Text Generation
Transformers
llama
Inference Endpoints
GodRain's picture
Update README.md
5e6a82d
|
raw
history blame
1.64 kB
metadata
license: bigcode-openrail-m
datasets:
  - WizardLM/WizardLM_evol_instruct_70k
language:
  - en

Here is an example to show how to use model quantized by auto_gptq

_3BITS_MODEL_PATH_V1_ = 'GodRain/WizardCoder-15B-V1.1-3bit'

# pip install auto_gptq
from auto_gptq import AutoGPTQForCausalLM
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained(_3BITS_MODEL_PATH_V1_)
model = AutoGPTQForCausalLM.from_quantized(_3BITS_MODEL_PATH_V1_)

out = evaluate("Hello, tell me a story about sun", model=model, tokenizer=tokenizer)
print(out[0].strip())
def evaluate(
        batch_data,
        tokenizer,
        model,
        temperature=1,
        top_p=0.9,
        top_k=40,
        num_beams=1,
        max_new_tokens=2048,
        **kwargs,
):
    prompts = generate_prompt(batch_data)
    inputs = tokenizer(prompts, return_tensors="pt", max_length=256, truncation=True)
    input_ids = inputs["input_ids"].to(device)
    generation_config = GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        num_beams=num_beams,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.pad_token_id,
        **kwargs,
    )
    with torch.no_grad():
        generation_output = model.generate(
            input_ids=input_ids,
            generation_config=generation_config,
            return_dict_in_generate=True,
            output_scores=True,
            max_new_tokens=max_new_tokens,
        )
    s = generation_output.sequences
    output = tokenizer.batch_decode(s, skip_special_tokens=True)
    return output