4 ms·
Have you noticed a gap between 8bit and 4bit quant? I've always ran 4bit quant cause less memory required
by nateb2022 5mo ago
Have you noticed a gap between 8bit and 4bit quant? I've always ran 4bit quant cause less memory required
- verdverm 5mo agoI run the biggest quant because it is more capable, spark has enough memory for two qwen at 8bit and full context length (roughly 48G each) I find gemini/gemma to have become worse at coding, they are better for non-coding tasks, but maybe not even that, the hallucinations and instruction following have both degraded ime