Skip to content

chatting module

Classes for chatting.

See vectorbtpro.utils.knowledge for the toy dataset.

memory_store dict

Object store by store id for MemoryStore.

complete function

complete(
    message,
    completions=None,
    **kwargs
)

Get completion for a message.

Resolves completions with resolve_completions. Keyword arguments are passed to either initialize a class or replace an instance of Completions.

def_metadata_template function

def_metadata_template(
    metadata_content
)

Default metadata template

detokenize function

detokenize(
    tokens,
    tokenizer=None,
    **kwargs
)

Detokenize text.

Resolves tokenizer with resolve_tokenizer. Keyword arguments are passed to either initialize a class or replace an instance of Tokenizer.

embed function

embed(
    query,
    embeddings=None,
    **kwargs
)

Get embedding(s) for one or more queries.

Resolves embeddings with resolve_embeddings. Keyword arguments are passed to either initialize a class or replace an instance of Embeddings.

embed_documents function

embed_documents(
    documents,
    refresh=False,
    refresh_documents=None,
    refresh_embeddings=None,
    return_embeddings=False,
    return_documents=False,
    doc_ranker=None,
    **kwargs
)

Embed documents.

Keyword arguments are passed to either initialize a class or replace an instance of DocumentRanker.

rank_documents function

rank_documents(
    query,
    documents=None,
    top_k=None,
    min_top_k=None,
    max_top_k=None,
    cutoff=None,
    refresh=False,
    refresh_documents=None,
    refresh_embeddings=None,
    return_chunks=False,
    return_scores=False,
    doc_ranker=None,
    **kwargs
)

Rank documents by their relevance to a query.

Keyword arguments are passed to either initialize a class or replace an instance of DocumentRanker.

resolve_completions function

resolve_completions(
    completions=None
)

Resolve a subclass or an instance of Completions.

The following values are supported:

resolve_embeddings function

resolve_embeddings(
    embeddings=None
)

Resolve a subclass or an instance of Embeddings.

The following values are supported:

resolve_obj_store function

resolve_obj_store(
    obj_store=None
)

Resolve a subclass or an instance of ObjectStore.

The following values are supported:

resolve_text_splitter function

resolve_text_splitter(
    text_splitter=None
)

Resolve a subclass or an instance of TextSplitter.

The following values are supported:

resolve_tokenizer function

resolve_tokenizer(
    tokenizer=None
)

Resolve a subclass or an instance of Tokenizer.

The following values are supported:

split_text function

split_text(
    text,
    text_splitter=None,
    **kwargs
)

Split text.

Resolves text_splitter with resolve_text_splitter. Keyword arguments are passed to either initialize a class or replace an instance of TextSplitter.

tokenize function

tokenize(
    text,
    tokenizer=None,
    **kwargs
)

Tokenize text.

Resolves tokenizer with resolve_tokenizer. Keyword arguments are passed to either initialize a class or replace an instance of Tokenizer.

CachedStore class

CachedStore(
    obj_store,
    lazy_open=None,
    mirror=None,
    **kwargs
)

Store class that acts as a (temporary) cache to another store.

For defaults, see chat.obj_store_configs.cached in knowledge.

Superclasses

Inherited members

force_open class property

Whether to open the store forcefully.

lazy_open class property

Whether to open the store lazily.

mirror class property

Whether to mirror the store in memory_store.

obj_store class property

Object store.

Completions class

Completions(
    context='',
    chat_history=None,
    stream=None,
    max_tokens=None,
    tokenizer=None,
    tokenizer_kwargs=None,
    system_prompt=None,
    system_as_user=None,
    context_prompt=None,
    formatter=None,
    formatter_kwargs=None,
    minimal_format=None,
    silence_warnings=None,
    template_context=None,
    **kwargs
)

Abstract class for completion providers.

For argument descriptions, see their properties, like Completions.chat_history.

For defaults, see knowledge.chat.completions_config in knowledge.

Superclasses

Inherited members

Subclasses

chat_history class property

Chat history.

Must be list of dictionaries with proper roles.

After generating a response, the output will be appended to this sequence as an assistant message.

context class property

Context.

Becomes a user message.

context_prompt class property

Context prompt.

A prompt template requiring the variable "context". The prompt can be either a custom template, or string or function that will become one. Once the prompt is evaluated, it becomes a user message.

formatter class property

A subclass or an instance of ContentFormatter.

