickle_module,
**pickle_load_args,
)
if f not in torch_checkpoint_file_handles:
torch_checkpoint_file_handles[f] = zipfile.ZipFile(f, "r")
for k, v in model_dict.items():
v.file_name = f
if callback is not None:
callback(
retval,
model_dict,
f=f,
map_location=map_location,
pickle_module=pickle_module,
is_safetensors=False,
**pickle_load_args,
)
return retval
return model_dict
torch.load = torch_load
@@ -500,30 +453,8 @@ def use_lazy_load(
patch_safetensors(callback)
if dematerialized_modules:
if use_accelerate_init_empty_weights:
import accelerate
init_empty_weights = accelerate.init_empty_weights()
init_empty_weights.__enter__()
else:
old_linear_init = torch.nn.Linear.__init__
old_embedding_init = torch.nn.Embedding.__init__
old_layernorm_init = torch.nn.LayerNorm.__init__
def linear_init(self, *args, device=None, **kwargs):
return old_linear_init(self, *args, device="meta", **kwargs)
def embedding_init(self, *args, device=None, **kwargs):
return old_embedding_init(self, *args, device="meta", **kwargs)
def layernorm_init(self, *args, device=None, **kwargs):
return old_layernorm_init(self, *args, device="meta", **kwargs)
torch.nn.Linear.__init__ = linear_init
torch.nn.Embedding.__init__ = embedding_init
torch.nn.LayerNorm.__init__ = layernorm_init
old_load_from_state_dict = torch.nn.Module._load_from_state_dict
torch.nn.Module._load_from_state_dict = _load_from_state_dict
init_empty_weights = accelerate.init_empty_weights()
init_empty_weights.__enter__()
with use_custom_unpickler(_LazyUnpickler):
yield True
@@ -531,11 +462,30 @@ def use_lazy_load(
finally:
torch._utils._rebuild_tensor = old_rebuild_tensor
torch.load = old_torch_load
post_load_cleanup()
logger.debug(
f"[lazy_load] Context closed in {round(time.time() - begin_time, 2)} seconds."
)
if dematerialized_modules:
if use_accelerate_init_empty_weights:
init_empty_weights.__exit__(None, None, None)
else:
torch.nn.Linear.__init__ = old_linear_init
torch.nn.Embedding.__init__ = old_embedding_init
torch.nn.LayerNorm.__init__ = old_layernorm_init
torch.nn.Module._load_from_state_dict = old_load_from_state_dict
init_empty_weights.__exit__(None, None, None)
def post_load_cleanup() -> None:
"""Close dangling file pointers and clear caches after the load is complete."""
global torch_checkpoint_file_handles
logger.debug(
f"[lazy_load] CheckpointChunkCache Hit Data: {CheckpointChunkCache.hit_data}"
)
CheckpointChunkCache.clear(unload_model=True)
# Bar is initialized in
# patches.patch_transformers_for_lazyload._load_state_dict_into_meta_model,
# as it has access to the state dict (for getting tensor count)
utils.bar = None
for v in torch_checkpoint_file_handles.values():
v.close()
torch_checkpoint_file_handles = {}

