question about configuration

[menglin0320]

In the examples you guys didn't mention how to specify parameters like batch size, max input length etc.
My first question is how to change the max input length, I tried the llama2 example for a RAG usage case. llama2 should be able to handle 4096 input tokens but it's limited to 1024 for some reason.
Similarly though I don't feel batching is a good idea on cpu, I still want to try batched inference with this package. is there a document for how to configure those things?

[menglin0320]

after trying mistral out, yeah you guys limit the ctx length to 1024 for every model.

[a32543254]

could you tell us which example you are using?

[menglin0320]

from transformers import AutoTokenizer
from intel_extension_for_transformers.transformers import AutoModelForCausalLM

# Specify the GGUF repo on the Hugginface
model_name = "TheBloke/Llama-2-7B-Chat-GGUF"
# Download the the specific gguf model file from the above repo
gguf_file = "llama-2-7b-chat.Q4_0.gguf"
# make sure you are granted to access this model on the Huggingface.
tokenizer_name = "meta-llama/Llama-2-7b-chat-hf"
prompt = "Once upon a time, there existed a little girl,"
tokenizer = AutoTokenizer.from_pretrained(tokenizer_name, trust_remote_code=True)
inputs = tokenizer(prompt, return_tensors="pt").input_ids

model = AutoModelForCausalLM.from_pretrained(model_name, gguf_file = gguf_file)
outputs = model.generate(inputs)

This one, and I believe that the input sequence length is limited to 1024 by default.
It's hard to know the arguments for "from_pretrained" and "model.generate" with current code.