code for pile training

.gitignore

@@ -5,6 +5,10 @@
 *.xlsx
 *.xls
 wandb/
+data/
+vocab.json
+*.sh
+*log/
 
 # Byte-compiled / optimized / DLL files
 __pycache__/

RWKV-v4neo/src/dataset.py

@@ -22,11 +22,11 @@ class MyDataset(Dataset):
             self.data_size = len(self.data._bin_buffer) // 2
             print(f"Data has {self.data_size} tokens.")
 
-            if args.my_pile_mode > 0:
+            if args.my_pile_stage > 0:
                 assert self.data_size == 332115325534 and self.vocab_size == 50277 and args.ctx_len == 1024
                 self.samples_per_epoch = args.epoch_steps * int(args.devices) * args.micro_bsz
                 assert self.samples_per_epoch == 40320
-                print(f"########## Pile 20b-tokenized mode {args.my_pile_mode} ##########")
+                print(f"########## Pile 20b-tokenized stage {args.my_pile_stage} ##########")
                 self.magic_prime = 324331313
                 dataset_slot = self.data_size // args.ctx_len
                 assert MaybeIsPrime(self.magic_prime)
@@ -46,7 +46,7 @@ class MyDataset(Dataset):
                     aa = (i) % 10000
                     bb = (i * i) % 10000
                     cc = aa + bb
-                    self.data += f'.{aa}+{bb}={cc}.'
+                    self.data += f".{aa}+{bb}={cc}."
             else:
                 self.data = open(args.data_file, "r", encoding=args.data_type).read()
             print("Building token list...")
@@ -84,7 +84,7 @@ class MyDataset(Dataset):
         ctx_len = args.ctx_len
         req_len = ctx_len + 1
 
-        if args.my_pile_mode > 0:
+        if args.my_pile_stage > 0:
             ii = 1 + epoch * self.samples_per_epoch + (idx * world_size) + rank
             factor = (math.sqrt(5) - 1) / 2
             factor = int(self.magic_prime * factor)

RWKV-v4neo/src/model.py

@@ -12,8 +12,10 @@ from pytorch_lightning.utilities import rank_zero_info, rank_zero_only
 from pytorch_lightning.strategies import DeepSpeedStrategy
 import deepspeed
 from deepspeed.ops.adam import DeepSpeedCPUAdam, FusedAdam
+
 # from deepspeed.runtime.fp16.onebit.zoadam import ZeroOneAdam
 
+
 def __nop(ob):
     return ob
 
@@ -278,7 +280,7 @@ class RWKV(pl.LightningModule):
             lr_2x = set()
             lr_3x = set()
             for n, p in self.named_parameters():
-                if ("time_mix" in n) and (self.args.my_pile_mode == 2):
+                if ("time_mix" in n) and (self.args.my_pile_stage == 2):
                     lr_2x.add(n)
                 elif "time_decay" in n:
                     lr_2x.add(n)
@@ -382,7 +384,7 @@ class RWKV(pl.LightningModule):
                 m[n] = p
             else:
                 if n == "emb.weight":
-                    scale = -25 * self.args.lr_init
+                    scale = -1 * self.args.lr_init
                 else:
                     if shape[0] > shape[1]:
                         gain = math.sqrt(shape[0] / shape[1])
@@ -406,7 +408,7 @@ class RWKV(pl.LightningModule):
                 if scale == 0:
                     nn.init.zeros_(m[n])
                 elif scale < 0:
-                    nn.init.normal_(m[n], mean=0.0, std=-scale)
+                    nn.init.uniform_(m[n], a=scale, b=-scale)
                 else:
                     nn.init.orthogonal_(m[n], gain=gain * scale)
 

RWKV-v4neo/src/trainer.py

@@ -2,9 +2,8 @@ import os, math, time, datetime
 import torch
 from torch.utils.data import DataLoader
 import pytorch_lightning as pl
-from pytorch_lightning import seed_everything
 from pytorch_lightning.utilities import rank_zero_info, rank_zero_only
-from pytorch_lightning.callbacks import TQDMProgressBar
+
 
 class train_callback(pl.Callback):
     def __init__(self, args):
@@ -33,9 +32,9 @@ class train_callback(pl.Callback):
 
         for param_group in trainer.optimizers[0].param_groups:
             if args.layerwise_lr > 0:
-                if self.args.my_pile_mode == 0:
+                if self.args.my_pile_stage != 2:
                     param_group["lr"] = lr * param_group["my_lr_scale"]
-                elif self.args.my_pile_mode == 2:
+                else:
                     if param_group["my_lr_scale"] > 1:
                         param_group["lr"] = lr * 5
                     else:
@@ -63,10 +62,10 @@ class train_callback(pl.Callback):
                     print("Login to wandb...")
                     import wandb
 
-                    model_name = str(args.vocab_size) + "-" + str(args.ctx_len) + "-" + str(args.n_layer) + "-" + str(args.n_embd)
+                    model_name = f"{args.vocab_size} ctx{args.ctx_len} L{args.n_layer} D{args.n_embd}"
                     wandb.init(
                         project=args.wandb,
-                        name=model_name + "-" + args.my_timestamp,
+                        name=model_name + " " + args.my_timestamp,
                         config=args,
                         save_code=False,
                     )
@@ -76,10 +75,12 @@ class train_callback(pl.Callback):
         args = self.args
         if trainer.is_global_zero:  # logging
             t_now = time.time_ns()
+            token_per_step = args.ctx_len * float(args.devices) * args.micro_bsz
+            real_step = trainer.global_step + args.epoch_begin * args.epoch_steps
             try:
                 t_cost = (t_now - trainer.my_time_ns) / 1e9
                 self.log("REAL it/s", 1.0 / t_cost, prog_bar=True, on_step=True)
-                self.log("token/s", args.ctx_len * float(args.devices) * args.micro_bsz / t_cost, prog_bar=True, on_step=True)
+                self.log("Kt/s", token_per_step / t_cost / 1000, prog_bar=True, on_step=True)
             except:
                 pass
             trainer.my_time_ns = t_now
@@ -89,11 +90,12 @@ class train_callback(pl.Callback):
             trainer.my_epoch_loss = trainer.my_loss_sum / trainer.my_loss_count
             self.log("lr", trainer.my_lr, prog_bar=True, on_step=True)
             self.log("loss", trainer.my_epoch_loss, prog_bar=True, on_step=True)
+            # self.log("s", real_step, prog_bar=True, on_step=True)
 