Resolved with resolve_formatter.

formatter_kwargs class property

Keyword arguments passed to Completions.formatter.

Used either to initialize a class or replace an instance of ContentFormatter.

get_chat_response method

Completions.get_chat_response(
    messages,
    **kwargs
)

Get chat response to messages.

get_completion method

Completions.get_completion(
    message,
    return_response=False
)

Get completion for a message.

get_delta_content method

Completions.get_delta_content(
    response
)

Get content from a streaming response chunk.

get_message_content method

Completions.get_message_content(
    response
)

Get content from a chat response.

get_stream_response method

Completions.get_stream_response(
    messages,
    **kwargs
)

Get streaming response to messages.

max_tokens class property

Maximum number of tokens in messages.

max_tokens_set class property

Whether the user provided max_tokens.

minimal_format class property

Whether input is minimally-formatted.

model class property

Model.

prepare_messages method

Completions.prepare_messages(
    message
)

Prepare messages for a completion.

silence_warnings class property

Whether to silence warnings.

stream class property

Whether to stream the response.

When streaming, appends chunks one by one and displays the intermediate result. Otherwise, displays the entire message.

system_as_user class property

Whether to use the user role for the system message.

Mainly for experimental models where the system role is not available.

system_prompt class property

System prompt.

Precedes the context prompt.

template_context class property

Context used to substitute templates.

tokenizer class property

A subclass or an instance of Tokenizer.

Resolved with resolve_tokenizer.

tokenizer_kwargs class property

Keyword arguments passed to Completions.tokenizer.

Used either to initialize a class or replace an instance of Tokenizer.

Contextable class

Contextable()

Abstract class that can be converted into a context.

Superclasses

Inherited members

Subclasses

chat class method

Contextable.chat(
    message,
    chat_history=None,
    *,
    return_chat=False,
    **kwargs
)

Chat with an LLM while using the instance as a context.

Uses Contextable.create_chat and then Completions.get_completion.

Note

Context is recalculated each time this method is invoked. For multiple turns, it's more efficient to use Contextable.create_chat.

Usage

>>> asset.chat("What's the value under 'xyz'?")
The value under 'xyz' is 123.

>>> chat_history = []
>>> asset.chat("What's the value under 'xyz'?", chat_history=chat_history)
The value under 'xyz' is 123.

>>> asset.chat("Are you sure?", chat_history=chat_history)
Yes, I am sure. The value under 'xyz' is 123 for the entry where `s` is "EFG".

count_tokens method

Contextable.count_tokens(
    to_context_kwargs=None,
    tokenizer=None,
    tokenizer_kwargs=None
)

Count the number of tokens in the context.

create_chat method

Contextable.create_chat(
    to_context_kwargs=None,
    completions=None,
    **kwargs
)

Create a chat by returning an instance of Completions.

Uses Contextable.to_context to turn this instance to a context.

Usage

>>> chat = asset.create_chat()

>>> chat.get_completion("What's the value under 'xyz'?")
The value under 'xyz' is 123.

>>> chat.get_completion("Are you sure?")
Yes, I am sure. The value under 'xyz' is 123 for the entry where `s` is "EFG".

to_context method

Contextable.to_context(
    *args,
    **kwargs
)

Convert to a context.

DictStore class

DictStore(
    **kwargs
)

Store class based on a dictionary.

For defaults, see chat.obj_store_configs.memory in knowledge.

Superclasses

Inherited members

Subclasses

store class property

Store dictionary.

DocumentRanker class

DocumentRanker(
    dataset_id=None,
    embeddings=None,
    embeddings_kwargs=None,
    doc_store=None,
    doc_store_kwargs=None,
    cache_doc_store=None,
    emb_store=None,
    emb_store_kwargs=None,
    cache_emb_store=None,
    score_func=None,
    score_agg_func=None,
    show_progress=None,
    pbar_kwargs=None,
    template_context=None,
    **kwargs
)

Class for embedding, scoring, and ranking documents.

For defaults, see knowledge.chat.doc_ranker_config in knowledge.

Superclasses

Inherited members

compute_score method

DocumentRanker.compute_score(
    emb1,
    emb2
)

Compute scores between embeddings, which can be either single or multiple.

Supported distance functions are 'cosine', 'euclidean', and 'dot'. A metric can also be a callable that should take two and return one 2-dim NumPy array.

doc_store class property

An instance of ObjectStore for documents.

emb_store class property

An instance of ObjectStore for embeddings.

embed_documents method

DocumentRanker.embed_documents(
    documents,
    refresh=False,
    refresh_documents=None,
    refresh_embeddings=None,
    return_embeddings=False,
    return_documents=False
)

Embed documents.

