Page MenuHomePhabricator
Paste P86468

Docker run
ActivePublic
Actions

Authored by gkyziridis on Dec 9 2025, 10:47 AM.
Tags
Referenced Files
F70958890: Docker run
Dec 9 2025, 10:59 AM
F70958842: Docker run
Dec 9 2025, 10:55 AM
F70958727: Docker run
Dec 9 2025, 10:47 AM
Subscribers
gkyziridis
$ docker run --rm -it --network=host torch_rocm3
Python 3.11.2 (main, Apr 28 2025, 14:11:48) [GCC 12.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> from transformers import AutoTokenizer, AutoModel
>>> import torch.nn as nn
>>> try:
... import torch._dynamo
... torch._dynamo.config.suppress_errors = True
... except Exception:
... pass
...
>>> DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
>>> MODEL_NAME = "answerdotai/ModernBERT-base"
>>> print("Device =", DEVICE)
Device = cpu
>>> tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
>>> encoder = AutoModel.from_pretrained(MODEL_NAME).to(DEVICE)
>>> head = nn.Linear(encoder.config.hidden_size, 2).to(DEVICE)
>>> criterion = nn.CrossEntropyLoss()
>>> optimizer = torch.optim.AdamW(
... list(encoder.parameters()) + list(head.parameters()),
... lr=1e-4
... )
>>> texts = ["hello world", "modernbert rocm test"]
>>> batch = tokenizer(texts, padding=True, return_tensors="pt").to(DEVICE)
>>> labels = torch.tensor([0, 1], device=DEVICE)
>>> encoder.train()
ModernBertModel(
(embeddings): ModernBertEmbeddings(
(tok_embeddings): Embedding(50368, 768, padding_idx=50283)
(norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(drop): Dropout(p=0.0, inplace=False)
)
(layers): ModuleList(
(0): ModernBertEncoderLayer(
(attn_norm): Identity()
(attn): ModernBertAttention(
(Wqkv): Linear(in_features=768, out_features=2304, bias=False)
(rotary_emb): ModernBertRotaryEmbedding()
(Wo): Linear(in_features=768, out_features=768, bias=False)
(out_drop): Identity()
)
(mlp_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(mlp): ModernBertMLP(
(Wi): Linear(in_features=768, out_features=2304, bias=False)
(act): GELUActivation()
(drop): Dropout(p=0.0, inplace=False)
(Wo): Linear(in_features=1152, out_features=768, bias=False)
)
)
(1-21): 21 x ModernBertEncoderLayer(
(attn_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(attn): ModernBertAttention(
(Wqkv): Linear(in_features=768, out_features=2304, bias=False)
(rotary_emb): ModernBertRotaryEmbedding()
(Wo): Linear(in_features=768, out_features=768, bias=False)
(out_drop): Identity()
)
(mlp_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(mlp): ModernBertMLP(
(Wi): Linear(in_features=768, out_features=2304, bias=False)
(act): GELUActivation()
(drop): Dropout(p=0.0, inplace=False)
(Wo): Linear(in_features=1152, out_features=768, bias=False)
)
)
)
(final_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
)
>>> head.train()
Linear(in_features=768, out_features=2, bias=True)
>>> outputs = encoder(**batch)
>>> pooled = outputs.last_hidden_state[:, 0, :] # CLS token
>>> logits = head(pooled)
>>> loss = criterion(logits, labels)
>>> print("Loss =", loss.item())
Loss = 0.7854277491569519
>>> loss.backward()
>>> optimizer.step()
# Inference
>>> encoder.eval()
ModernBertModel(
(embeddings): ModernBertEmbeddings(
(tok_embeddings): Embedding(50368, 768, padding_idx=50283)
(norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(drop): Dropout(p=0.0, inplace=False)
)
(layers): ModuleList(
(0): ModernBertEncoderLayer(
(attn_norm): Identity()
(attn): ModernBertAttention(
(Wqkv): Linear(in_features=768, out_features=2304, bias=False)
(rotary_emb): ModernBertRotaryEmbedding()
(Wo): Linear(in_features=768, out_features=768, bias=False)
(out_drop): Identity()
)
(mlp_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(mlp): ModernBertMLP(
(Wi): Linear(in_features=768, out_features=2304, bias=False)
(act): GELUActivation()
(drop): Dropout(p=0.0, inplace=False)
(Wo): Linear(in_features=1152, out_features=768, bias=False)
)
)
(1-21): 21 x ModernBertEncoderLayer(
(attn_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(attn): ModernBertAttention(
(Wqkv): Linear(in_features=768, out_features=2304, bias=False)
(rotary_emb): ModernBertRotaryEmbedding()
(Wo): Linear(in_features=768, out_features=768, bias=False)
(out_drop): Identity()
)
(mlp_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
(mlp): ModernBertMLP(
(Wi): Linear(in_features=768, out_features=2304, bias=False)
(act): GELUActivation()
(drop): Dropout(p=0.0, inplace=False)
(Wo): Linear(in_features=1152, out_features=768, bias=False)
)
)
)
(final_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
)
>>> head.eval()
Linear(in_features=768, out_features=2, bias=True)
>>> with torch.no_grad():
... outputs = encoder(**batch)
... pooled = outputs.last_hidden_state[:, 0, :]
... logits = head(pooled)
>>> all_zero = torch.allclose(logits, torch.zeros_like(logits))
>>> has_nan = torch.isnan(logits).any()
>>> print("\n===== RESULT =====")
>>> print("All-zero logits? :", bool(all_zero))
>>> print("Contains NaNs? :", bool(has_nan))
>>> print("Logits:", logits.cpu())
===== RESULT =====
All-zero logits? : False
Contains NaNs? : False
Logits: tensor([[-0.8974, 0.5434],
[ 0.5851, -0.1151]])
Content licensed under Creative Commons Attribution-ShareAlike (CC BY-SA) 4.0 unless otherwise noted; code licensed under GNU General Public License (GPL) 2.0 or later and other open source licenses. By using this site, you agree to the Terms of Use, Privacy Policy, and Code of Conduct. · Wikimedia Foundation · Privacy Policy · Code of Conduct · Terms of Use · Disclaimer · CC-BY-SA · GPL · Credits