mirror/OpenLLM
1
0
Fork 0
mirror of https://github.com/bentoml/OpenLLM.git synced 2026-06-12 18:39:16 -04:00
Code Issues Packages Projects Releases Wiki Activity

chore(style): synchronized style across packages [skip ci]

Browse Source 
Signed-off-by: Aaron <29749331+aarnphm@users.noreply.github.com>
This commit is contained in:
Aaron
2023-08-23 08:46:22 -04:00
parent bbd9aa7646
commit 787ce1b3b6
124 changed files with 2775 additions and 2771 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
openllm-python/src/openllm/models/opt/__init__.py
View File

@@ -9,28 +9,28 @@ try:
except MissingDependencyError:
pass
else:
_import_structure["modeling_opt"] = ["OPT"]
_import_structure['modeling_opt'] = ['OPT']
if t.TYPE_CHECKING: from .modeling_opt import OPT as OPT
try:
if not is_flax_available(): raise MissingDependencyError
except MissingDependencyError:
pass
else:
_import_structure["modeling_flax_opt"] = ["FlaxOPT"]
_import_structure['modeling_flax_opt'] = ['FlaxOPT']
if t.TYPE_CHECKING: from .modeling_flax_opt import FlaxOPT as FlaxOPT
try:
if not is_vllm_available(): raise MissingDependencyError
except MissingDependencyError:
pass
else:
_import_structure["modeling_vllm_opt"] = ["VLLMOPT"]
_import_structure['modeling_vllm_opt'] = ['VLLMOPT']
if t.TYPE_CHECKING: from .modeling_vllm_opt import VLLMOPT as VLLMOPT
try:
if not is_tf_available(): raise MissingDependencyError
except MissingDependencyError:
pass
else:
_import_structure["modeling_tf_opt"] = ["TFOPT"]
_import_structure['modeling_tf_opt'] = ['TFOPT']
if t.TYPE_CHECKING: from .modeling_tf_opt import TFOPT as TFOPT
sys.modules[__name__] = LazyModule(__name__, globals()["__file__"], _import_structure)
sys.modules[__name__] = LazyModule(__name__, globals()['__file__'], _import_structure)

10
openllm-python/src/openllm/models/opt/modeling_flax_opt.py
View File

@@ -4,17 +4,17 @@ from openllm._prompt import process_prompt
from openllm.utils import generate_labels
from openllm_core.config.configuration_opt import DEFAULT_PROMPT_TEMPLATE
if t.TYPE_CHECKING: import transformers
else: transformers = openllm.utils.LazyLoader("transformers", globals(), "transformers")
else: transformers = openllm.utils.LazyLoader('transformers', globals(), 'transformers')
logger = logging.getLogger(__name__)
class FlaxOPT(openllm.LLM["transformers.TFOPTForCausalLM", "transformers.GPT2Tokenizer"]):
class FlaxOPT(openllm.LLM['transformers.TFOPTForCausalLM', 'transformers.GPT2Tokenizer']):
__openllm_internal__ = True
def import_model(self, *args: t.Any, trust_remote_code: bool = False, **attrs: t.Any) -> bentoml.Model:
config, tokenizer = transformers.AutoConfig.from_pretrained(self.model_id), transformers.AutoTokenizer.from_pretrained(self.model_id, **self.llm_parameters[-1])
tokenizer.pad_token_id = config.pad_token_id
return bentoml.transformers.save_model(
self.tag, transformers.FlaxAutoModelForCausalLM.from_pretrained(self.model_id, **attrs), custom_objects={"tokenizer": tokenizer}, labels=generate_labels(self)
self.tag, transformers.FlaxAutoModelForCausalLM.from_pretrained(self.model_id, **attrs), custom_objects={'tokenizer': tokenizer}, labels=generate_labels(self)
)
def sanitize_parameters(
@@ -29,11 +29,11 @@ class FlaxOPT(openllm.LLM["transformers.TFOPTForCausalLM", "transformers.GPT2Tok
**attrs: t.Any
) -> tuple[str, dict[str, t.Any], dict[str, t.Any]]:
return process_prompt(prompt, DEFAULT_PROMPT_TEMPLATE, use_default_prompt_template, **attrs), {
"max_new_tokens": max_new_tokens, "temperature": temperature, "top_k": top_k, "num_return_sequences": num_return_sequences, "repetition_penalty": repetition_penalty
'max_new_tokens': max_new_tokens, 'temperature': temperature, 'top_k': top_k, 'num_return_sequences': num_return_sequences, 'repetition_penalty': repetition_penalty
}, {}
def generate(self, prompt: str, **attrs: t.Any) -> list[str]:
return self.tokenizer.batch_decode(
self.model.generate(**self.tokenizer(prompt, return_tensors="np"), do_sample=True, generation_config=self.config.model_construct_env(**attrs).to_generation_config()).sequences,
self.model.generate(**self.tokenizer(prompt, return_tensors='np'), do_sample=True, generation_config=self.config.model_construct_env(**attrs).to_generation_config()).sequences,
skip_special_tokens=True
)

