A
Alpha
20260223071621_y46kcompletedchat1.85M params16m 9s elapsed · Updated 2d ago
4L / 128D / 4H · helios · bpe-4k · adamw· Created Feb 23, 2026 7:16 AM
Step 2,000 / 2,000100.0%
5.7951
Loss?
5.1923
Best Loss?
-30.4% from start
5.8732
Val Loss?
best: 5.8732
2.28e-10
Learning Rate?
4,196
Throughput?
tok/s (avg)
495
Speed?
ms/iter (avg)
1043.369
Grad Norm?
avg: 552.975
4.08M
Tokens
processed
116ms
Forward
23% of step
340ms
Backward
69% of step
6ms
GPU Sync
1% of step
737
GPU Ops
per step
0.2%
MFU
model FLOPS util
2.9x
Bwd/Fwd
ratio
Loss Curve ?
Validation / selection
Run diagnostics
Gradient + Loss
Train/val loss with validation, warmup, overfit, and instability markers.
Search-aware view
Architecture
Layers?4
Embedding?128
Heads?4
Vocab?4,000
Context?256
Dropout?0
Parameters?1.85M
Training Config
Total iters?2,000
Batch size?8
Max LR?0.0003
Optimizer?adamw
Backend?helios
Tokenizer?bpe-4k
Seed?42
Weight decay?0.01
Grad clip?1
Eval interval?500
GPU & VRAM
No GPU data
Learning Rate
Grad Norm
Step Time Breakdown
Step Time Breakdown
Checkpoints (0) ?
No checkpoints saved
Sample Generations (5)
#CheckpointPrompt (preview)Generated
1-<|user|> Hello, how are you? <|assistant|>2d ago
Prompt
<|user|> Hello, how are you? <|assistant|>
Output
<|user|> Hello, how are you? <|assistant|> <|user|> d_t ds t|> tbs kos g_texnof_text|> <|user|> O t|> <|assistant|> <|user|> ie<|end_md_<|assistant|> <|user|> I d_<|assistant|>
2-<|user|> What do you like to do for fun? <|assistant|>2d ago
Prompt
<|user|> What do you like to do for fun? <|assistant|>
Output
<|user|> What do you like to do for fun? <|assistant|> <|assistant|> ia <|assistant|> d_t|> s <|assistant|> itexofys t|> _oft <|user|> n_Eh s<|end_t <|user|> a to it ein io s <|user|> I in
3-<|user|> Tell me about yourself. <|assistant|>2d ago
Prompt
<|user|> Tell me about yourself. <|assistant|>
Output
<|user|> Tell me about yourself. <|assistant|>ayenc d_of_text|> <|assistant|> itext|> <|assistant|> k <|assistant|> you tex <|assistant|> it <|user|> p ttld_tex text|> h<|user|>
4-In the beginning 2d ago
Prompt
In the beginning
Output
In the beginning stm uing my aa g <|assistant|> <|assistant|> i ce <|assistant|> Si d_texin a seg<|assistant|> I d_of_of_t|> to retexlutext|> reu <|assistant|> ruotexof_ce<|end_of_t|> <|user|> kad_of_of_texh te it that a<|endnhrtexof_text
5-We the People of 2d ago
Prompt
We the People of
Output
We the People of the <|assistant|> Tn<|assistant|> the pa <|assistant|> Gm<|assistant|> wai k <|assistant|> i d<|user|> Ttexti it is in <|assistant|> H<|user|> Gi <|assistant|> m <|user|> A<|end_t’_t|> <|user|> i like s, I ds <|assistant|> Am, <|assistant|> s|> k <|assistant|> cing <|assistant|> hend_t_texof_of_textexof_textexof_text<|assistant|> re<|user|> 
{
  "vocabSize": 4000,
  "blockSize": 256,
  "nLayer": 4,
  "nEmbd": 128,
  "nHead": 4,
  "dropout": 0
}
{
  "iters": 2000,
  "batchSize": 8,
  "lr": 0.0003,
  "beta1": 0.9,
  "beta2": 0.999,
  "eps": 0.000001,
  "weightDecay": 0.01,
  "gradClip": 1,
  "evalInterval": 500,
  "evalIters": 10,
  "seed": 42,
  "backend": "helios",
  "tokenizer": "bpe-4k",
  "optimizer": "adamw",
  "logLevel": "info",
  "trace": false,
  "gradAccumSteps": 1,
  "sampleInterval": 500
}