renaudb1999/le-harnais-ft-counsel
renaudb1999/le-harnais-ft-counsel is a 1 billion parameter instruction-tuned causal language model built upon Meta Llama 3.2. This model is specifically fine-tuned for providing wisdom-grounded counseling, excelling in a verify-repair advice game. It is designed to generate and retrieve advice for couple-related scenarios, leveraging a context length of 32768 tokens.
Loading preview...
Model Overview
renaudb1999/le-harnais-ft-counsel is a 1 billion parameter model, based on the meta-llama/Llama-3.2-1B-Instruct architecture. It is specifically designed as a "wisdom-grounded counselor" for an advice-giving game, focusing on a "verify→repair" methodology. The model operates within a 32768 token context window.
Key Capabilities
- Counseling and Advice Generation: Specialized in providing advice, particularly for couple-related scenarios, by generating and retrieving relevant guidance.
- Wisdom-Grounded Responses: Trained on a dataset of 270 examples (
datasets/counsel_train.jsonl) derived from wisdom and commentary from public domain sources, ensuring its advice is grounded in established principles. - Advice Game Participation: Optimized for a "generate→retrieve→judge→repair" advice game, indicating its ability to iteratively refine suggestions.
Formats and Usage
This model is provided in multiple formats for diverse deployment needs:
.safetensors: For bf16 inference usingtransformersorle-harnais'slh-serve/candle..Q4_K_M.gguf: A portable 4-bit quantization suitable forollamaorllama.cpp, including Mac environments..Q8_0.gguf: A higher-fidelity 8-bit quantization, typically used for "hero models."
Important Note
The developers emphasize that "Orchestration amplifies a capable generator; it does not create competence," highlighting the model's inherent capabilities as a foundation for more complex systems.