Update to remote service

app.py

@@ -1,9 +1,8 @@
 from typing import Iterator
 
 import gradio as gr
-import torch
 
-from model import get_input_token_length, run
+from model import run
 
 DEFAULT_SYSTEM_PROMPT = """\
 You are a helpful, respectful and honest assistant with a deep knowledge of code and software design. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.\n\nIf a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.\
@@ -13,9 +12,9 @@ DEFAULT_MAX_NEW_TOKENS = 1024
 MAX_INPUT_TOKEN_LENGTH = 4000
 
 DESCRIPTION = """
-# Code Llama 13B Chat
+# Code Llama 34B Chat
 
-This Space demonstrates model [CodeLlama-13b-Instruct](https://huggingface.co/codellama/CodeLlama-13b-Instruct-hf) by Meta, a Code Llama model with 13B parameters fine-tuned for chat instructions and specialized on code tasks. Feel free to play with it, or duplicate to run generations without a queue! If you want to run your own service, you can also [deploy the model on Inference Endpoints](https://huggingface.co/inference-endpoints).
+This Space demonstrates model [CodeLlama-34b-Instruct](https://huggingface.co/codellama/CodeLlama-34b-Instruct-hf) by Meta, a Code Llama model with 34B parameters fine-tuned for chat instructions and specialized on code tasks. Feel free to play with it, or duplicate to run generations without a queue! If you want to run your own service, you can also [deploy the model on Inference Endpoints](https://huggingface.co/inference-endpoints).
 
 🔎 For more details about the Code Llama family of models and how to use them with `transformers`, take a look [at our blog post](https://huggingface.co/blog/codellama) or [the paper](https://huggingface.co/papers/2308.12950).
 
@@ -28,12 +27,9 @@ LICENSE = """
 
 ---
 As a derivate work of Code Llama by Meta,
-this demo is governed by the original [license](https://huggingface.co/spaces/huggingface-projects/codellama-2-13b-chat/blob/main/LICENSE.txt) and [acceptable use policy](https://huggingface.co/spaces/huggingface-projects/codellama-2-13b-chat/blob/main/USE_POLICY.md).
+this demo is governed by the original [license](https://huggingface.co/spaces/huggingface-projects/codellama-2-34b-chat/blob/main/LICENSE.txt) and [acceptable use policy](https://huggingface.co/spaces/huggingface-projects/codellama-2-34b-chat/blob/main/USE_POLICY.md).
 """
 
-if not torch.cuda.is_available():
-    DESCRIPTION += '\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>'
-
 
 def clear_and_save_textbox(message: str) -> tuple[str, str]:
     return '', message
@@ -85,7 +81,7 @@ def process_example(message: str) -> tuple[str, list[tuple[str, str]]]:
 
 
 def check_input_token_length(message: str, chat_history: list[tuple[str, str]], system_prompt: str) -> None:
-    input_token_length = get_input_token_length(message, chat_history, system_prompt)
+    input_token_length = len(message) + len(chat_history)
     if input_token_length > MAX_INPUT_TOKEN_LENGTH:
         raise gr.Error(f'The accumulated input is too long ({input_token_length} > {MAX_INPUT_TOKEN_LENGTH}). Clear your chat history and try again.')
 
@@ -274,4 +270,4 @@ with gr.Blocks(css='style.css') as demo:
         api_name=False,
     )
 
-demo.queue(max_size=20).launch()
+demo.queue(max_size=32).launch(debug=True)

model.py

@@ -1,25 +1,19 @@
-from threading import Thread
+import os
 from typing import Iterator
 
-import torch
-from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+from text_generation import Client
 
-model_id = 'codellama/CodeLlama-13b-Instruct-hf'
+model_id = 'codellama/CodeLlama-34b-Instruct-hf'
 
-if torch.cuda.is_available():
-    config = AutoConfig.from_pretrained(model_id)
-    config.pretraining_tp = 1
-    model = AutoModelForCausalLM.from_pretrained(
-        model_id,
-        config=config,
-        torch_dtype=torch.float16,
-        load_in_4bit=True,
-        device_map='auto',
-        use_safetensors=False,
-    )
-else:
-    model = None
-tokenizer = AutoTokenizer.from_pretrained(model_id)
+API_URL = "https://api-inference.huggingface.co/models/" + model_id
+HF_TOKEN = os.environ.get("HF_TOKEN", "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")
+
+client = Client(
+    API_URL,
+    headers={"Authorization": f"Bearer {HF_TOKEN}"},
+)
+EOS_STRING = "</s>"
+EOT_STRING = "<EOT>"
 
 
 def get_prompt(message: str, chat_history: list[tuple[str, str]],
@@ -36,12 +30,6 @@ def get_prompt(message: str, chat_history: list[tuple[str, str]],
     return ''.join(texts)
 
 
-def get_input_token_length(message: str, chat_history: list[tuple[str, str]], system_prompt: str) -> int:
-    prompt = get_prompt(message, chat_history, system_prompt)
-    input_ids = tokenizer([prompt], return_tensors='np', add_special_tokens=False)['input_ids']
-    return input_ids.shape[-1]
-
-
 def run(message: str,
         chat_history: list[tuple[str, str]],
         system_prompt: str,
@@ -50,26 +38,20 @@ def run(message: str,
         top_p: float = 0.9,
         top_k: int = 50) -> Iterator[str]:
     prompt = get_prompt(message, chat_history, system_prompt)
-    inputs = tokenizer([prompt], return_tensors='pt', add_special_tokens=False).to('cuda')
 
-    streamer = TextIteratorStreamer(tokenizer,
-                                    timeout=10.,
-                                    skip_prompt=True,
-                                    skip_special_tokens=True)
     generate_kwargs = dict(
-        inputs,
-        streamer=streamer,
         max_new_tokens=max_new_tokens,
         do_sample=True,
         top_p=top_p,
         top_k=top_k,
         temperature=temperature,
-        num_beams=1,
     )
-    t = Thread(target=model.generate, kwargs=generate_kwargs)
-    t.start()
-
-    outputs = []
-    for text in streamer:
-        outputs.append(text)
-        yield ''.join(outputs)
+    stream = client.generate_stream(prompt, **generate_kwargs)
+    output = ""
+    for response in stream:
+        if any([end_token in response.token.text for end_token in [EOS_STRING, EOT_STRING]]):
+            return output
+        else:
+            output += response.token.text
+        yield output
+    return output

requirements.txt

@@ -5,4 +5,5 @@ protobuf
 scipy
 sentencepiece
 torch
+text_generation
 git+https://github.com/huggingface/transformers@main