Update app.py

app.py

@@ -6,7 +6,7 @@ from pynvml import *
 nvmlInit()
 gpu_h = nvmlDeviceGetHandleByIndex(0)
 
-title = "RWKV-4 14B fp16 ctx1024"
+title = "RWKV-4 14B fp16 (DEMO, limited to ctxlen 824)"
 desc = '''Links:
 <a href='https://github.com/BlinkDL/ChatRWKV' target="_blank" style="margin:0 1em">ChatRWKV</a>
 <a href='https://github.com/BlinkDL/RWKV-LM' target="_blank" style="margin:0 1em">RWKV-LM</a>
@@ -17,10 +17,10 @@ os.environ["RWKV_JIT_ON"] = '1'
 os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (much faster)
 
 from rwkv.model import RWKV
-# model_path = hf_hub_download(repo_id="BlinkDL/rwkv-4-pile-14b", filename="RWKV-4-Pile-14B-20230213-8019.pth")
-# model = RWKV(model=model_path, strategy='cuda fp16 *34 -> cpu fp32')
-model_path = hf_hub_download(repo_id="BlinkDL/rwkv-4-pile-169m", filename="RWKV-4-Pile-169M-20220807-8023.pth")
-model = RWKV(model=model_path, strategy='cuda fp16')
+model_path = hf_hub_download(repo_id="BlinkDL/rwkv-4-pile-14b", filename="RWKV-4-Pile-14B-20230213-8019.pth")
+model = RWKV(model=model_path, strategy='cuda fp16 *34 -> cpu fp32')
+# model_path = hf_hub_download(repo_id="BlinkDL/rwkv-4-pile-169m", filename="RWKV-4-Pile-169M-20220807-8023.pth")
+# model = RWKV(model=model_path, strategy='cuda fp16')
 from rwkv.utils import PIPELINE, PIPELINE_ARGS
 pipeline = PIPELINE(model, "20B_tokenizer.json")
 
@@ -53,7 +53,7 @@ def infer(
     occurrence = {}
     state = None
     for i in range(int(token_count)):
-        out, state = model.forward(pipeline.encode(ctx)[:1024] if i == 0 else [token], state)
+        out, state = model.forward(pipeline.encode(ctx)[:824] if i == 0 else [token], state)
         for n in args.token_ban:
             out[n] = -float('inf')
         for n in occurrence: