LLM · Text Generation
Llama 3.2 1B Synthia I Redmond (Instruction Fine-Tune)
Fine-tuned Llama 3.2 1B for instruction following, trained on the Synthia-v1.5-I dataset with RedmondAI GPU support.
About this model
Llama 3.2 1B Synthia I Redmond is a text-generation model that fine-tunes Meta's Llama 3.2 1B (via the NousResearch/Llama-3.2-1B checkpoint) on the Synthia-v1.5-I instruction dataset. The goal of the fine-tune is to improve instruction-following behaviour for conversational AI and research use, while keeping the tiny 1B parameter footprint that runs on modest hardware. Training was supported by RedmondAI’s GPUs, per the model card.
The Synthia-v1.5-I training set contains 20.7k examples, and the model was trained for 3 epochs with a learning rate of 2e-05, a total batch size of 8 (1 per device with 8 gradient accumulation steps), a cosine LR schedule with 100 warmup steps, and the Paged AdamW 8bit optimizer. The fine-tune was built with the Axolotl framework 0.5.0 on Transformers 4.46.1 and PyTorch 2.3.1. The base model is multilingual, covering en, de, fr, it, pt, hi, es, and th.
Because this is an instruction-tuned LLM rather than an image or LoRA model, there are no trigger words or instance prompts: use it as a normal generative text model, prompting it directly through Hugging Face Transformers' text-generation pipeline or a chat template. No specific inference settings are published on the card, so start with the standard generation defaults for the base model and adjust temperature/top-p for your use case.
The model is released under the Llama 3.2 Community License Agreement, and all usage must comply with its terms.
Learn more
Read the in-depth guide about Llama 3.2 1B Synthia I Redmond (Instruction Fine-Tune)
Project signals
- 35 Hugging Face downloads
- 1 Hugging Face likes
Topics
transformers · pytorch · safetensors · llama · text-generation · facebook · meta · llama-3 · base_model:nousresearch/llama-3.2-1b · base_model:finetune:nousresearch/llama-3.2-1b