Enable refresh or its sub-arguments to refresh documents and/or embeddings in their particular stores. Without refreshing, will rely on the persisted objects.

If return_embeddings and return_documents are both False, returns nothing. If return_embeddings and return_documents are both True, for each document, returns the document and either an embedding or a list of document chunks and their embeddings. If return_documents is False, returns only embeddings.

embeddings class property

An instance of Embeddings.

pbar_kwargs class property

Keyword arguments passed to ProgressBar.

rank_documents method

DocumentRanker.rank_documents(
    query,
    documents=None,
    top_k=None,
    min_top_k=None,
    max_top_k=None,
    cutoff=None,
    refresh=False,
    refresh_documents=None,
    refresh_embeddings=None,
    return_chunks=False,
    return_scores=False
)

Sort documents by relevance to a query.

Top-k, minimum top-k, and maximum top-k are resolved with DocumentRanker.resolve_top_k. Score cutoff is converted into top-k with DocumentRanker.top_k_from_cutoff. Minimum and maximum top-k are used to override non-integer top-k and cutoff; it has no effect on the integer top-k, which can be outside the top-k bounds and won't be overridden.

resolve_top_k class method

DocumentRanker.resolve_top_k(
    scores,
    top_k=None
)

Resolve top_k based on sorted scores.

Supported values are integers (top number), floats (top %), strings (supported methods are 'elbow' and 'kmeans'), as well as callables that should take a 1-dim NumPy array and return an integer or a float. Filters out NaN before computation (requires them to be at the tail).

score_agg_func class property

Score aggregation function.

score_documents method

DocumentRanker.score_documents(
    query,
    documents=None,
    refresh=False,
    refresh_documents=None,
    refresh_embeddings=None,
    return_chunks=False,
    return_documents=False
)

Score documents by relevance to a query.

score_func class property

Score function.

See DocumentRanker.compute_score.

show_progress class property

Whether to show progress bar.

template_context class property

Context used to substitute templates.

top_k_from_cutoff class method

DocumentRanker.top_k_from_cutoff(
    scores,
    cutoff=None
)

Get top_k from cutoff based on sorted scores.

EmbeddedDocument class

EmbeddedDocument(
    *args,
    **kwargs
)

Abstract class for embedded documents.

Superclasses

Inherited members

child_documents field

Embedded child documents.

document field

Document.

embedding field

Embedding.

Embeddings class

Embeddings(
    batch_size=None,
    show_progress=None,
    pbar_kwargs=None,
    template_context=None,
    **kwargs
)

Abstract class for embedding providers.

For defaults, see knowledge.chat.embeddings_config in knowledge.

Superclasses

Inherited members

Subclasses

batch_size class property

Batch size.

Set to None to disable batching.

get_embedding method

Embeddings.get_embedding(
    query
)

Get embedding for a query.

get_embedding_batch method

Embeddings.get_embedding_batch(
    batch
)

Get embeddings for one batch of queries.

get_embeddings method

Embeddings.get_embeddings(
    queries
)

Get embeddings for multiple queries.

iter_embedding_batches method

Embeddings.iter_embedding_batches(
    queries
)

Get iterator of embedding batches.

model class property

Model.

pbar_kwargs class property

Keyword arguments passed to ProgressBar.

show_progress class property

Whether to show progress bar.

template_context class property

Context used to substitute templates.

FileStore class

FileStore(
    dir_path=None,
    compression=None,
    save_kwargs=None,
    load_kwargs=None,
    use_patching=None,
    consolidate=None,
    **kwargs
)

Store class based on files.

Either commits changes to a single file (with index id being the file name), or commits the initial changes to the base file and any other change to patch file(s) (with index id being the directory name).

For defaults, see chat.obj_store_configs.file in knowledge.

Superclasses

Inherited members

compression class property

Compression.

consolidate class property

Whether to consolidate patch files.

dir_path class property

Path to the directory.

get_next_patch_path method

FileStore.get_next_patch_path()

Get path to the next patch file to be saved.

load_kwargs class property

Keyword arguments passed to load.

new_keys class property

Keys that haven't been added to the store.

reset_state method

FileStore.reset_state()

Reset state.

save_kwargs class property

Keyword arguments passed to save.

store_changes class property

Store with new or modified objects only.

store_path class property

Path to the directory with patch files or a single file.

use_patching class property

Whether to use directory with patch files or create a single file.

