Could you provide a sample script to start the model with openai access

[garyyang85]

No description provided.

[mayank31398]

HI @garyyang85 , having trouble understanding the request.
Can you clarify?
There is an example in the README.md

you will need to install transformers from source for this though.

cross-pasting here as well:

from transformers import AutoModelForCausalLM, AutoTokenizer

device = "cuda" # or "cpu"
model_path = "ibm-granite/granite-3b-code-base" # pick anyone from above list

tokenizer = AutoTokenizer.from_pretrained(model_path)

# drop device_map if running on CPU
model = AutoModelForCausalLM.from_pretrained(model_path, device_map=device)
model.eval()

# change input text as desired
input_text = "def generate():"
# tokenize the text
input_tokens = tokenizer(input_text, return_tensors="pt")

# transfer tokenized inputs to the device
for i in input_tokens:
    input_tokens[i] = input_tokens[i].to(device)

# generate output tokens
output = model.generate(**input_tokens)
# decode output tokens into text
output = tokenizer.batch_decode(output)

# loop over the batch to print, in this example the batch size is 1
for i in output:
    print(i)

[garyyang85]

Hi @mayank31398 Thanks for your response.
Popular models may included in fastchat. I mean something like the api server in fastchat. It will load model only once and then accept standard openai request, return the stream answer. Like this: https://github.com/baichuan-inc/Baichuan2/blob/main/OpenAI_api.py
For test purpose, maybe Flask is enough. Of course I can follow the sample code to build one myself. :)

[mayank31398]

@garyyang85 I see
currently, there is VLLM integration underway: vllm-project/vllm#4636

[mayank31398]

@garyyang85 vllm-project/vllm#4636 is merged now.
closing this.