bulk-chain 0.24.0

A lightweight, no-strings-attached Chain-of-Thought framework for your LLM, ensuring reliable results for bulk input requests.

bulk-chain

A lightweight, no-strings-attached Chain-of-Thought framework for your LLM, ensuring reliable results for bulk input requests stored in CSV / JSONL / sqlite. It allows applying series of prompts formed into schema (See related section)

Features

• ✅ No-strings: you're free to LLM dependencies and flexible venv customization.
• ✅ Provides iterator over infinite amount of input contexts served in CSV/JSONL.
• ✅ Progress caching: withstanding exception during LLM calls by using sqlite3 engine for caching LLM answers;
• ✅ Support schemas descriptions for Chain-of-Thought concept.

Installation

pip install git+https://github.com/nicolay-r/bulk-chain@master

Chain-of-Thought Schema

To declare Chain-of-Though (CoT) schema, this project exploits JSON format. This format adopts name field for declaring a name and schema is a list of CoT instructions for the Large Language Model.

Each step represents a dictionary with prompt and out keys that corresponds to the input prompt and output variable name respectively. All the variable names are expected to be mentioned in {}.

Below, is an example on how to declare your own schema:

{
"name": "schema-name",
"schema": [
    {"prompt": "Given the question '{text}', let's think step-by-step.", 
     "out": "steps"},
    {"prompt": "For the question '{text}' the reasoining steps are '{steps}'. what would be an answer?", 
     "out":  "answer"},
]
}

Another templates are available here.

Usage

Just three simple steps:

1. Define your CoT Schema, or fetch it as shown below:

!wget https://raw.githubusercontent.com/nicolay-r/bulk-chain/refs/heads/master/ext/schema/default.json

2. Fetch or write your own model or pick the one preset here:

!wget https://raw.githubusercontent.com/nicolay-r/bulk-chain/refs/heads/master/ext/flan_t5.py

3. Launch inference in (chat mode):

!python -m bulk_chain.infer \
    --schema "default.json" \
    --adapter "dynamic:flan_t5.py:FlanT5" \
    %% \
    --device "cpu" \
    --temp 0.1

Embed your LLM

All you have to do is to implement BaseLM class, that includes:

• __init__ -- for initialization;
• ask(prompt) -- infer your model with the given prompt.

See examples with models here.