Update app.py

app.py

@@ -12,7 +12,6 @@ from io import BytesIO
 import torch
 import torch.nn.functional as F
 from datetime import datetime
-from transformers import CLIPImageProcessor
 from huggingface_hub import hf_hub_download
 from pynvml import *
 nvmlInit()
@@ -20,14 +19,13 @@ gpu_h = nvmlDeviceGetHandleByIndex(0)
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
 ctx_limit = 4000
-gen_limit = 700
+gen_limit = 1000
 
 ########################## text rwkv ################################################################
 from rwkv.utils import PIPELINE, PIPELINE_ARGS
 
-title_v6 = "RWKV-x070-World-2.9B-v3-20250211-ctx4096"
-model_path_v6 = hf_hub_download(repo_id="BlinkDL/rwkv-7-world", filename=f"{title_v6}.pth")
-# model_path_v6 = f'/mnt/e/RWKV-Runner/models/{title_v6}' # conda activate torch2; cd /mnt/program/git-public/RWKV-Gradio-1; python app.py
+title_v6 = "rwkv7-g1-2.9b-20250519-ctx4096"
+model_path_v6 = hf_hub_download(repo_id="BlinkDL/rwkv7-g1", filename=f"{title_v6}.pth")
 model_v6 = RWKV(model=model_path_v6.replace('.pth',''), strategy='cuda fp16')
 pipeline_v6 = PIPELINE(model_v6, "rwkv_vocab_v20230424")
 
@@ -108,7 +106,7 @@ def evaluate(
     yield out_str.strip()
 
 examples = [
-    [generate_prompt("I have a dataset which contains a list of 2D images, given a new image, how to find the closest image in the dataset"), gen_limit, 1, 0.3, 0.5, 0.5],
+    ["User: simulate SpaceX mars landing using python\n\nAssistant: <think", gen_limit, 1, 0.3, 0.5, 0.5],
     [generate_prompt("Please give the pros and cons of hodl versus active trading."), gen_limit, 1, 0.3, 0.5, 0.5],
     ["Assistant: How can we craft an engaging story featuring vampires on Mars? Let's think step by step and provide an expert response:", gen_limit, 1, 0.3, 0.5, 0.5],
     ["Assistant: How can we persuade Elon Musk to follow you on Twitter? Let's think step by step and provide an expert response:", gen_limit, 1, 0.3, 0.5, 0.5],
@@ -129,10 +127,10 @@ with gr.Blocks(title=title_v6) as demo:
     gr.HTML(f"<div style=\"text-align: center;\">\n<h1>{title_v6}</h1>\n</div>")
 
     with gr.Tab("=== Base Model (Raw Generation) ==="):
-        gr.Markdown(f"This is [RWKV-7 World v3](https://huggingface.co/BlinkDL/rwkv-7-world) 2.9B (L32-D2560) base LM - an attention-free pure RNN [RWKV-LM](https://github.com/BlinkDL/RWKV-LM). Supports 100+ world languages and code. Check [400+ Github RWKV projects](https://github.com/search?o=desc&p=1&q=rwkv&s=updated&type=Repositories). *** Can try examples (bottom of page) *** (can edit them). Demo limited to ctxlen {ctx_limit}.")
+        gr.Markdown(f'This is [RWKV7 G1](https://huggingface.co/BlinkDL/rwkv7-g1) 2.9B reasoning base LM - an attention-free pure RNN [RWKV-LM](https://github.com/BlinkDL/RWKV-LM). Supports 100+ world languages and code. Check [400+ Github RWKV projects](https://github.com/search?o=desc&p=1&q=rwkv&s=updated&type=Repositories). *** Can try examples (bottom of page) *** (can edit them). Demo limited to ctxlen {ctx_limit}.')
         with gr.Row():
             with gr.Column():
-                prompt = gr.Textbox(lines=6, label="Prompt", value=generate_prompt("I have a dataset which contains a list of 2D images, given a new image, how to find the closest image in the dataset"))
+                prompt = gr.Textbox(lines=6, label="Prompt", value="User: simulate SpaceX mars landing using python\n\nAssistant: <think")
                 token_count = gr.Slider(10, gen_limit, label="Max Tokens", step=10, value=gen_limit)
                 temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
                 top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.3)
@@ -149,4 +147,4 @@ with gr.Blocks(title=title_v6) as demo:
         data.click(lambda x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])
 
 demo.queue(concurrency_count=1, max_size=10)
-demo.launch(share=False)
+demo.launch(share=False)