185
modeling/patches.py
View File

@@ -2,7 +2,7 @@ from __future__ import annotations
import copy
import requests
from typing import Iterable, List
from typing import List
from tqdm.auto import tqdm
import transformers
@@ -10,6 +10,7 @@ from transformers import (
PreTrainedModel,
modeling_utils,
)
from modeling.lazy_loader import LazyTensor
import utils
@@ -125,9 +126,185 @@ def patch_transformers_generation() -> None:
transformers.generation.logits_process.NoBadWordsLogitsProcessor.__init__ = new_init
def patch_transformers() -> None:
def patch_transformers_for_lazyload() -> None:
"""
Most of the code is modified code from the Accelerate and Transformers
projects, made by HuggingFace. The license for these projects are as follows:
---
Copyright The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
"""
import torch
from accelerate.utils import set_module_tensor_to_device, offload_weight
def _load_state_dict_into_meta_model(
model,
state_dict,
loaded_state_dict_keys,
start_prefix,
expected_keys,
device_map=None,
offload_folder=None,
offload_index=None,
state_dict_folder=None,
state_dict_index=None,
dtype=None,
# PATCH: load_in_8bit was renamed to is_quantized in Transformers 4.30, keep
# both for short term compatibility
load_in_8bit=False,
is_quantized=False,
is_safetensors=False,
keep_in_fp32_modules=None,
):
is_quantized = is_quantized or load_in_8bit
if is_quantized:
from .utils.bitsandbytes import set_module_8bit_tensor_to_device
error_msgs = []
old_keys = []
new_keys = []
for key in state_dict.keys():
new_key = None
if "gamma" in key:
new_key = key.replace("gamma", "weight")
if "beta" in key:
new_key = key.replace("beta", "bias")
if new_key:
old_keys.append(key)
new_keys.append(new_key)
for old_key, new_key in zip(old_keys, new_keys):
state_dict[new_key] = state_dict.pop(old_key)
# BEGIN PATCH
utils.bar = tqdm(total=len(state_dict), desc="Loading model tensors", file=utils.UIProgressBarFile(), position=1)
for param_name, param in sorted(
state_dict.items(),
# State dict must be ordered in this manner to make the caching in
# lazy_loader.py effective
key=lambda x: (
x[1].key,
x[1].seek_offset,
),
):
if isinstance(param, LazyTensor):
# Should always be true
param = param.materialize(map_location="cpu")
utils.bar.update(1)
# END PATCH
# First part of the test is always true as load_state_dict_keys always contains state_dict keys.
if (
param_name not in loaded_state_dict_keys
or param_name not in expected_keys
):
continue
if param_name.startswith(start_prefix):
param_name = param_name[len(start_prefix) :]
module_name = param_name
# We convert floating dtypes to the `dtype` passed. We want to keep the buffers/params
# in int/uint/bool and not cast them.
if dtype is not None and torch.is_floating_point(param):
if (
keep_in_fp32_modules is not None
and any(
module_to_keep_in_fp32 in param_name
for module_to_keep_in_fp32 in keep_in_fp32_modules
)
and dtype == torch.float16
):
param = param.to(torch.float32)
else:
param = param.to(dtype)
# For compatibility with PyTorch load_state_dict which converts state dict dtype to existing dtype in model
if dtype is None:
old_param = model
splits = param_name.split(".")
for split in splits:
old_param = getattr(old_param, split)
if old_param is None:
break
if old_param is not None:
param = param.to(old_param.dtype)
if device_map is None:
param_device = "cpu"
else:
# find next higher level module that is defined in device_map:
# bert.lm_head.weight -> bert.lm_head -> bert -> ''
while len(module_name) > 0 and module_name not in device_map:
module_name = ".".join(module_name.split(".")[:-1])
if module_name == "" and "" not in device_map:
# TODO: group all errors and raise at the end.
raise ValueError(f"{param_name} doesn't have any device set.")
param_device = device_map[module_name]
if param_device == "disk":
if not is_safetensors:
offload_index = offload_weight(
param, param_name, offload_folder, offload_index
)
elif param_device == "cpu" and state_dict_index is not None:
state_dict_index = offload_weight(
param, param_name, state_dict_folder, state_dict_index
)
elif not is_quantized:
# For backward compatibility with older versions of `accelerate`
set_module_tensor_to_device(
model,
tensor_name=param_name,
device=param_device,
value=param,
dtype=dtype,
)
else:
if (
param.dtype == torch.int8
and param_name.replace("weight", "SCB") in state_dict.keys()
):
fp16_statistics = state_dict[param_name.replace("weight", "SCB")]
else:
fp16_statistics = None
if "SCB" not in param_name:
set_module_8bit_tensor_to_device(
model,
param_name,
param_device,
value=param,
fp16_statistics=fp16_statistics,
)
return error_msgs, offload_index, state_dict_index
transformers.modeling_utils._load_state_dict_into_meta_model = (
_load_state_dict_into_meta_model
)
def patch_transformers(use_tpu: bool) -> None:
patch_transformers_download()
patch_transformers_loader()
# Doesn't do anything for TPU
patch_transformers_generation()
if not use_tpu:
patch_transformers_generation()
patch_transformers_for_lazyload()

