tacit/nanochat
1
0
Fork 0
mirror of https://github.com/karpathy/nanochat.git synced 2025-12-06 04:12:13 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #13 from LokiMetaSmith/fix-cpu-ddp-init

Browse Source 
Fix CPU DDP crashes, enable ROCm detection, and prevent single-proces…
This commit is contained in:
Lawrence R Kincheloe III 2025-11-22 12:07:29 -06:00 committed by GitHub
commit 3b3113c8d2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 38 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
nanochat/gpt.py
View File

@ -228,11 +228,13 @@ class GPT(nn.Module):
dict(params=embedding_params, lr=embedding_lr * dmodel_lr_scale),
]
adamw_kwargs = dict(betas=(0.8, 0.95), eps=1e-10, weight_decay=weight_decay)
AdamWFactory = DistAdamW if ddp else partial(torch.optim.AdamW, fused=True)
# Use distributed optimizers only if we are actually distributed (world_size > 1)
use_dist_optim = ddp and world_size > 1
AdamWFactory = DistAdamW if use_dist_optim else partial(torch.optim.AdamW, fused=True)
adamw_optimizer = AdamWFactory(adam_groups, **adamw_kwargs)
# Create the Muon optimizer for the linear layers
muon_kwargs = dict(lr=matrix_lr, momentum=0.95)
MuonFactory = DistMuon if ddp else Muon
MuonFactory = DistMuon if use_dist_optim else Muon
muon_optimizer = MuonFactory(matrix_params, **muon_kwargs)
# Combine them the two optimizers into one list
optimizers = [adamw_optimizer, muon_optimizer]

34
output.txt Normal file
View File

@ -0,0 +1,34 @@
█████ █████
░░███ ░░███
████████ ██████ ████████ ██████ ██████ ░███████ ██████ ███████
░░███░░███ ░░░░░███ ░░███░░███ ███░░███ ███░░███ ░███░░███ ░░░░░███░░░███░
░███ ░███ ███████ ░███ ░███ ░███ ░███░███ ░░░ ░███ ░███ ███████ ░███
░███ ░███ ███░░███ ░███ ░███ ░███ ░███░███ ███ ░███ ░███ ███░░███ ░███ ███
████ █████░░████████ ████ █████░░██████ ░░██████ ████ █████░░███████ ░░█████
░░░░ ░░░░░ ░░░░░░░░ ░░░░ ░░░░░ ░░░░░░ ░░░░░░ ░░░░ ░░░░░ ░░░░░░░░ ░░░░░
Overriding: depth = 2
Overriding: device_batch_size = 2
Overriding: max_seq_len = 128
Overriding: num_iterations = 10
Overriding: run = dummy
Autodetected device type: cpu
2025-11-22 17:01:10,440 - nanochat.common - INFO - Distributed world size: 1
/app/nanochat/tokenizer.py:397: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
token_bytes = torch.load(f, map_location=device)
Vocab size: 265
num_layers: 2
model_dim: 128
num_heads: 1
num_kv_heads: 1
Tokens / micro-batch / rank: 2 x 128 = 256
Tokens / micro-batch: 256
Total batch size 524,288 => gradient accumulation steps: 2048
Number of parameters: 461,056
Estimated FLOPs per token: 2.956032e+06
Using user-provided number of iterations: 10
Total number of training tokens: 5,242,880
Tokens : Params ratio: 11.37
Total training FLOPs estimate: 1.549812e+13
Scaling the LR for the AdamW parameters ∝1/√(128/768) = 2.449490