MaziyarPanahi

/

Starling-LM-7B-beta-GPTQ

like 4

Might be the model, but I would double check the chat template, special tokens, etc. Since you are using HF Pipelines, you have to make sure there is a template, it's a correct one, it ads the spacial tokens, etc.

Thanks, I'm using exactly the same settings as in your example.

from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
import torch

model_id = "MaziyarPanahi/Starling-LM-7B-beta-GPTQ"

quantize_config = BaseQuantizeConfig(
        bits=4,
        group_size=128,
        desc_act=False
    )

model = AutoGPTQForCausalLM.from_quantized(
        model_id,
        use_safetensors=True,
        device="cuda:0",
        quantize_config=quantize_config)

tokenizer = AutoTokenizer.from_pretrained(model_id)

pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=512,
    temperature=0.7,
    top_p=0.95,
    repetition_penalty=1.1
)

outputs = pipe("What is love?")
print(outputs[0]["generated_text"])

My README is a very basic template, I suggest looking here for examples that are compatible with the original model: https://huggingface.co/Nexusflow/Starling-LM-7B-beta