Skip to content

Release Artifacts

llamatelemetry v0.1.1 ships as both source archives and CUDA binary bundles. This page provides a complete breakdown of release contents, installation methods, and artifact usage.

Release Overview

Property Value
Version v0.1.1
Release date 2026-02-02
License MIT
Python >= 3.11
CUDA 12.x
Target GPU Tesla T4 (SM 7.5)
Repository github.com/llamatelemetry/llamatelemetry

Installation

pip install git+https://github.com/llamatelemetry/llamatelemetry.git@v0.1.1

From Source

git clone https://github.com/llamatelemetry/llamatelemetry.git
cd llamatelemetry
git checkout v0.1.1
pip install -e .

With Optional Dependencies

# Telemetry support
pip install git+https://github.com/llamatelemetry/llamatelemetry.git@v0.1.1 \
    opentelemetry-api opentelemetry-sdk opentelemetry-exporter-otlp

# Graphistry support
pip install git+https://github.com/llamatelemetry/llamatelemetry.git@v0.1.1 \
    pygraphistry pandas

# Full installation (all optional dependencies)
pip install -e ".[dev]"

On Kaggle

In a Kaggle notebook cell:

!pip install -q git+https://github.com/llamatelemetry/llamatelemetry.git@v0.1.1

Source Archives

Files:

  • llamatelemetry-v0.1.1-source.tar.gz
  • llamatelemetry-v0.1.1-source.zip

Contents (top level):

llamatelemetry-v0.1.1/
  llamatelemetry/        Python package source (~40 files, 13K+ lines)
  csrc/                  C++/CUDA extension sources (7 files, ~650 lines)
  docs/                  MkDocs documentation content
  notebooks/             18 curated Kaggle-ready notebooks
  examples/              Runnable example scripts
  tests/                 Unit and end-to-end tests (246 tests)
  scripts/               Release and HuggingFace helper scripts
  pyproject.toml         Build configuration
  README.md              Project README
  LICENSE                MIT License
  CHANGELOG.md           Version history

These archives match the repository source tree exactly.

CUDA Binary Bundle (Kaggle T4 x2)

File:

  • llamatelemetry-v0.1.1-cuda12-kaggle-t4x2.tar.gz

This bundle contains pre-compiled CUDA 12 binaries targeting Tesla T4 GPUs (SM 7.5, compute capability 7.5). Total size is approximately 961 MB.

Binaries Included

Binary Purpose
llama-server OpenAI-compatible HTTP server for GGUF inference
llama-cli Command-line inference tool
llama-bench Performance benchmarking tool
llama-embedding Embedding generation tool
llama-tokenize Tokenization utility
llama-gguf GGUF file inspection tool
llama-gguf-hash GGUF file hash verification
llama-gguf-split GGUF file splitting utility
llama-quantize Model quantization tool
llama-imatrix Importance matrix generation
llama-perplexity Perplexity evaluation tool
llama-export-lora LoRA adapter export tool
llama-cvector-generator Control vector generation tool

Libraries Included

Library Purpose
lib/libnccl.so NVIDIA Collective Communications Library
lib/libnccl.so.2 NCCL versioned symlink

Scripts Included

Script Purpose
start-server.sh Quick-start script for llama-server
quantize.sh Model quantization helper script

Binary Bootstrap Process

When llamatelemetry is imported for the first time and the CUDA binaries are not found locally, the bootstrap layer automatically downloads and extracts the binary bundle.

import llamatelemetry
  --> Check for llama-server in llamatelemetry/binaries/cuda12/
  --> If missing: download llamatelemetry-v0.1.1-cuda12-kaggle-t4x2.tar.gz
  --> Extract to llamatelemetry/binaries/cuda12/
  --> Set LLAMA_SERVER_PATH environment variable
  --> Set LD_LIBRARY_PATH to include lib/ directory

The download URL is configured in ServerManager._BINARY_BUNDLES and points to the GitHub Releases page:

https://github.com/llamatelemetry/llamatelemetry/releases/download/v0.1.1/
    llamatelemetry-v0.1.1-cuda12-kaggle-t4x2.tar.gz

Cache Locations

The bootstrap layer checks these locations in order:

  1. llamatelemetry/binaries/cuda12/llama-server (package directory)
  2. /usr/local/bin/llama-server (system-wide)
  3. /usr/bin/llama-server (system-wide)
  4. ~/.cache/llamatelemetry/llama-server (user cache)

Kaggle Dataset Integration

For Kaggle notebooks, GGUF model files are typically uploaded as Kaggle Datasets and attached to the notebook. The SDK's model registry and KaggleEnvironment handle path resolution automatically.

from llamatelemetry.kaggle import KaggleEnvironment

env = KaggleEnvironment.setup()
engine = env.create_engine("gemma-3-1b-Q4_K_M")

The MODEL_REGISTRY in _internal/registry.py contains 30+ curated models with HuggingFace repo references, filenames, sizes, and VRAM requirements.

Checksums

Each archive includes a .sha256 file in releases/v0.1.1/ for integrity verification:

sha256sum -c llamatelemetry-v0.1.1-cuda12-kaggle-t4x2.tar.gz.sha256
sha256sum -c llamatelemetry-v0.1.1-source.tar.gz.sha256

Release Directory Structure

releases/
  v0.1.1/
    llamatelemetry-v0.1.1-source.tar.gz
    llamatelemetry-v0.1.1-source.zip
    llamatelemetry-v0.1.1-cuda12-kaggle-t4x2.tar.gz
    *.sha256 checksum files
  v1.2.0/
    Binary archive + CUDA tar.gz (future release)

Where to Go Next