PyPI page
Home page
Author:
Tej Desai
Summary:
CARL (Coherence-Aware Reinforcement Learning) — information-theoretic reward signals for LLM training via token-level probability distributions
Latest version:
0.21.1
Required dependencies:
anyio
|
carl-core
|
httpx
|
numpy
|
pydantic
|
pydantic-settings
|
pyyaml
|
rich
|
typer
Optional dependencies:
anthropic
|
arxiv
|
bitsandbytes
|
build
|
coinbase-agentkit
|
cryptography
|
cuda-tile
|
cyclonedx-bom
|
datasets
|
flash-attn-4
|
gradio-client
|
huggingface-hub
|
hypothesis
|
keyring
|
mcp
|
peft
|
prometheus-client
|
pynacl
|
pyperclip
|
pyright
|
pytest
|
pytest-asyncio
|
questionary
|
respx
|
ruff
|
runpod
|
sglang
|
stripe-agent-toolkit
|
textual
|
textual-plotext
|
tinker
|
torch
|
trackio
|
transformers
|
trl
|
uv
|
x402
Downloads last day:
20
Downloads last week:
78
Downloads last month:
27
7-day smoothing