             if len(args.wandb) > 0:
                 trainer.my_wandb.log(
-                    {"loss": trainer.my_loss, "lr": trainer.my_lr},
-                    step=trainer.global_step,
+                    {"loss": trainer.my_loss, "lr": trainer.my_lr, "Gtokens": real_step * token_per_step / 1e9},
+                    step=int(real_step),
                 )
 
     def on_train_epoch_start(self, trainer, pl_module):
@@ -120,11 +122,19 @@ class train_callback(pl.Callback):
 
 
 @rank_zero_only
-def generate_init_weight(model, temp_name):
-    try:
-        os.remove(temp_name)
-    except:
-        pass
+def generate_init_weight(model, init_weight_name):
     mm = model.generate_init_weight()
-    print(f"Saving to {temp_name}...")
-    torch.save(mm, temp_name)
+
+    if model.args.my_pile_stage == 1:
+        print(f"Combine weights from {model.args.load_model}...")
+        load_dict = torch.load(model.args.load_model, map_location="cpu")
+        for k in load_dict:
+            assert k in mm
+            mm[k] = load_dict[k].reshape(mm[k].shape)
+
+    print(f"Save to {init_weight_name}...")
+    torch.save(mm, init_weight_name)
+
+    if model.args.my_pile_stage == 1:
+        print("Done. Now go for stage 2.")
+        exit(0)

RWKV-v4neo/train.py

@@ -3,7 +3,7 @@
 ########################################################################################################
 
 if __name__ == "__main__":
-    print("\n!!! NOTE: THIS IS STILL WIP !!!\n")
+    print("\n!!! work in progress !!!\n")
     import os, warnings, math, datetime, sys, time
     import numpy as np
     from argparse import ArgumentParser
@@ -23,7 +23,7 @@ if __name__ == "__main__":
     warnings.filterwarnings("ignore", ".*The progress bar already tracks a metric with the*")
 
     ########################################################################################################
-    
+    #
     # example: train a simple L12-D768 RWKV on dummy data
     #
     # python train.py --load_model "" --wandb "" --proj_dir "out" \
@@ -91,7 +91,7 @@ if __name__ == "__main__":
     parser.add_argument("--adam_eps", default=1e-8, type=float)
 
     parser.add_argument("--grad_cp", default=0, type=int)  # gradient checkpt: saves VRAM, but slower
-    parser.add_argument("--my_pile_mode", default=0, type=int)  # my special pile mode
+    parser.add_argument("--my_pile_stage", default=0, type=int)  # my special pile mode
     parser.add_argument("--layerwise_lr", default=1, type=int)  # layerwise lr for faster convergence (but slower it/s)
     parser.add_argument("--ds_bucket_mb", default=200, type=int)  # deepspeed bucket size in MB. 200 seems enough
 
@@ -107,11 +107,32 @@ if __name__ == "__main__":
     args.max_epochs = -1  # continue forever
     args.betas = (args.beta1, args.beta2)
 
-    if args.my_pile_mode > 0:
+    if args.my_pile_stage > 0:
         args.epoch_steps = 40320 // (int(args.devices) * args.micro_bsz)
         assert args.epoch_steps * int(args.devices) * args.micro_bsz == 40320
-        if args.my_pile_mode == 2:
+        if args.my_pile_stage == 2:
             assert args.lr_final == args.lr_init
+        if args.my_pile_stage >= 2:  # find latest saved model
+            pths = os.listdir(args.proj_dir)
+            max_p = -1
+            for p in pths:
+                if p.startswith("rwkv") and p.endswith(".pth"):
+                    p = ((p.split("-"))[1].split("."))[0]
+                    if p == "init":
+                        p = -1
+                    else:
+                        p = int(p)
+                    if p > max_p:
+                        max_p = p
+            if max_p == -1:
+                args.load_model = f"{args.proj_dir}/rwkv-init.pth"
+            else:
+                args.load_model = f"{args.proj_dir}/rwkv-{max_p}.pth"
+                if args.my_pile_stage == 2:
+                    args.warmup_steps = 10
+                else:
+                    args.warmup_steps = 50
+            args.epoch_begin = max_p + 1
 
     samples_per_epoch = args.epoch_steps * int(args.devices) * args.micro_bsz
     tokens_per_epoch = samples_per_epoch * args.ctx_len
@@ -185,9 +206,10 @@ if __name__ == "__main__":
 
     model = RWKV(args)
 
-    if len(args.load_model) == 0:
-        args.load_model = f"{args.proj_dir}/rwkv-init.pth"
-        generate_init_weight(model, args.load_model)  # save initial weights to tmp file
+    if len(args.load_model) == 0 or args.my_pile_stage == 1:  # shall we build the initial weights?
+        init_weight_name = f"{args.proj_dir}/rwkv-init.pth"
+        generate_init_weight(model, init_weight_name)  # save initial weights
+        args.load_model = init_weight_name
 
     print(f"########## Loading {args.load_model}... ##########")
     load_dict = torch.load(args.load_model, map_location="cpu")