All tags
Topic: "multi-gpu"
12/27/2023: NYT vs OpenAI
phi2 openhermes-2.5-mistral-7b llama-2-7b llama-2-13b microsoft-research mistral-ai apple amd model-performance fine-tuning llm-api gpu-optimization hardware-configuration multi-gpu inference-speed plugin-release conversation-history
The LM Studio Discord community extensively discussed model performance comparisons, notably between Phi2 by Microsoft Research and OpenHermes 2.5 Mistral 7b, with focus on U.S. history knowledge and fine-tuning for improved accuracy. Technical challenges around LLM API usage, conversation history maintenance, and GPU optimization for inference speed were addressed. Hardware discussions covered DDR4 vs DDR5, multi-GPU setups, and potential of Apple M1/M3 and AMD AI CPUs for AI workloads. The community also announced the ChromaDB Plugin v3.0.2 release enabling image search in vector databases. Users shared practical tips on running multiple LM Studio instances and optimizing resource usage.
12/9/2023: The Mixtral Rush
mixtral hermes-2.5 hermes-2 mistral-yarn ultrachat discoresearch fireworks-ai hugging-face mistral-ai benchmarking gpu-requirements multi-gpu quantization gptq chain-of-thought min-p-sampling top-p-sampling model-sampling model-merging model-performance small-models reasoning-consistency temperature-sampling bjoernp the_bloke rtyax kalomaze solbus calytrix
Mixtral's weights were released without code, prompting the Disco Research community and Fireworks AI to implement it rapidly. Despite efforts, no significant benchmark improvements were reported, limiting its usefulness for local LLM usage but marking progress for the small models community. Discussions in the DiscoResearch Discord covered Mixtral's performance compared to models like Hermes 2.5 and Hermes 2, with evaluations on benchmarks such as winogrande, truthfulqa_mc2, and arc_challenge. Technical topics included GPU requirements, multi-GPU setups, and quantization via GPTQ. Benchmarking strategies like grammar-based evaluation, chain of thought (CoT), and min_p sampling were explored, alongside model sampling techniques like Min P and Top P to enhance response stability and creativity. Users also discussed GPTs' learning limitations and the adaptability of models under varying conditions, emphasizing min_p sampling's role in enabling higher temperature settings for creativity.