adding rwkv

app.py

@@ -4,12 +4,23 @@ import discord
 from discord.ext import commands
 from dotenv import load_dotenv
 from threading import Thread
-
+from rwkvstic.load import RWKV
+from rwkvstic.agnostic.backends import TORCH
+import torch
 
 load_dotenv()
 bot = commands.Bot("", intents=discord.Intents.all())
 
 # model_pipe = pickle.load(open('pipe.pkl', 'rb'))
+# this is the dtype used for trivial operations, such as vector->vector operations and is the dtype that will determine the accuracy of the model
+runtimedtype = torch.float32 # torch.float64, torch.bfloat16
+
+# this is the dtype used for matrix-vector operations, and is the dtype that will determine the performance and memory usage of the model
+dtype = torch.bfloat16 # torch.float32, torch.float64, torch.bfloat16
+
+useGPU = False # False
+
+model = RWKV("RWKV-4-Pile-3B-Instruct-test2-20230209.pth", mode=TORCH, useGPU=useGPU, runtimedtype=runtimedtype, dtype=dtype)
 
 @bot.event
 async def on_ready():
@@ -20,6 +31,26 @@ async def on_message_event(message):
     if message.author == bot.user:
         return   
 
+    if message.content.startswith("!rwkv "):
+            mess = await message.channel.send("Loading...")
+            model.resetState()
+            model.loadContext(
+                newctx=f"\n\nQuestion: {message.content[6:]}\n\nExpert Long Detailed Response: ")
+            tex = ""
+            for i in range(10):
+                print(i)
+                curr = model.forward(number=10)[
+                    "output"]
+                tex = tex + curr
+                print(curr)
+
+                if ("<|endoftext|>" in curr):
+                    break
+                mess = await mess.edit(content=tex)
+
+            await asyncio.sleep(1)
+            await mess.edit(content=tex)
+
     if message.content:
         print(message.content)
         # await message.channel.send(model_pipe.predict([message.content]))

chatbot.py

@@ -0,0 +1,21 @@
+from rwkvstic.load import RWKV
+from rwkvstic.agnostic.backends import TORCH
+import torch
+import os
+
+os.environ["RWKV_JIT_ON"] = '1'
+os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (much faster)
+# this is the dtype used for trivial operations, such as vector->vector operations and is the dtype that will determine the accuracy of the model
+runtimedtype = torch.float32 # torch.float64, torch.bfloat16
+
+# this is the dtype used for matrix-vector operations, and is the dtype that will determine the performance and memory usage of the model
+dtype = torch.bfloat16 # torch.float32, torch.float64, torch.bfloat16
+
+useGPU = False # False
+
+model = RWKV("RWKV-4-Pile-3B-Instruct-test2-20230209.pth", mode=TORCH, useGPU=useGPU, runtimedtype=runtimedtype, dtype=dtype)
+
+model.loadContext(newctx=f"Q: How many hours are there in a day?\n\nA:")
+output = model.forward(number=100)["output"]
+
+print(output)

requirements copy.txt

@@ -0,0 +1,8 @@
+discord.py==2.1.1
+python-dotenv==1.0.0
+rwkvstic==0.7.0
+torch==1.13.1
+inquirer==3.1.2
+scipy==1.10.1
+nvidia-cuda-runtime-cu12==12.0.146
+--extra-index-url https://pypi.ngc.nvidia.com

requirements.txt

@@ -1,4 +1,10 @@
+# Automatically generated by https://github.com/damnever/pigar.
+
 discord==2.1.0
-requests
-asyncio==3.4.3
-python-dotenv==1.0.0
+python-dotenv==0.21.1
+rwkvstic==0.7.0
+torch==1.13.1+cu116
+inquirer==3.1.2
+scipy==1.10.1
+nvidia-cuda-runtime-cu12==12.0.146
+--extra-index-url https://pypi.ngc.nvidia.com