rixty
/
KoboldAI-Client
mirror of https://github.com/KoboldAI/KoboldAI-Client.git
1
1
Fork 0
Code Issues Projects Releases Wiki Activity

Tokenizer Fixes (Slow first to keep coherency)

This commit is contained in:
Henk 2022-09-27 17:57:18 +02:00
parent 07896867b2
commit 11455697ef
2 changed files with 30 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
aiserver.py
View File

@ -2489,12 +2489,11 @@ def load_model(use_gpu=True, gpu_layers=None, disk_layers=None, initial_load=Fal
if(vars.lazy_load): # torch_lazy_loader.py and low_cpu_mem_usage can't be used at the same time if(vars.lazy_load): # torch_lazy_loader.py and low_cpu_mem_usage can't be used at the same time
lowmem = {} lowmem = {}
if(os.path.isdir(vars.custmodpth)): if(os.path.isdir(vars.custmodpth)):
try:
tokenizer = AutoTokenizer.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache")
except Exception as e:
pass
try: try:
tokenizer = AutoTokenizer.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache", use_fast=False) tokenizer = AutoTokenizer.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache", use_fast=False)
except Exception as e:
try:
tokenizer = AutoTokenizer.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache")
except Exception as e: except Exception as e:
try: try:
tokenizer = GPT2TokenizerFast.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache") tokenizer = GPT2TokenizerFast.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache")
@ -2507,12 +2506,11 @@ def load_model(use_gpu=True, gpu_layers=None, disk_layers=None, initial_load=Fal
raise RuntimeError("One of your GPUs ran out of memory when KoboldAI tried to load your model.") raise RuntimeError("One of your GPUs ran out of memory when KoboldAI tried to load your model.")
model = GPTNeoForCausalLM.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache", **lowmem) model = GPTNeoForCausalLM.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache", **lowmem)
elif(os.path.isdir("models/{}".format(vars.model.replace('/', '_')))): elif(os.path.isdir("models/{}".format(vars.model.replace('/', '_')))):
try:
tokenizer = AutoTokenizer.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache")
except Exception as e:
pass
try: try:
tokenizer = AutoTokenizer.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache", use_fast=False) tokenizer = AutoTokenizer.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache", use_fast=False)
except Exception as e:
try:
tokenizer = AutoTokenizer.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache")
except Exception as e: except Exception as e:
try: try:
tokenizer = GPT2TokenizerFast.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache") tokenizer = GPT2TokenizerFast.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache")
@ -2538,12 +2536,11 @@ def load_model(use_gpu=True, gpu_layers=None, disk_layers=None, initial_load=Fal
return old_rebuild_tensor(storage, storage_offset, shape, stride) return old_rebuild_tensor(storage, storage_offset, shape, stride)
torch._utils._rebuild_tensor = new_rebuild_tensor torch._utils._rebuild_tensor = new_rebuild_tensor
try:
tokenizer = AutoTokenizer.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache")
except Exception as e:
pass
try: try:
tokenizer = AutoTokenizer.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache", use_fast=False) tokenizer = AutoTokenizer.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache", use_fast=False)
except Exception as e:
try:
tokenizer = AutoTokenizer.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache")
except Exception as e: except Exception as e:
try: try:
tokenizer = GPT2TokenizerFast.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache") tokenizer = GPT2TokenizerFast.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache")

21
tpu_mtj_backend.py
View File

@ -1350,12 +1350,11 @@ def load_model(path: str, driver_version="tpu_driver0.1_dev20210607", hf_checkpo
print("\n", flush=True) print("\n", flush=True)
with torch_lazy_loader.use_lazy_torch_load(callback=callback, dematerialized_modules=True): with torch_lazy_loader.use_lazy_torch_load(callback=callback, dematerialized_modules=True):
if(os.path.isdir(vars.custmodpth)): if(os.path.isdir(vars.custmodpth)):
try:
tokenizer = AutoTokenizer.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache")
except Exception as e:
pass
try: try:
tokenizer = AutoTokenizer.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache", use_fast=False) tokenizer = AutoTokenizer.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache", use_fast=False)
except Exception as e:
try:
tokenizer = AutoTokenizer.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache")
except Exception as e: except Exception as e:
try: try:
tokenizer = GPT2TokenizerFast.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache") tokenizer = GPT2TokenizerFast.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache")
@ -1366,12 +1365,11 @@ def load_model(path: str, driver_version="tpu_driver0.1_dev20210607", hf_checkpo
except Exception as e: except Exception as e:
model = GPTNeoForCausalLM.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache") model = GPTNeoForCausalLM.from_pretrained(vars.custmodpth, revision=vars.revision, cache_dir="cache")
elif(os.path.isdir("models/{}".format(vars.model.replace('/', '_')))): elif(os.path.isdir("models/{}".format(vars.model.replace('/', '_')))):
try:
tokenizer = AutoTokenizer.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache")
except Exception as e:
pass
try: try:
tokenizer = AutoTokenizer.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache", use_fast=False) tokenizer = AutoTokenizer.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache", use_fast=False)
except Exception as e:
try:
tokenizer = AutoTokenizer.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache")
except Exception as e: except Exception as e:
try: try:
tokenizer = GPT2TokenizerFast.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache") tokenizer = GPT2TokenizerFast.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache")
@ -1382,12 +1380,11 @@ def load_model(path: str, driver_version="tpu_driver0.1_dev20210607", hf_checkpo
except Exception as e: except Exception as e:
model = GPTNeoForCausalLM.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache") model = GPTNeoForCausalLM.from_pretrained("models/{}".format(vars.model.replace('/', '_')), revision=vars.revision, cache_dir="cache")
else: else:
try:
tokenizer = AutoTokenizer.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache")
except Exception as e:
pass
try: try:
tokenizer = AutoTokenizer.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache", use_fast=False) tokenizer = AutoTokenizer.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache", use_fast=False)
except Exception as e:
try:
tokenizer = AutoTokenizer.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache")
except Exception as e: except Exception as e:
try: try:
tokenizer = GPT2TokenizerFast.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache") tokenizer = GPT2TokenizerFast.from_pretrained(vars.model, revision=vars.revision, cache_dir="cache")