2
prompt_tuner.py
View File

@@ -855,7 +855,7 @@ class TrainerBase(abc.ABC):
lazy_load_callback.nested = False
# Since we're using lazy loader, we need to figure out what the model's hidden layers are called
with lazy_loader.use_lazy_load(dematerialized_modules=True, use_accelerate_init_empty_weights=True):
with lazy_loader.use_lazy_load(dematerialized_modules=True):
try:
metamodel = AutoModelForCausalLM.from_config(model_config)
except Exception as e:

3
requirements_mtj.txt
View File

@@ -1,4 +1,4 @@
torch >= 1.9, < 1.13
torch == 2.0
numpy
tqdm
requests
@@ -34,3 +34,4 @@ ijson
ftfy
pydub
sentencepiece
accelerate==0.18.0

225
tpu_mtj_backend.py
View File

@@ -1171,139 +1171,118 @@ def load_model(path: str, model_type: str, badwordsids=koboldai_settings.badword
if callback.nested:
return
callback.nested = True
with zipfile.ZipFile(f, "r") as z:
try:
last_storage_key = None
zipfolder = os.path.basename(os.path.normpath(f)).split('.')[0]
f = None
current_offset = 0
if utils.current_shard == 0:
print("\n\n\nThis model has ", f"{hk.data_structures.tree_size(network.state['params']):,d}".replace(",", " "), " parameters.\n")
if utils.num_shards is None or utils.current_shard == 0:
if utils.num_shards is not None:
num_tensors = len(utils.get_sharded_checkpoint_num_tensors(utils.from_pretrained_model_name, utils.from_pretrained_index_filename, **utils.from_pretrained_kwargs))
else:
num_tensors = len(model_dict)
if socketio is None:
utils.bar = tqdm(total=num_tensors, desc="Loading model tensors")
else:
utils.bar = tqdm(total=num_tensors, desc="Loading model tensors", file=utils.UIProgressBarFile(socketio.emit))
koboldai_vars.status_message = "Loading model"
koboldai_vars.loaded_layers = 0
koboldai_vars.total_layers = num_tensors
try:
if utils.current_shard == 0:
print("\n\n\nThis model has ", f"{hk.data_structures.tree_size(network.state['params']):,d}".replace(",", " "), " parameters.\n")
if utils.num_shards is None or utils.current_shard == 0:
if utils.num_shards is not None:
utils.current_shard += 1
for key in sorted(model_dict.keys(), key=lambda k: (model_dict[k].key, model_dict[k].seek_offset)):
model_spec_key = max((k for k in model_spec.keys() if key.endswith(k)), key=len, default=None)
num_tensors = len(utils.get_sharded_checkpoint_num_tensors(utils.from_pretrained_model_name, utils.from_pretrained_index_filename, **utils.from_pretrained_kwargs))
else:
num_tensors = len(model_dict)
# Some model weights are used by transformers but not by MTJ.
# We have to materialize these weights anyways because
# transformers will throw a tantrum otherwise. To attain
# the least possible memory usage, we create them as meta
# tensors, which don't take up any actual CPU or TPU memory.
if model_spec_key is None:
model_dict[key] = torch.empty(model_dict[key].shape, dtype=model_dict[key].dtype, device="meta")
utils.bar.update(1)
koboldai_vars.loaded_layers += 1
continue
if socketio is None:
utils.bar = tqdm(total=num_tensors, desc="Loading model tensors")
else:
utils.bar = tqdm(total=num_tensors, desc="Loading model tensors", file=utils.UIProgressBarFile(socketio.emit))
koboldai_vars.status_message = "Loading model"
koboldai_vars.loaded_layers = 0
koboldai_vars.total_layers = num_tensors
storage_key = model_dict[key].key
if storage_key != last_storage_key or model_dict[key].seek_offset < current_offset:
last_storage_key = storage_key
if isinstance(f, zipfile.ZipExtFile):
f.close()
try:
f = z.open(f"archive/data/{storage_key}")
except:
f = z.open(f"{zipfolder}/data/{storage_key}")
current_offset = 0
