Latest Threat Research:SANDWORM_MODE: Shai-Hulud-Style npm Worm Hijacks CI Workflows and Poisons AI Toolchains.Details →
Socket
Book a DemoSign in
Socket
Book a DemoSign in

tgi

Package Overview
Dependencies
Maintainers
1
Versions
7
Alerts
File Explorer

Advanced tools

License
Socket logo

Install Socket

Detect and block malicious and high-risk dependencies

Install

tgi - pypi Package Compare versions

Comparing version
1.4.3
 to
1.4.4
+32
-4
PKG-INFO
Metadata-Version: 2.1
Name: tgi
Version: 1.4.3
Version: 1.4.4
Summary: Nightly release of Hugging Face Text Generation Python Client

@@ -22,3 +22,3 @@ Home-page: https://github.com/huggingface/text-generation-inference

Requires-Dist: huggingface-hub (>=0.12,<1.0)
Requires-Dist: pydantic (>1.10,<3)
Requires-Dist: pydantic (>2,<3)
Project-URL: Repository, https://github.com/huggingface/text-generation-inference

@@ -136,3 +136,15 @@ Description-Content-Type: text/markdown

```python
# Request Parameters
# enum for grammar type
class GrammarType(Enum):
Json = "json"
Regex = "regex"
# Grammar type and value
class Grammar:
# Grammar type
type: GrammarType
# Grammar value
value: Union[str, dict]
class Parameters:

@@ -146,2 +158,6 @@ # Activate logits sampling

repetition_penalty: Optional[float]
# The parameter for frequency penalty. 1.0 means no penalty
# Penalize new tokens based on their existing frequency in the text so far,
# decreasing the model's likelihood to repeat the same line verbatim.
frequency_penalty: Optional[float]
# Whether to prepend the prompt to the generated text

@@ -169,2 +185,4 @@ return_full_text: bool

watermark: bool
# Get generation details
details: bool
# Get decoder input token logprobs and ids

@@ -174,3 +192,13 @@ decoder_input_details: bool

top_n_tokens: Optional[int]
# grammar to use for generation
grammar: Optional[Grammar]
class Request:
# Prompt
inputs: str
# Generation parameters
parameters: Optional[Parameters]
# Whether to stream output tokens
stream: bool
# Decoder input tokens

@@ -194,3 +222,3 @@ class InputToken:

# Logprob
logprob: float
logprob: Optional[float]
# Is the token a special token

@@ -197,0 +225,0 @@ # Can be used to ignore tokens when concatenating

+2
-2
pyproject.toml
[tool.poetry]
name = "tgi"
version = "1.4.3"
version = "1.4.4"
description = "Nightly release of Hugging Face Text Generation Python Client"

@@ -15,3 +15,3 @@ license = "Apache-2.0"

python = "^3.7"
pydantic = "> 1.10, < 3"
pydantic = "> 2, < 3"
aiohttp = "^3.8"

@@ -18,0 +18,0 @@ huggingface-hub = ">= 0.12, < 1.0"

+30
-2
README.md

@@ -110,3 +110,15 @@ # Text Generation

```python
# Request Parameters
# enum for grammar type
class GrammarType(Enum):
Json = "json"
Regex = "regex"
# Grammar type and value
class Grammar:
# Grammar type
type: GrammarType
# Grammar value
value: Union[str, dict]
class Parameters:

@@ -120,2 +132,6 @@ # Activate logits sampling

repetition_penalty: Optional[float]
# The parameter for frequency penalty. 1.0 means no penalty
# Penalize new tokens based on their existing frequency in the text so far,
# decreasing the model's likelihood to repeat the same line verbatim.
frequency_penalty: Optional[float]
# Whether to prepend the prompt to the generated text

@@ -143,2 +159,4 @@ return_full_text: bool

watermark: bool
# Get generation details
details: bool
# Get decoder input token logprobs and ids

@@ -148,3 +166,13 @@ decoder_input_details: bool

top_n_tokens: Optional[int]
# grammar to use for generation
grammar: Optional[Grammar]
class Request:
# Prompt
inputs: str
# Generation parameters
parameters: Optional[Parameters]
# Whether to stream output tokens
stream: bool
# Decoder input tokens

@@ -168,3 +196,3 @@ class InputToken:

# Logprob
logprob: float
logprob: Optional[float]
# Is the token a special token

@@ -171,0 +199,0 @@ # Can be used to ignore tokens when concatenating

+39
-3
tgi/client.py

@@ -70,2 +70,3 @@ import json

messages: List[Message],
repetition_penalty: Optional[float] = None,
frequency_penalty: Optional[float] = None,

@@ -91,5 +92,9 @@ logit_bias: Optional[List[float]] = None,

List of messages
repetition_penalty (`float`):
The parameter for repetition penalty. 0.0 means no penalty. See [this
paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
frequency_penalty (`float`):
The parameter for frequency penalty. 0.0 means no penalty. See [this
paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
The parameter for frequency penalty. 0.0 means no penalty
Penalize new tokens based on their existing frequency in the text so far,
decreasing the model's likelihood to repeat the same line verbatim.
logit_bias (`List[float]`):

@@ -126,2 +131,3 @@ Adjust the likelihood of specified tokens

messages=messages,
repetition_penalty=repetition_penalty,
frequency_penalty=frequency_penalty,

@@ -185,2 +191,3 @@ logit_bias=logit_bias,

repetition_penalty: Optional[float] = None,
frequency_penalty: Optional[float] = None,
return_full_text: bool = False,

@@ -214,2 +221,6 @@ seed: Optional[int] = None,

paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
frequency_penalty (`float`):
The parameter for frequency penalty. 1.0 means no penalty
Penalize new tokens based on their existing frequency in the text so far,
decreasing the model's likelihood to repeat the same line verbatim.
return_full_text (`bool`):

@@ -253,2 +264,3 @@ Whether to prepend the prompt to the generated text

repetition_penalty=repetition_penalty,
frequency_penalty=frequency_penalty,
return_full_text=return_full_text,

@@ -287,2 +299,3 @@ seed=seed,

repetition_penalty: Optional[float] = None,
frequency_penalty: Optional[float] = None,
return_full_text: bool = False,

@@ -313,2 +326,6 @@ seed: Optional[int] = None,

paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
frequency_penalty (`float`):
The parameter for frequency penalty. 1.0 means no penalty
Penalize new tokens based on their existing frequency in the text so far,
decreasing the model's likelihood to repeat the same line verbatim.
return_full_text (`bool`):

@@ -351,2 +368,3 @@ Whether to prepend the prompt to the generated text

repetition_penalty=repetition_penalty,
frequency_penalty=frequency_penalty,
return_full_text=return_full_text,

@@ -447,2 +465,3 @@ seed=seed,

messages: List[Message],
repetition_penalty: Optional[float] = None,
frequency_penalty: Optional[float] = None,

@@ -468,5 +487,9 @@ logit_bias: Optional[List[float]] = None,

List of messages
frequency_penalty (`float`):
repetition_penalty (`float`):
The parameter for frequency penalty. 0.0 means no penalty. See [this
paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
frequency_penalty (`float`):
The parameter for frequency penalty. 0.0 means no penalty
Penalize new tokens based on their existing frequency in the text so far,
decreasing the model's likelihood to repeat the same line verbatim.
logit_bias (`List[float]`):

@@ -503,2 +526,3 @@ Adjust the likelihood of specified tokens

messages=messages,
repetition_penalty=repetition_penalty,
frequency_penalty=frequency_penalty,

@@ -561,2 +585,3 @@ logit_bias=logit_bias,

repetition_penalty: Optional[float] = None,
frequency_penalty: Optional[float] = None,
return_full_text: bool = False,

@@ -590,2 +615,6 @@ seed: Optional[int] = None,

paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
frequency_penalty (`float`):
The parameter for frequency penalty. 1.0 means no penalty
Penalize new tokens based on their existing frequency in the text so far,
decreasing the model's likelihood to repeat the same line verbatim.
return_full_text (`bool`):

@@ -631,2 +660,3 @@ Whether to prepend the prompt to the generated text

repetition_penalty=repetition_penalty,
frequency_penalty=frequency_penalty,
return_full_text=return_full_text,

@@ -662,2 +692,3 @@ seed=seed,

repetition_penalty: Optional[float] = None,
frequency_penalty: Optional[float] = None,
return_full_text: bool = False,

@@ -688,2 +719,6 @@ seed: Optional[int] = None,

paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
frequency_penalty (`float`):
The parameter for frequency penalty. 1.0 means no penalty
Penalize new tokens based on their existing frequency in the text so far,
decreasing the model's likelihood to repeat the same line verbatim.
return_full_text (`bool`):

@@ -726,2 +761,3 @@ Whether to prepend the prompt to the generated text

repetition_penalty=repetition_penalty,
frequency_penalty=frequency_penalty,
return_full_text=return_full_text,

