Spaces:
Configuration error
Configuration error
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,40 +0,0 @@
|
|
| 1 |
-
---
|
| 2 |
-
emoji: ⚡
|
| 3 |
-
license: apache-2.0
|
| 4 |
-
title: meeTARA Spark
|
| 5 |
-
sdk: static
|
| 6 |
-
colorFrom: indigo
|
| 7 |
-
colorTo: blue
|
| 8 |
-
---
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
# meeTARA Spark
|
| 12 |
-
|
| 13 |
-
**GGUF models for Apache Spark.** Run quantized LLMs as a step in your data pipeline — inference where the data lives.
|
| 14 |
-
|
| 15 |
-
We publish Spark-optimized GGUF models (Qwen, Phi, SmolLM, and more) with **embedded metadata** so [GGUF-Spark](https://github.com/rbasina/meetara-spark) can run them on your cluster. Use them via **SparkSQL UDFs**: predict, summarize, classify, extract, data-quality checks, anomaly detection. No external APIs; data stays on your cluster.
|
| 16 |
-
|
| 17 |
-
## Why GGUF-Spark?
|
| 18 |
-
|
| 19 |
-
Analytics and LLM features in one place. Your tables and files stay in Spark; you add sentiment, summaries, or entities by registering a UDF and pointing it at one of our models. The model runs on the same executors that hold the data — one pipeline, at scale.
|
| 20 |
-
|
| 21 |
-
```mermaid
|
| 22 |
-
flowchart LR
|
| 23 |
-
T[Tables / Files] --> DF[Spark DataFrame]
|
| 24 |
-
DF --> P[Partitions]
|
| 25 |
-
P --> UDF[GGUF UDF]
|
| 26 |
-
UDF --> M[(Our models)]
|
| 27 |
-
M --> E[Enriched table]
|
| 28 |
-
```
|
| 29 |
-
|
| 30 |
-
## Models
|
| 31 |
-
|
| 32 |
-
All models in this org are ready for GGUF-Spark. Each repo includes Q4_K_M and Q8_0 quantizations and a model card with usage examples and requirements (Java, Python, `pyspark`, `llama-cpp-python`, etc.).
|
| 33 |
-
|
| 34 |
-
**[Browse models →](https://huggingface.co/meetara-spark)**
|
| 35 |
-
|
| 36 |
-
## Research
|
| 37 |
-
|
| 38 |
-
These models support the work described in **"Distributed Inference of Quantized Large Language Models on Apache Spark: A Memory-Aware Architectural Analysis"** (Basina, 2026). The research introduces memory-aware quantization selection, partition-resident model (PRM) deployment, thread orchestration to avoid CPU oversubscription, and SparkSQL UDF integration with GGUF/llama.cpp. If you use these models or the GGUF-Spark system in your work, we encourage citing the paper; the manuscript and implementation details are in the [meetara-spark repo](https://github.com/rbasina/meetara-spark) (see `docs/GGUF-SPARK-RESEARCH-PAPER.md` and `docs/INDEX.md`).
|
| 39 |
-
- **GitHub:** [rbasina/meetara-spark](https://github.com/rbasina/meetara-spark)
|
| 40 |
-
- **Part of:** [meeTARA](https://github.com/rbasina/meetara) — empathetic AI assistant and tooling
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|