LMDBStore class

LMDBStore(
    dir_path=None,
    mkdir_kwargs=None,
    dumps_kwargs=None,
    loads_kwargs=None,
    **kwargs
)

Store class based on LMDB (Lightning Memory-Mapped Database).

Uses lmdbm package.

For defaults, see chat.obj_store_configs.lmdb in knowledge.

Superclasses

Inherited members

db class property

Database.

db_path class property

Path to the database.

decode method

LMDBStore.decode(
    bytes_
)

Decode an object.

dir_path class property

Path to the directory.

dumps_kwargs class property

Keyword arguments passed to dumps.

encode method

LMDBStore.encode(
    obj
)

Encode an object.

loads_kwargs class property

Keyword arguments passed to loads.

mkdir_kwargs class property

Keyword arguments passed to check_mkdir.

open_kwargs class property

Keyword arguments passed to lmdbm.lmdbm.Lmdb.open.

LiteLLMCompletions class

LiteLLMCompletions(
    context='',
    chat_history=None,
    stream=None,
    max_tokens=None,
    tokenizer=None,
    tokenizer_kwargs=None,
    system_prompt=None,
    system_as_user=None,
    context_prompt=None,
    formatter=None,
    formatter_kwargs=None,
    silence_warnings=None,
    template_context=None,
    model=None,
    **kwargs
)

Completions class for LiteLLM.

Keyword arguments are passed to the completion call.

For defaults, see chat.completions_configs.litellm in knowledge.

Superclasses

Inherited members

completion_kwargs class property

Keyword arguments passed to litellm.completion.

LiteLLMEmbeddings class

LiteLLMEmbeddings(
    model=None,
    batch_size=None,
    show_progress=None,
    pbar_kwargs=None,
    template_context=None,
    **kwargs
)

Embeddings class for LiteLLM.

For defaults, see chat.embeddings_configs.litellm in knowledge.

Superclasses

Inherited members

embedding_kwargs class property

Keyword arguments passed to litellm.embedding.

LlamaIndexCompletions class

LlamaIndexCompletions(
    context='',
    chat_history=None,
    stream=None,
    max_tokens=None,
    tokenizer=None,
    tokenizer_kwargs=None,
    system_prompt=None,
    system_as_user=None,
    context_prompt=None,
    formatter=None,
    formatter_kwargs=None,
    silence_warnings=None,
    template_context=None,
    llm=None,
    **kwargs
)

Completions class for LlamaIndex.

LLM can be provided via llm, which can be either the name of the class (case doesn't matter), the path or its suffix to the class (case matters), or a subclass or an instance of llama_index.core.llms.LLM.

Keyword arguments are passed to the resolved LLM.

For defaults, see chat.completions_configs.llama_index in knowledge.

Superclasses

Inherited members

llm class property

LLM.

LlamaIndexEmbeddings class

LlamaIndexEmbeddings(
    embedding=None,
    batch_size=None,
    show_progress=None,
    pbar_kwargs=None,
    template_context=None,
    **kwargs
)

Embeddings class for LlamaIndex.

For defaults, see chat.embeddings_configs.llama_index in knowledge.

Superclasses

Inherited members

embedding class property

Embedding.

LlamaIndexSplitter class

LlamaIndexSplitter(
    node_parser=None,
    template_context=None,
    **kwargs
)

Splitter class based on a node parser from LlamaIndex.

For defaults, see chat.text_splitter_configs.llama_index in knowledge.

Superclasses

Inherited members

node_parser class property

An instance of llama_index.core.node_parser.interface.NodeParser.

MemoryStore class

MemoryStore(
    **kwargs
)

Store class based in memory.

Commits changes to memory_store.

For defaults, see chat.obj_store_configs.memory in knowledge.

Superclasses

Inherited members

store_exists method

MemoryStore.store_exists()

Whether store exists.

MetaObjectStore class

MetaObjectStore(
    name,
    bases,
    attrs
)

Metaclass for ObjectStore.

Superclasses

ObjectStore class

ObjectStore(
    store_id=None,
    purge_on_open=None,
    template_context=None,
    **kwargs
)

Abstract class for managing an object store.

For defaults, see knowledge.chat.obj_store_config in knowledge.

