File size: 2,781 Bytes
0523608 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 | import sys
import os
import torch
import shlex
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from nanochat.gpt import GPT, GPTConfig
from scripts._sweep_utils import model_dims
from scripts.base_train import get_args
def parse_args_string(cmd):
parser = get_args()
return parser.parse_args(shlex.split(cmd))
def build_model(args):
_, head_dim, model_dim, _ = model_dims(args.depth)
num_heads = model_dim // head_dim
vocab_size = 100288 # dummy for test
config = GPTConfig(
sequence_len=args.max_seq_len,
vocab_size=vocab_size,
n_layer=args.depth,
n_head=num_heads,
n_kv_head=num_heads,
n_embd=model_dim,
use_remixed_linear=args.use_remix_linear,
remixed_linear_kwargs={
'use_basis_gate': args.remix_use_basis_gate == 1,
'use_output_gate': args.remix_use_output_gate == 1,
'use_context': args.remix_use_context == 1,
'basis_gate_mode': args.remix_basis_gate_mode,
'output_gate_rank': 8,
'film_gate': False,
},
remix_basis_size=args.remix_basis_size,
remix_context_dim=args.remix_context_dim,
)
# add all the cclblock args
for k, v in vars(args).items():
if k.startswith('cclblock_'):
setattr(config, k, v)
model = GPT(config)
return model
# Dense cmd
cmd_dense = "--depth 12 --aspect-ratio 64 --head-dim 128 --model-dim 768 --max-seq-len 2048"
args_dense = parse_args_string(cmd_dense)
model_dense = build_model(args_dense)
print(f"Dense: {sum(p.numel() for p in model_dense.parameters()):,}")
# Remix cmd (from the log)
cmd_remix = "--depth 12 --aspect-ratio 64 --head-dim 128 --model-dim 768 --max-seq-len 2048 --remix-use-basis-gate 1 --remix-use-output-gate 1 --remix-use-context 1 --remix-basis-gate-mode mlp --use-remix-linear --remix-basis-size 256 --remix-context-dim 256"
args_remix = parse_args_string(cmd_remix)
model_remix = build_model(args_remix)
print(f"Remixed: {sum(p.numel() for p in model_remix.parameters()):,}")
# Let's count where the params are
def print_param_summary(model):
from collections import defaultdict
counts = defaultdict(int)
for name, p in model.named_parameters():
if 'attn' in name: counts['attn'] += p.numel()
elif 'mlp' in name: counts['mlp'] += p.numel()
elif 'cclblock' in name: counts['cclblock'] += p.numel()
elif 'wte' in name or 'lm_head' in name: counts['embed'] += p.numel()
else: counts['other'] += p.numel()
for k, v in counts.items():
print(f" {k}: {v:,}")
print("\nDense breakdown:")
print_param_summary(model_dense)
print("\nRemixed breakdown:")
print_param_summary(model_remix)
|