if current_offset != model_dict[key].seek_offset:
f.read(model_dict[key].seek_offset - current_offset)
current_offset = model_dict[key].seek_offset
spec = model_spec[model_spec_key]
transforms = set(spec.get("transforms", ()))
if not isinstance(model_dict[key], lazy_loader.LazyTensor):
error = f"Duplicate key {repr(key)}"
print("\n\nERROR: " + error, file=sys.stderr)
raise RuntimeError(error)
size = functools.reduce(lambda x, y: x * y, model_dict[key].shape, 1)
dtype = model_dict[key].dtype
nbytes = size if dtype is torch.bool else size * ((torch.finfo if dtype.is_floating_point else torch.iinfo)(dtype).bits >> 3)
tensor = model_dict[key].materialize(f, map_location="cpu")
model_dict[key] = tensor.to("meta")
current_offset += nbytes
if utils.num_shards is not None:
utils.current_shard += 1
# MTJ requires certain mathematical operations to be performed
# on tensors in order for them to be in the correct format
if "remove_first_two_rows" in transforms:
tensor = tensor[2:]
if "divide_by_shards" in transforms:
tensor /= params["cores_per_replica"]
if "vocab_pad" in transforms:
tensor = torch.nn.functional.pad(tensor, (0,) * (tensor.ndim * 2 - 1) + (params["n_vocab_padding"],))
# We don't need to transpose linear module weights anymore because MTJ will do it for us if `transposed_linear` is set to True in the config
#if "no_transpose" not in transforms and tensor.ndim == 2:
# tensor = tensor.T
tensor.unsqueeze_(0)
for key in sorted(model_dict.keys(), key=lambda k: (model_dict[k].key, model_dict[k].seek_offset)):
model_spec_key = max((k for k in model_spec.keys() if key.endswith(k)), key=len, default=None)
# Shard the tensor so that parts of the tensor can be used
# on different TPU cores
tensor = reshard_reverse(
tensor,
params["cores_per_replica"],
network.state["params"][spec["module"]][spec["param"]].shape,
)
tensor = tensor.detach()
# numpy does not support bfloat16
if tensor.dtype is torch.bfloat16:
tensor = tensor.to(torch.float32)
tensor = jnp.array(tensor)
if tensor.dtype is torch.float16 or tensor.dtype is torch.float32:
tensor = tensor.bfloat16()
network.state["params"][spec["module"]][spec["param"]] = move_xmap(
tensor,
np.empty(params["cores_per_replica"]),
)
koboldai_vars.loaded_layers += 1
try:
time.sleep(0.01)
except:
pass
# Some model weights are used by transformers but not by MTJ.
# We have to materialize these weights anyways because
# transformers will throw a tantrum otherwise. To attain
# the least possible memory usage, we create them as meta
# tensors, which don't take up any actual CPU or TPU memory.
if model_spec_key is None:
model_dict[key] = torch.empty(model_dict[key].shape, dtype=model_dict[key].dtype, device="meta")
utils.bar.update(1)
koboldai_vars.loaded_layers += 1
continue
if utils.num_shards is not None and utils.current_shard < utils.num_shards:
return
spec = model_spec[model_spec_key]
transforms = set(spec.get("transforms", ()))
# Check for tensors that MTJ needs that were not provided in the
# HF model
for mk, mv in network.state["params"].items():
for pk, pv in mv.items():
if isinstance(pv, PlaceholderTensor):
# The transformers GPT-J models apparently do not
# have embedding bias, whereas MTJ GPT-J models do,
# so we have to supplement an embedding bias tensor
# by creating a tensor with the necessary shape, filled
# with zeros.
if mk == "causal_transformer_shard/~/embedding_shard/~/linear" and pk == "b":
mv[pk] = move_xmap(jnp.zeros(mv[pk].shape, dtype=jnp.bfloat16), np.empty(params["cores_per_replica"]))
if not isinstance(model_dict[key], lazy_loader.LazyTensor):
error = f"Duplicate key {repr(key)}"