Superclasses

Inherited members

Subclasses

check_opened method

ObjectStore.check_opened()

Check the store is opened.

close method

ObjectStore.close()

Close the store.

commit method

ObjectStore.commit()

Commit changes.

enter_calls class property

Number of enter calls.

mirror_store_id class property

Mirror store id.

open method

ObjectStore.open()

Open the store.

opened class property

Whether the store has been opened.

purge method

ObjectStore.purge()

Purge the store.

purge_on_open class property

Whether to purge on open.

store_id class property

Store id.

template_context class property

Context used to substitute templates.

OpenAICompletions class

OpenAICompletions(
    context='',
    chat_history=None,
    stream=None,
    max_tokens=None,
    tokenizer=None,
    tokenizer_kwargs=None,
    system_prompt=None,
    system_as_user=None,
    context_prompt=None,
    formatter=None,
    formatter_kwargs=None,
    silence_warnings=None,
    template_context=None,
    model=None,
    **kwargs
)

Completions class for OpenAI.

Keyword arguments are distributed between the client call and the completion call.

For defaults, see chat.completions_configs.openai in knowledge.

Superclasses

Inherited members

client class property

Client.

completion_kwargs class property

Keyword arguments passed to openai.resources.chat.completions_configs.Completions.create.

OpenAIEmbeddings class

OpenAIEmbeddings(
    model=None,
    batch_size=None,
    show_progress=None,
    pbar_kwargs=None,
    template_context=None,
    **kwargs
)

Embeddings class for OpenAI.

For defaults, see chat.embeddings_configs.openai in knowledge.

Superclasses

Inherited members

client class property

Client.

embeddings_kwargs class property

Keyword arguments passed to openai.resources.embeddings.Embeddings.create.

RankContextable class

RankContextable()

Abstract class that combines both Rankable and Contextable to rank a context.

Superclasses

Inherited members

Subclasses

chat class method

RankContextable.chat(
    message,
    chat_history=None,
    *,
    incl_past_queries=None,
    rank=None,
    top_k=None,
    min_top_k=None,
    max_top_k=None,
    cutoff=None,
    return_chunks=None,
    rank_kwargs=None,
    **kwargs
)

See Contextable.chat.

If rank is True, or rank is None and any of top_k, min_top_k, max_top_k, cutoff, or return_chunks is set, will rank the documents with Rankable.rank first.

Rankable class

Rankable()

Abstract class that can be ranked.

Superclasses

Inherited members

Subclasses

embed method

Rankable.embed(
    refresh=False,
    refresh_documents=None,
    refresh_embeddings=None,
    return_embeddings=False,
    return_documents=False,
    **kwargs
)

Embed documents.

rank method

Rankable.rank(
    query,
    top_k=None,
    min_top_k=None,
    max_top_k=None,
    cutoff=None,
    refresh=False,
    refresh_documents=None,
    refresh_embeddings=None,
    return_chunks=False,
    return_scores=False,
    **kwargs
)

Rank documents by their relevance to a query.

ScoredDocument class

ScoredDocument(
    *args,
    **kwargs
)

Abstract class for scored documents.

Superclasses

Inherited members

child_documents field

Scored child documents.

document field

Document.

score field

Score.

SegmentSplitter class

SegmentSplitter(
    separators=None,
    min_chunk_size=None,
    fixed_overlap=None,
    **kwargs
)

Splitter class for segments based on separators.

If a segment is too big, the next separator within the same layer is taken to split the segment into smaller segments. If a segment is too big and there are no segments previously added to the chunk, or, if the number of tokens is less than the minimal count, the next layer is taken. To split into tokens, set any separator to None. To split into characters, use an empty string.

For defaults, see chat.text_splitter_configs.segment in knowledge.

Superclasses

Inherited members

fixed_overlap class property

Whether overlap should be fixed.

min_chunk_size class property

Minimum number of tokens per chunk.

Can also be provided as a floating number relative to SegmentSplitter.chunk_size.

separators class property

Nested list of separators grouped into layers.

split_into_segments method

SegmentSplitter.split_into_segments(
    text,
    separator=None
)

Split text into segments.

StoreDocument class

StoreDocument(
    *args,
    **kwargs
)

Abstract class for documents to be stored.

Superclasses

Inherited members

Subclasses

data field

Data.

from_data class method

StoreDocument.from_data(
    data,
    id_=None,
    **kwargs
)

Create an instance of StoreDocument from data.

get_content method

StoreDocument.get_content(
    for_embed=False
)

Get content.

