BlinkDL commited on
Commit
db3b05f
1 Parent(s): 2d6fa0c

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +6 -6
app.py CHANGED
@@ -6,11 +6,11 @@ from pynvml import *
6
  nvmlInit()
7
  gpu_h = nvmlDeviceGetHandleByIndex(0)
8
 
9
- title = "RWKV-4 14B fp16 (DEMO, limited to ctxlen 824)"
10
- desc = '''Links:
11
- <a href='https://github.com/BlinkDL/ChatRWKV' target="_blank" style="margin:0 1em">ChatRWKV</a>
12
- <a href='https://github.com/BlinkDL/RWKV-LM' target="_blank" style="margin:0 1em">RWKV-LM</a>
13
- <a href="https://pypi.org/project/rwkv/" target="_blank" style="margin:0 1em">RWKV pip package</a>
14
  '''
15
 
16
  os.environ["RWKV_JIT_ON"] = '1'
@@ -18,7 +18,7 @@ os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (muc
18
 
19
  from rwkv.model import RWKV
20
  model_path = hf_hub_download(repo_id="BlinkDL/rwkv-4-pile-14b", filename="RWKV-4-Pile-14B-20230213-8019.pth")
21
- model = RWKV(model=model_path, strategy='cuda fp16 *34 -> cpu fp32')
22
  # model_path = hf_hub_download(repo_id="BlinkDL/rwkv-4-pile-169m", filename="RWKV-4-Pile-169M-20220807-8023.pth")
23
  # model = RWKV(model=model_path, strategy='cuda fp16')
24
  from rwkv.utils import PIPELINE, PIPELINE_ARGS
 
6
  nvmlInit()
7
  gpu_h = nvmlDeviceGetHandleByIndex(0)
8
 
9
+ title = "RWKV-4 14B fp16"
10
+ desc = '''DEMO limited to ctxlen 824, and slow because A10g does not have enough VRAM (some layers are computed on CPU instead). Links:
11
+ <a href='https://github.com/BlinkDL/ChatRWKV' target="_blank" style="margin:0 0.5em">ChatRWKV</a>
12
+ <a href='https://github.com/BlinkDL/RWKV-LM' target="_blank" style="margin:0 0.5em">RWKV-LM</a>
13
+ <a href="https://pypi.org/project/rwkv/" target="_blank" style="margin:0 0.5em">RWKV pip package</a>
14
  '''
15
 
16
  os.environ["RWKV_JIT_ON"] = '1'
 
18
 
19
  from rwkv.model import RWKV
20
  model_path = hf_hub_download(repo_id="BlinkDL/rwkv-4-pile-14b", filename="RWKV-4-Pile-14B-20230213-8019.pth")
21
+ model = RWKV(model=model_path, strategy='cuda fp16 *30 -> cpu fp32')
22
  # model_path = hf_hub_download(repo_id="BlinkDL/rwkv-4-pile-169m", filename="RWKV-4-Pile-169M-20220807-8023.pth")
23
  # model = RWKV(model=model_path, strategy='cuda fp16')
24
  from rwkv.utils import PIPELINE, PIPELINE_ARGS