print("\n\nERROR: " + error, file=sys.stderr)
raise RuntimeError(error)
else:
error = f"{mk} {pk} could not be found in the model checkpoint"
print("\n\nERROR: " + error, file=sys.stderr)
raise RuntimeError(error)
finally:
if utils.num_shards is None or utils.current_shard >= utils.num_shards:
utils.bar.close()
utils.bar = None
koboldai_vars.status_message = ""
callback.nested = False
if isinstance(f, zipfile.ZipExtFile):
f.close()
tensor = model_dict[key].materialize(map_location="cpu")
model_dict[key] = tensor.to("meta")
# MTJ requires certain mathematical operations to be performed
# on tensors in order for them to be in the correct format
if "remove_first_two_rows" in transforms:
tensor = tensor[2:]
if "divide_by_shards" in transforms:
tensor /= params["cores_per_replica"]
if "vocab_pad" in transforms:
tensor = torch.nn.functional.pad(tensor, (0,) * (tensor.ndim * 2 - 1) + (params["n_vocab_padding"],))
# We don't need to transpose linear module weights anymore because MTJ will do it for us if `transposed_linear` is set to True in the config
#if "no_transpose" not in transforms and tensor.ndim == 2:
# tensor = tensor.T
tensor.unsqueeze_(0)
# Shard the tensor so that parts of the tensor can be used
# on different TPU cores
tensor = reshard_reverse(
tensor,
params["cores_per_replica"],
network.state["params"][spec["module"]][spec["param"]].shape,
)
tensor = tensor.detach()
# numpy does not support bfloat16
if tensor.dtype is torch.bfloat16:
tensor = tensor.to(torch.float32)
tensor = jnp.array(tensor)
if tensor.dtype is torch.float16 or tensor.dtype is torch.float32:
tensor = tensor.bfloat16()
network.state["params"][spec["module"]][spec["param"]] = move_xmap(
tensor,
np.empty(params["cores_per_replica"]),
)
koboldai_vars.loaded_layers += 1
try:
time.sleep(0.01)
except:
pass
utils.bar.update(1)
if utils.num_shards is not None and utils.current_shard < utils.num_shards:
return
# Check for tensors that MTJ needs that were not provided in the
# HF model
for mk, mv in network.state["params"].items():
for pk, pv in mv.items():
if isinstance(pv, PlaceholderTensor):
# The transformers GPT-J models apparently do not
# have embedding bias, whereas MTJ GPT-J models do,
# so we have to supplement an embedding bias tensor
# by creating a tensor with the necessary shape, filled
# with zeros.
if mk == "causal_transformer_shard/~/embedding_shard/~/linear" and pk == "b":
mv[pk] = move_xmap(jnp.zeros(mv[pk].shape, dtype=jnp.bfloat16), np.empty(params["cores_per_replica"]))
else:
error = f"{mk} {pk} could not be found in the model checkpoint"
print("\n\nERROR: " + error, file=sys.stderr)
raise RuntimeError(error)
finally:
if utils.num_shards is None or utils.current_shard >= utils.num_shards:
utils.bar.close()
utils.bar = None
koboldai_vars.status_message = ""
callback.nested = False
callback.nested = False
if os.path.isdir(koboldai_vars.model.replace('/', '_')):

11
utils.py
View File

@@ -650,17 +650,6 @@ class UIProgressBarFile(object):
def flush(self):
pass
def get_auxilary_device():
"""Get device auxilary tensors like inputs should be stored on."""
# NOTE: TPU isn't a torch device, so TPU stuff gets sent to CPU.
if koboldai_vars.hascuda and koboldai_vars.usegpu:
return koboldai_vars.gpu_device
elif koboldai_vars.hascuda and koboldai_vars.breakmodel:
import breakmodel
return breakmodel.primary_device
return "cpu"
#==================================================================#
# Strips submitted text from the text returned by the AI
#==================================================================#