INIT

app.py

@@ -1,146 +1,48 @@
+from dotenv import load_dotenv
+from generators import *
 import gradio as gr
-import numpy as np
-import random
-from diffusers import DiffusionPipeline
-import torch
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-
-if torch.cuda.is_available():
-    torch.cuda.max_memory_allocated(device=device)
-    pipe = DiffusionPipeline.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16", use_safetensors=True)
-    pipe.enable_xformers_memory_efficient_attention()
-    pipe = pipe.to(device)
-else: 
-    pipe = DiffusionPipeline.from_pretrained("stabilityai/sdxl-turbo", use_safetensors=True)
-    pipe = pipe.to(device)
-
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
-def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
-
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-        
-    generator = torch.Generator().manual_seed(seed)
-    
-    image = pipe(
-        prompt = prompt, 
-        negative_prompt = negative_prompt,
-        guidance_scale = guidance_scale, 
-        num_inference_steps = num_inference_steps, 
-        width = width, 
-        height = height,
-        generator = generator
-    ).images[0] 
-    
-    return image
-
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css="""
-#col-container {
-    margin: 0 auto;
-    max-width: 520px;
-}
-"""
-
-if torch.cuda.is_available():
-    power_device = "GPU"
-else:
-    power_device = "CPU"
-
-with gr.Blocks(css=css) as demo:
-    
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(f"""
-        # Text-to-Image Gradio Template
-        Currently running on {power_device}.
-        """)
-        
-        with gr.Row():
-            
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-            
-            run_button = gr.Button("Run", scale=0)
-        
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-            
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-            
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-            
-            with gr.Row():
-                
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=512,
-                )
-                
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=512,
-                )
-            
-            with gr.Row():
-                
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,
-                )
-                
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=12,
-                    step=1,
-                    value=2,
-                )
-        
-        gr.Examples(
-            examples = examples,
-            inputs = [prompt]
-        )
-
-    run_button.click(
-        fn = infer,
-        inputs = [prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
-        outputs = [result]
+from utils import async_zip_stream
+
+load_dotenv()
+
+
+async def handle(system_input: str, user_input: str):
+    print(system_input, user_input)
+    buffers = ["", "", "", ""]
+    async for outputs in async_zip_stream(
+            generate_gpt2(system_input, user_input),
+            generate_mistral_7bvo1(system_input, user_input),
+            generate_llama2(system_input, user_input),
+            generate_llama3(system_input, user_input),
+    ):
+        # gpt_output, mistral_output, llama_output, llama2_output, llama3_output, llama4_output = outputs
+        for i, b in enumerate(buffers):
+            buffers[i] += str(outputs[i])
+
+        yield list(buffers) + ["", ""]
+    yield list(buffers) + [await generate_openllama(system_input, user_input),
+                           await generate_bloom(system_input, user_input)]
+
+
+with gr.Blocks() as demo:
+    system_input = gr.Textbox(label='System Input', value='You are AI assistant', lines=2)
+    with gr.Row():
+        gpt = gr.Textbox(label='gpt-2', lines=4, interactive=False)
+        mistral = gr.Textbox(label='mistral', lines=4, interactive=False)
+        llama = gr.Textbox(label='openllama', lines=4, interactive=False)
+    with gr.Row():
+        llama2 = gr.Textbox(label='llama-2', lines=4, interactive=False)
+        llama3 = gr.Textbox(label='llama-3', lines=4, interactive=False)
+        bloom = gr.Textbox(label='bloom', lines=4, interactive=False)
+
+    user_input = gr.Textbox(label='User Input', lines=2)
+    gen_button = gr.Button('Generate')
+
+    gen_button.click(
+        fn=handle,
+        inputs=[system_input, user_input],
+        outputs=[gpt, mistral, llama2, llama3, llama, bloom],
     )
 
-demo.queue().launch()
+demo.launch()

generators.py

@@ -0,0 +1,106 @@
+import asyncio
+import json
+import os
+
+import aiohttp
+import gradio as gr
+import numpy as np
+import spaces
+from huggingface_hub import InferenceClient
+
+import random
+import torch
+from huggingface_hub import AsyncInferenceClient
+from transformers import LlamaTokenizer, LlamaForCausalLM, AutoTokenizer
+
+
+async def query_llm(payload, model_name):
+    headers = {"Authorization": f"Bearer {os.getenv('HF_TOKEN')}"}
+    async with aiohttp.ClientSession() as session:
+        async with session.post(f"https://api-inference.huggingface.co/models/{model_name}", headers=headers,
+                                json=payload) as response:
+            return await response.json()
+
+
+async def generate_mistral_7bvo1(system_input, user_input):
+    client = AsyncInferenceClient(
+        "mistralai/Mistral-7B-Instruct-v0.1",
+        token=os.getenv('HF_TOKEN'),
+    )
+
+    async for message in await client.chat_completion(
+            messages=[
+                {"role": "system", "content": system_input},
+                {"role": "user", "content": user_input}, ],
+            max_tokens=256,
+            stream=True,
+    ):
+        yield message.choices[0].delta.content
+
+
+async def generate_gpt2(system_input, user_input):
+    output = await query_llm({
+        "inputs": (inputs:=f"{system_input}\n{user_input}"),
+    }, "openai-community/gpt2")
+    yield output[0]["generated_text"].replace(inputs, '')
+
+
+async def generate_llama2(system_input, user_input):
+    client = AsyncInferenceClient(
+        "meta-llama/Llama-2-7b-chat-hf",
+        token=os.getenv('HF_TOKEN')
+    )
+    async for message in await client.chat_completion(
+            messages=[
+                {"role": "system", "content": system_input},
+                {"role": "user", "content": user_input}, ],
+            max_tokens=256,
+            stream=True,
+    ):
+        yield message.choices[0].delta.content
+
+
+@spaces.GPU
+async def generate_openllama(system_input, user_input):
+    model_path = 'openlm-research/open_llama_3b_v2'
+    tokenizer = LlamaTokenizer.from_pretrained(model_path)
+    model = LlamaForCausalLM.from_pretrained(
+        model_path, torch_dtype=torch.float16, device_map='cuda',
+    )
+    # model = model.to("cuda")
+    input_text = f"{system_input}\n{user_input}"
+    input_ids = tokenizer(input_text, return_tensors="pt").input_ids.to("cuda")
+    output = model.generate(input_ids, max_length=128)
+    return tokenizer.decode(output[0], skip_special_tokens=True)
+
+
+@spaces.GPU
+async def generate_bloom(system_input, user_input):
+    model_path = 'bigscience/bloom-7b1'
+    tokenizer = AutoTokenizer.from_pretrained(model_path)
+    model = LlamaForCausalLM.from_pretrained(
+        model_path, torch_dtype=torch.float16, device_map='cuda',
+    )
+    input_text = f"{system_input}\n{user_input}"
+    input_ids = tokenizer(input_text, return_tensors="pt").input_ids.to("cuda")
+    output = model.generate(input_ids, max_length=128)
+    return tokenizer.decode(output[0], skip_special_tokens=True)
+
+
+
+async def generate_llama3(system_input, user_input):
+    client = AsyncInferenceClient(
+        "meta-llama/Meta-Llama-3.1-8B-Instruct",
+        token=os.getenv('HF_TOKEN')
+    )
+    try:
+        async for message in await client.chat_completion(
+                messages=[
+                    {"role": "system", "content": system_input},
+                    {"role": "user", "content": user_input}, ],
+                max_tokens=256,
+                stream=True,
+        ):
+            yield message.choices[0].delta.content
+    except json.JSONDecodeError:
+        pass

utils.py

@@ -0,0 +1,25 @@
+import asyncio
+
+
+async def async_zip_stream(*iterators, default_value=''):
+    tasks = [asyncio.create_task(iterator.__anext__()) for iterator in iterators]
+    done = [False] * len(iterators)
+
+    while not all(done):
+        results = []
+
+        for i, task in enumerate(tasks):
+            if done[i]:
+                results.append(default_value)
+            elif task.done():
+                try:
+                    results.append(task.result())
+                    tasks[i] = asyncio.create_task(iterators[i].__anext__())
+                except StopAsyncIteration:
+                    done[i] = True
+                    results.append(default_value)
+            else:
+                results.append(default_value)
+
+        yield tuple(results)
+        await asyncio.sleep(0.01)  # Slight delay to allow other tasks to progress