Spaces:
Running
on
T4
Running
on
T4
Update app.py
Browse files
app.py
CHANGED
@@ -21,6 +21,12 @@ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
21 |
ctx_limit = 3500
|
22 |
########################## text rwkv ################################################################
|
23 |
from rwkv.utils import PIPELINE, PIPELINE_ARGS
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
title = "RWKV-5-World-1B5-v2-20231025-ctx4096"
|
25 |
model_path = hf_hub_download(repo_id="BlinkDL/rwkv-5-world", filename=f"{title}.pth")
|
26 |
model = RWKV(model=model_path, strategy='cuda fp16')
|
@@ -65,7 +71,7 @@ def evaluate(
|
|
65 |
state = None
|
66 |
for i in range(int(token_count)):
|
67 |
input_ids = pipeline.encode(ctx)[-ctx_limit:] if i == 0 else [token]
|
68 |
-
out, state =
|
69 |
for n in occurrence:
|
70 |
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
71 |
|
@@ -261,9 +267,9 @@ def chatbot(image, question):
|
|
261 |
|
262 |
##################################################################################################################
|
263 |
with gr.Blocks(title=title) as demo:
|
264 |
-
gr.HTML(f"<div style=\"text-align: center;\">\n<h1>
|
265 |
with gr.Tab("Raw Generation"):
|
266 |
-
gr.Markdown(f"This is [RWKV-
|
267 |
with gr.Row():
|
268 |
with gr.Column():
|
269 |
prompt = gr.Textbox(lines=2, label="Prompt", value="Assistant: Sure! Here is a very detailed plan to create flying pigs:")
|
|
|
21 |
ctx_limit = 3500
|
22 |
########################## text rwkv ################################################################
|
23 |
from rwkv.utils import PIPELINE, PIPELINE_ARGS
|
24 |
+
|
25 |
+
title_v6 = "RWKV-x060-World-1B6-v2-20240208-ctx4096"
|
26 |
+
model_path_v6 = hf_hub_download(repo_id="BlinkDL/rwkv-6-world", filename=f"{title_v6}.pth")
|
27 |
+
model_v6 = RWKV(model=model_path_v6, strategy='cuda fp16')
|
28 |
+
pipeline_v6 = PIPELINE(model_v6, "rwkv_vocab_v20230424")
|
29 |
+
|
30 |
title = "RWKV-5-World-1B5-v2-20231025-ctx4096"
|
31 |
model_path = hf_hub_download(repo_id="BlinkDL/rwkv-5-world", filename=f"{title}.pth")
|
32 |
model = RWKV(model=model_path, strategy='cuda fp16')
|
|
|
71 |
state = None
|
72 |
for i in range(int(token_count)):
|
73 |
input_ids = pipeline.encode(ctx)[-ctx_limit:] if i == 0 else [token]
|
74 |
+
out, state = model_v6.forward(tokens=input_ids, state=state)
|
75 |
for n in occurrence:
|
76 |
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
77 |
|
|
|
267 |
|
268 |
##################################################################################################################
|
269 |
with gr.Blocks(title=title) as demo:
|
270 |
+
gr.HTML(f"<div style=\"text-align: center;\">\n<h1>{title_v6}</h1>\n</div>")
|
271 |
with gr.Tab("Raw Generation"):
|
272 |
+
gr.Markdown(f"This is [RWKV-6 World v2](https://huggingface.co/BlinkDL/rwkv-6-world) with 1.6B params - a 100% attention-free RNN [RWKV-LM](https://github.com/BlinkDL/RWKV-LM). Supports all 100+ world languages and code. And we have [200+ Github RWKV projects](https://github.com/search?o=desc&p=1&q=rwkv&s=updated&type=Repositories). *** Please try examples first (bottom of page) *** (edit them to use your question). Demo limited to ctxlen {ctx_limit}.")
|
273 |
with gr.Row():
|
274 |
with gr.Column():
|
275 |
prompt = gr.Textbox(lines=2, label="Prompt", value="Assistant: Sure! Here is a very detailed plan to create flying pigs:")
|