@@ -728,0 +764,0 @@ seed=seed,

+39
-24
tgi/types.py
from enum import Enum
from pydantic import BaseModel, validator
from pydantic import BaseModel, field_validator
from typing import Optional, List, Union, Any

@@ -35,3 +35,3 @@

# Content of the message
content: Optional[str]
content: Optional[str] = None
# Optional name of the message sender

@@ -60,3 +60,3 @@ name: Optional[str] = None

# Usage details of the chat completion
usage: Any
usage: Optional[Any] = None

@@ -78,3 +78,3 @@

role: str
content: Optional[str]
content: Optional[str] = None
tool_calls: Optional[ChoiceDeltaToolCall]

@@ -115,3 +115,8 @@

messages: List[Message]
# Penalty for frequency of new tokens
# The parameter for repetition penalty. 1.0 means no penalty.
# See [this paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
repetition_penalty: Optional[float] = None
# The parameter for frequency penalty. 1.0 means no penalty
# Penalize new tokens based on their existing frequency in the text so far,
# decreasing the model's likelihood to repeat the same line verbatim.
frequency_penalty: Optional[float] = None

@@ -152,2 +157,6 @@ # Bias values for token selection

repetition_penalty: Optional[float] = None
# The parameter for frequency penalty. 1.0 means no penalty
# Penalize new tokens based on their existing frequency in the text so far,
# decreasing the model's likelihood to repeat the same line verbatim.
frequency_penalty: Optional[float] = None
# Whether to prepend the prompt to the generated text

@@ -184,3 +193,3 @@ return_full_text: bool = False

@validator("best_of")
@field_validator("best_of")
def valid_best_of(cls, field_value, values):

@@ -190,10 +199,10 @@ if field_value is not None:

raise ValidationError("`best_of` must be strictly positive")
if field_value > 1 and values["seed"] is not None:
if field_value > 1 and values.data["seed"] is not None:
raise ValidationError("`seed` must not be set when `best_of` is > 1")
sampling = (
values["do_sample"]
| (values["temperature"] is not None)
| (values["top_k"] is not None)
| (values["top_p"] is not None)
| (values["typical_p"] is not None)
values.data["do_sample"]
| (values.data["temperature"] is not None)
| (values.data["top_k"] is not None)
| (values.data["top_p"] is not None)
| (values.data["typical_p"] is not None)
)

@@ -205,3 +214,3 @@ if field_value > 1 and not sampling:

@validator("repetition_penalty")
@field_validator("repetition_penalty")
def valid_repetition_penalty(cls, v):

@@ -212,3 +221,9 @@ if v is not None and v <= 0:

@validator("seed")
@field_validator("frequency_penalty")
def valid_frequency_penalty(cls, v):
if v is not None and v <= 0:
raise ValidationError("`frequency_penalty` must be strictly positive")
return v
@field_validator("seed")
def valid_seed(cls, v):

@@ -219,3 +234,3 @@ if v is not None and v < 0:

@validator("temperature")
@field_validator("temperature")
def valid_temp(cls, v):

@@ -226,3 +241,3 @@ if v is not None and v <= 0:

@validator("top_k")
@field_validator("top_k")
def valid_top_k(cls, v):

@@ -233,3 +248,3 @@ if v is not None and v <= 0:

@validator("top_p")
@field_validator("top_p")
def valid_top_p(cls, v):

@@ -240,3 +255,3 @@ if v is not None and (v <= 0 or v >= 1.0):

@validator("truncate")
@field_validator("truncate")
def valid_truncate(cls, v):

@@ -247,3 +262,3 @@ if v is not None and v <= 0:

@validator("typical_p")
@field_validator("typical_p")
def valid_typical_p(cls, v):

@@ -254,3 +269,3 @@ if v is not None and (v <= 0 or v >= 1.0):

@validator("top_n_tokens")
@field_validator("top_n_tokens")
def valid_top_n_tokens(cls, v):

@@ -261,3 +276,3 @@ if v is not None and v <= 0:

@validator("grammar")
@field_validator("grammar")
def valid_grammar(cls, v):

@@ -280,3 +295,3 @@ if v is not None:

@validator("inputs")
@field_validator("inputs")
def valid_input(cls, v):

@@ -287,5 +302,5 @@ if not v:

@validator("stream")
@field_validator("stream")
def valid_best_of_stream(cls, field_value, values):
parameters = values["parameters"]
parameters = values.data["parameters"]
if (

@@ -292,0 +307,0 @@ parameters is not None