accelerate

e3d32424 - Check if the buffers fit GPU memory after device map auto inferred (#2412)

Check if the buffers fit GPU memory after device map auto inferred (#2412) * Check if the buffers fit GPU memory after device map auto inferred * For some models, like TheBloke/WizardCoder-33B-V1.1-GPTQ, contain a huge buffer, which may cause OOM on GPU memory if not using offload_buffers. This commit adds a check for such case. * Minor refactors. * Add missing assertions