fp16

RWKV-v4neo/run.py

@@ -22,7 +22,7 @@ args = types.SimpleNamespace()
 ########################################################################################################
 
 args.RUN_DEVICE = "cpu"  # 'cpu' (already very fast) // 'cuda'
-args.FLOAT_MODE = "fp32" # fp32 // bf16 (saves VRAM, slightly less accurate)
+args.FLOAT_MODE = "fp32" # fp32 (good for cpu) // fp16 (might overflow) // bf16 (less accurate)
 # if args.RUN_DEVICE == "cuda":
 #     os.environ["RWKV_RUN_BACKEND"] = 'nvfuser' # !!!BUGGY!!! wrong output
 
@@ -34,7 +34,9 @@ WORD_NAME = [
 UNKNOWN_CHAR = None
 vocab_size = 50277
 
-# note; you can set MODEL_NAME to your fine-tuned model
+# Download Pile models: https://huggingface.co/BlinkDL
+# or, set MODEL_NAME to your fine-tuned model
+
 # MODEL_NAME = "/fsx/BlinkDL/rwkv-release/RWKV-4-Pile-169M-20220807-8023"
 # n_layer = 12
 # n_embd = 768
@@ -50,21 +52,16 @@ vocab_size = 50277
 # n_embd = 2048
 # ctx_len = 1024
 
-# MODEL_NAME = '/fsx/BlinkDL/HF-MODEL/rwkv-4-pile-1b5/RWKV-4-Pile-1B5-20220929-ctx4096'
+# MODEL_NAME = '/fsx/BlinkDL/HF-MODEL/rwkv-4-pile-3b/RWKV-4-Pile-3B-20221008-8023'
-# n_layer = 24
+# n_layer = 32
-# n_embd = 2048
+# n_embd = 2560
-# ctx_len = 4096
+# ctx_len = 1024
 
-MODEL_NAME = '/fsx/BlinkDL/HF-MODEL/rwkv-4-pile-3b/RWKV-4-Pile-3B-20221003-6783'
+MODEL_NAME = '/fsx/BlinkDL/HF-MODEL/rwkv-4-pile-7b/RWKV-4-Pile-7B-20221115-8047'
 n_layer = 32
-n_embd = 2560
+n_embd = 4096
 ctx_len = 1024
 
-# MODEL_NAME = '/fsx/BlinkDL/HF-MODEL/rwkv-4-pile-7b/RWKV-4-Pile-7B-20221004-3047'
-# n_layer = 32
-# n_embd = 4096
-# ctx_len = 1024
-
 args.MODEL_NAME = MODEL_NAME
 args.n_layer = n_layer
 args.n_embd = n_embd

RWKV-v4neo/src/model_run.py

@@ -55,6 +55,8 @@ class RWKV_RNN(nn.Module):
                         w[x] = w[x].float()
                     elif self.FLOAT_MODE == "bf16":
                         w[x] = w[x].bfloat16()
+                    elif self.FLOAT_MODE == "fp16":
+                        w[x] = w[x].half()
 
                 w[x].requires_grad = False
                 if args.RUN_DEVICE == 'cuda' and x != 'emb.weight':
@@ -106,6 +108,10 @@ class RWKV_RNN(nn.Module):
             xk = x * time_mix_k + state[5*i+0].type(torch.bfloat16) * (1 - time_mix_k)
             xr = x * time_mix_r + state[5*i+0].type(torch.bfloat16) * (1 - time_mix_r)
             state[5*i+0] = x.float()
+        elif self.FLOAT_MODE == "fp16":
+            xk = x * time_mix_k + state[5*i+0].half() * (1 - time_mix_k)
+            xr = x * time_mix_r + state[5*i+0].half() * (1 - time_mix_r)
+            state[5*i+0] = x.float()            
         else:
             xk = x * time_mix_k + state[5*i+0] * (1 - time_mix_k)
             xr = x * time_mix_r + state[5*i+0] * (1 - time_mix_r)
@@ -124,6 +130,11 @@ class RWKV_RNN(nn.Module):
             xv = x * time_mix_v + state[5*i+1].type(torch.bfloat16) * (1 - time_mix_v)
             xr = x * time_mix_r + state[5*i+1].type(torch.bfloat16) * (1 - time_mix_r)
             state[5*i+1] = x.float()
+        elif self.FLOAT_MODE == "fp16":
+            xk = x * time_mix_k + state[5*i+1].half() * (1 - time_mix_k)
+            xv = x * time_mix_v + state[5*i+1].half() * (1 - time_mix_v)
+            xr = x * time_mix_r + state[5*i+1].half() * (1 - time_mix_r)
+            state[5*i+1] = x.float()            
         else:
             xk = x * time_mix_k + state[5*i+1] * (1 - time_mix_k)
             xv = x * time_mix_v + state[5*i+1] * (1 - time_mix_v)
@@ -134,7 +145,7 @@ class RWKV_RNN(nn.Module):
         k = kw @ xk
         v = vw @ xv
 
-        if self.FLOAT_MODE == "bf16":
+        if '16' in self.FLOAT_MODE:
             kk = k.float()
             vv = v.float()
         else:
@@ -158,6 +169,8 @@ class RWKV_RNN(nn.Module):
         state[5*i+4] = p
         if self.FLOAT_MODE == "bf16":
             wkv = (a / b).type(torch.bfloat16)
+        elif self.FLOAT_MODE == "fp16":
+            wkv = (a / b).half()
         else:
             wkv = a / b