Returns None if there's no content.

id_from_data class method

StoreDocument.id_from_data(
    data
)

Generate a unique identifier from data.

split method

StoreDocument.split()

Split document into multiple documents.

template_context field

Context used to substitute templates.

StoreEmbedding class

StoreEmbedding(
    *args,
    **kwargs
)

Class for embeddings to be stored.

Superclasses

Inherited members

child_ids field

Child object identifiers.

embedding field

Embedding.

parent_id field

Parent object identifier.

StoreObject class

StoreObject(
    *args,
    **kwargs
)

Class for objects to be managed by a store.

Superclasses

Inherited members

Subclasses

id_ field

Object identifier.

TextDocument class

TextDocument(
    *args,
    **kwargs
)

Class for text documents.

Superclasses

Inherited members

content_template field

Content template.

Must be suitable for formatting via the format() method.

dump_kwargs field

Keyword arguments passed to dump.

excl_embed_metadata field

Whether to exclude metadata and which fields to exclude for embeddings.

If None, becomes TextDocument.excl_metadata.

excl_metadata field

Whether to exclude metadata and which fields to exclude.

If False, metadata becomes everything except text.

get_metadata method

TextDocument.get_metadata(
    for_embed=False
)

Get metadata.

Returns None if no metadata.

get_metadata_content method

TextDocument.get_metadata_content(
    for_embed=False
)

Get metadata content.

Returns None if no metadata.

get_text method

TextDocument.get_text()

Get text.

Returns None if no text.

metadata_template field

Metadata template.

Must be suitable for formatting via the format() method.

skip_missing field

Set missing text or metadata to None rather than raise an error.

split_text_kwargs field

Keyword arguments passed to split_text.

text_path field

Path to the text field.

TextSplitter class

TextSplitter(
    chunk_template=None,
    template_context=None,
    **kwargs
)

Abstract class for text splitters.

For defaults, see knowledge.chat.text_splitter_config in knowledge.

Superclasses

Inherited members

Subclasses

chunk_template class property

Chunk template.

Can use the following context: chunk_idx, chunk_start, chunk_end, chunk_text, and text.

split method

TextSplitter.split(
    text
)

Split text and yield start character and end character position of each chunk.

split_text method

TextSplitter.split_text(
    text
)

Split text and return text chunks.

template_context class property

Context used to substitute templates.

TikTokenizer class

TikTokenizer(
    encoding=None,
    model=None,
    tokens_per_message=None,
    tokens_per_name=None,
    **kwargs
)

Tokenizer class for tiktoken.

Encoding can be a model name, an encoding name, or an encoding object for tokenization.

For defaults, see chat.tokenizer_configs.tiktoken in knowledge.

Superclasses

Inherited members

encoding class property

Encoding.

tokens_per_message class property

Tokens per message.

tokens_per_name class property

Tokens per name.

TokenSplitter class

TokenSplitter(
    chunk_size=None,
    chunk_overlap=None,
    tokenizer=None,
    tokenizer_kwargs=None,
    **kwargs
)

Splitter class for tokens.

For defaults, see chat.text_splitter_configs.token in knowledge.

Superclasses

Inherited members

Subclasses

chunk_overlap class property

Number of overlapping tokens between chunks.

Can also be provided as a floating number relative to SegmentSplitter.chunk_size.

chunk_size class property

Maximum number of tokens per chunk.

split_into_tokens method

TokenSplitter.split_into_tokens(
    text
)

Split text into tokens.

tokenizer class property

An instance of Tokenizer.

Tokenizer class

Tokenizer(
    template_context=None,
    **kwargs
)

Abstract class for tokenizers.

For defaults, see knowledge.chat.tokenizer_config in knowledge.

Superclasses

Inherited members

Subclasses

count_tokens method

Tokenizer.count_tokens(
    text
)

Count tokens in a text.

count_tokens_in_messages method

Tokenizer.count_tokens_in_messages(
    messages
)

Count tokens in messages.

decode method

Tokenizer.decode(
    tokens
)

Decode a list of tokens into text.

decode_single method

Tokenizer.decode_single(
    token
)

Decode a single token into text.

encode method

Tokenizer.encode(
    text
)

Encode text into a list of tokens.

encode_single method

Tokenizer.encode_single(
    text
)

Encode text into a single token.

template_context class property

Context used to substitute templates.