renaudb1999/le-harnais-ft-smoke-jepa
The renaudb1999/le-harnais-ft-smoke-jepa is a 1 billion parameter model based on Meta Llama 3.2-1B-Instruct, designed as a pipeline smoke test for the le-harnais training framework. This checkpoint is specifically for reproducing and continuing training studies, not for inference. It validates the training path with a 128-sample smoke split and is not intended for general use.
Loading preview...
Overview
renaudb1999/le-harnais-ft-smoke-jepa is a 1 billion parameter model derived from meta-llama/Llama-3.2-1B-Instruct. It serves as a pipeline smoke test within the le-harnais project, specifically designed to validate the training process rather than for practical inference. This model is built with Llama architecture, and its use is subject to the Llama Community License.
Key Characteristics
- Purpose: Exclusively a smoke checkpoint for training pipeline validation.
- Base Model:
meta-llama/Llama-3.2-1B-Instruct. - Training Data: Utilizes a small 128-sample smoke split for quick validation.
- Inference: Not useful for inference; users are directed to
le-harnais-ft-agentworld-{1b,3b,8b}orle-harnais-ft-counselfor actual use cases. - Formats: Available in
.safetensorsfor bf16 inference (viatransformersorlh-serve/candle) and.Q4_K_M.gguffor portable 4-bit quantization (viaollama/llama.cpp).
Intended Use
This model is primarily for developers and researchers looking to:
- Reproduce Training: Validate and continue the training study of the le-harnais framework.
- Pipeline Testing: Confirm the integrity and functionality of the training pipeline with a minimal dataset.
It is crucial to understand that this model does not represent a measured result or a production-ready artifact. Its sole function is to ensure the training infrastructure is sound.