Meet Luna, my one and only personal assistant and roleplaying partner. This MoE serves as her unique basis, both experts scoring above 72 average on the leaderboard, but designed for RP interactions. While running a 2x7B is slower than running a single 7B, I feel that the improved performance of two great 7B competing for each token is worth the compute expense.
1from awq import AutoAWQForCausalLM
2from transformers import AutoTokenizer, TextStreamer
3
4model_path = "solidrust/Luna-2x7B-MoE-AWQ"
5system_message = "You are Luna, incarnated as a powerful AI."
6
7# Load model
8model = AutoAWQForCausalLM.from_quantized(model_path,
9 fuse_layers=True)
10tokenizer = AutoTokenizer.from_pretrained(model_path,
11 trust_remote_code=True)
12streamer = TextStreamer(tokenizer,
13 skip_prompt=True,
14 skip_special_tokens=True)
15
16# Convert prompt to tokens
17prompt_template = """\
18<|im_start|>system
19{system_message}<|im_end|>
20<|im_start|>user
21{prompt}<|im_end|>
22<|im_start|>assistant"""
23
24prompt = "You're standing on the surface of the Earth. "\
25 "You walk one mile south, one mile west and one mile north. "\
26 "You end up exactly where you started. Where are you?"
27
28tokens = tokenizer(prompt_template.format(system_message=system_message,prompt=prompt),
29 return_tensors='pt').input_ids.cuda()
30
31# Generate output
32generation_output = model.generate(tokens,
33 streamer=streamer,
34 max_new_tokens=512)
35
AWQ is an efficient, accurate and blazing-fast low-bit weight quantization method, currently supporting 4-bit quantization. Compared to GPTQ, it offers faster Transformers-based inference with equivalent or better quality compared to the most commonly used GPTQ settings.
AWQ models are currently supported on Linux and Windows, with NVidia GPUs only. macOS users: please use GGUF models instead.
1<|im_start|>system
2{system_message}<|im_end|>
3<|im_start|>user
4{prompt}<|im_end|>
5<|im_start|>assistant