6
openllm-python/src/openllm/models/opt/modeling_opt.py
View File

@@ -3,18 +3,18 @@ import logging, typing as t, openllm
if t.TYPE_CHECKING: import transformers
logger = logging.getLogger(__name__)
class OPT(openllm.LLM["transformers.OPTForCausalLM", "transformers.GPT2Tokenizer"]):
class OPT(openllm.LLM['transformers.OPTForCausalLM', 'transformers.GPT2Tokenizer']):
__openllm_internal__ = True
@property
def import_kwargs(self) -> tuple[dict[str, t.Any], dict[str, t.Any]]:
import torch
return {"torch_dtype": torch.float16 if torch.cuda.is_available() else torch.float32}, {}
return {'torch_dtype': torch.float16 if torch.cuda.is_available() else torch.float32}, {}
def generate(self, prompt: str, **attrs: t.Any) -> list[str]:
import torch
with torch.inference_mode():
return self.tokenizer.batch_decode(
self.model.generate(**self.tokenizer(prompt, return_tensors="pt").to(self.device), do_sample=True, generation_config=self.config.model_construct_env(**attrs).to_generation_config()),
self.model.generate(**self.tokenizer(prompt, return_tensors='pt').to(self.device), do_sample=True, generation_config=self.config.model_construct_env(**attrs).to_generation_config()),
skip_special_tokens=True
)

6
openllm-python/src/openllm/models/opt/modeling_tf_opt.py
View File

@@ -2,7 +2,7 @@ from __future__ import annotations
import typing as t, bentoml, openllm
from openllm_core.utils import generate_labels
if t.TYPE_CHECKING: import transformers
class TFOPT(openllm.LLM["transformers.TFOPTForCausalLM", "transformers.GPT2Tokenizer"]):
class TFOPT(openllm.LLM['transformers.TFOPTForCausalLM', 'transformers.GPT2Tokenizer']):
__openllm_internal__ = True
def import_model(self, *args: t.Any, trust_remote_code: bool = False, **attrs: t.Any) -> bentoml.Model:
@@ -12,12 +12,12 @@ class TFOPT(openllm.LLM["transformers.TFOPTForCausalLM", "transformers.GPT2Token
return bentoml.transformers.save_model(
self.tag,
transformers.TFOPTForCausalLM.from_pretrained(self.model_id, trust_remote_code=trust_remote_code, **attrs),
custom_objects={"tokenizer": tokenizer},
custom_objects={'tokenizer': tokenizer},
labels=generate_labels(self)
)
def generate(self, prompt: str, **attrs: t.Any) -> list[str]:
return self.tokenizer.batch_decode(
self.model.generate(**self.tokenizer(prompt, return_tensors="tf"), do_sample=True, generation_config=self.config.model_construct_env(**attrs).to_generation_config()),
self.model.generate(**self.tokenizer(prompt, return_tensors='tf'), do_sample=True, generation_config=self.config.model_construct_env(**attrs).to_generation_config()),
skip_special_tokens=True
)

6
openllm-python/src/openllm/models/opt/modeling_vllm_opt.py
View File

@@ -3,9 +3,9 @@ import typing as t, openllm
from openllm_core._prompt import process_prompt
from openllm_core.config.configuration_opt import DEFAULT_PROMPT_TEMPLATE
if t.TYPE_CHECKING: import vllm, transformers
class VLLMOPT(openllm.LLM["vllm.LLMEngine", "transformers.GPT2Tokenizer"]):
class VLLMOPT(openllm.LLM['vllm.LLMEngine', 'transformers.GPT2Tokenizer']):
__openllm_internal__ = True
tokenizer_id = "local"
tokenizer_id = 'local'
def sanitize_parameters(
self,
@@ -18,5 +18,5 @@ class VLLMOPT(openllm.LLM["vllm.LLMEngine", "transformers.GPT2Tokenizer"]):
**attrs: t.Any
) -> tuple[str, dict[str, t.Any], dict[str, t.Any]]:
return process_prompt(prompt, DEFAULT_PROMPT_TEMPLATE, use_default_prompt_template, **attrs), {
"max_new_tokens": max_new_tokens, "temperature": temperature, "top_k": top_k, "num_return_sequences": num_return_sequences
'max_new_tokens': max_new_tokens, 'temperature': temperature, 'top_k': top_k, 'num_return_sequences': num_return_sequences
}, {}