Upload Chimera 279M at step 36500

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
+training_curves.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -24,8 +24,12 @@ language:
 - **Topology:** 4 unique bottom + 4×3 shared top
 - **GDN:Attn ratio:** 3:1 (every 4th layer is attention)
 
+## Training Curves
+
+![Training Curves](training_curves.png)
+
 ## Training
-- **Step:** 40,000
+- **Step:** 36,500
 - **Data:** Mixed (75% FineWeb-Edu, 18% StarCoder, 5% FineMath, 2% UltraChat)
 - **Framework:** Zara-ML (custom PyTorch)
 

config.json

@@ -26,7 +26,7 @@
   "architecture": "Chimera",
   "config_class": "ChimeraConfig",
   "topology": "4 bottom + 4x3 top = 16 virtual",
-  "step": 40000,
+  "step": 36500,
   "total_params": 278664160,
   "size_label": "279M",
   "model_type": "zara-ml"

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1edf6a5161d145aa61346e55804f248152a9266c9cffd2c684044089a47c52e
+oid sha256:dc675b0c28dd0a60866e25ed6a56e66734c83359c0782f40c9fe72dfe6446834
 size 868508712

training.log

@@ -51,9 +51,8 @@ Scalar params: 119,776 (77 tensors) @ LR 8.0e-05
 Auto-detected checkpoint: checkpoints/spark_05b/best.pt
 
 Resuming from checkpoint: checkpoints/spark_05b/best.pt
-  Restored optimizer state
-  Resuming from step 39500, best_val_loss=3.5300
-  Remaining: 210500 steps
+  Resuming from step 32000, best_val_loss=4.0522
+  Remaining: 218000 steps
 Compiling model with torch.compile...
 
 Training for 250000 steps (warmup=100)
@@ -61,84 +60,665 @@ LR: 0.0008, batch_size: 16
 Tokens/step: 32,768
 
 Starting step 1 (first step may be slow — Triton kernel compilation)...
-step  39510/250000 | loss 3.1408 | lr 8.00e-04 emb 4.00e-04 | 4236ms/step | 7,735 tok/s | epoch 1
-step  39520/250000 | loss 3.1581 | lr 8.00e-04 emb 4.00e-04 | 3852ms/step | 8,507 tok/s | epoch 1
-step  39530/250000 | loss 3.1745 | lr 8.00e-04 emb 4.00e-04 | 3729ms/step | 8,788 tok/s | epoch 1
-step  39540/250000 | loss 3.2080 | lr 8.00e-04 emb 4.00e-04 | 3666ms/step | 8,937 tok/s | epoch 1
-step  39550/250000 | loss 3.2982 | lr 8.00e-04 emb 4.00e-04 | 3628ms/step | 9,031 tok/s | epoch 1
-step  39560/250000 | loss 3.2445 | lr 8.00e-04 emb 4.00e-04 | 3604ms/step | 9,093 tok/s | epoch 1
-step  39570/250000 | loss 3.2455 | lr 8.00e-04 emb 4.00e-04 | 3586ms/step | 9,138 tok/s | epoch 1
-step  39580/250000 | loss 3.2044 | lr 8.00e-04 emb 4.00e-04 | 3575ms/step | 9,166 tok/s | epoch 1
-step  39590/250000 | loss 3.2730 | lr 8.00e-04 emb 4.00e-04 | 3565ms/step | 9,192 tok/s | epoch 1
-step  39600/250000 | loss 3.2686 | lr 8.00e-04 emb 4.00e-04 | 3556ms/step | 9,214 tok/s | epoch 1
-step  39610/250000 | loss 3.1287 | lr 8.00e-04 emb 4.00e-04 | 3550ms/step | 9,230 tok/s | epoch 1
-step  39620/250000 | loss 3.3391 | lr 8.00e-04 emb 4.00e-04 | 3544ms/step | 9,247 tok/s | epoch 1
-step  39630/250000 | loss 3.2220 | lr 8.00e-04 emb 4.00e-04 | 3538ms/step | 9,262 tok/s | epoch 1
-step  39640/250000 | loss 3.2508 | lr 8.00e-04 emb 4.00e-04 | 3533ms/step | 9,274 tok/s | epoch 1
-step  39650/250000 | loss 3.1966 | lr 8.00e-04 emb 4.00e-04 | 3529ms/step | 9,285 tok/s | epoch 1
-step  39660/250000 | loss 3.2953 | lr 8.00e-04 emb 4.00e-04 | 3525ms/step | 9,296 tok/s | epoch 1
-step  39670/250000 | loss 3.2249 | lr 8.00e-04 emb 4.00e-04 | 3522ms/step | 9,304 tok/s | epoch 1
-step  39680/250000 | loss 3.1715 | lr 8.00e-04 emb 4.00e-04 | 3518ms/step | 9,313 tok/s | epoch 1
-step  39690/250000 | loss 3.2230 | lr 8.00e-04 emb 4.00e-04 | 3516ms/step | 9,320 tok/s | epoch 1
-step  39700/250000 | loss 3.2068 | lr 8.00e-04 emb 4.00e-04 | 3513ms/step | 9,327 tok/s | epoch 1
-step  39710/250000 | loss 3.0613 | lr 8.00e-04 emb 4.00e-04 | 3512ms/step | 9,331 tok/s | epoch 1
-step  39720/250000 | loss 3.2250 | lr 8.00e-04 emb 4.00e-04 | 3510ms/step | 9,335 tok/s | epoch 1
-step  39730/250000 | loss 3.1740 | lr 8.00e-04 emb 4.00e-04 | 3509ms/step | 9,339 tok/s | epoch 1
-step  39740/250000 | loss 3.2110 | lr 8.00e-04 emb 4.00e-04 | 3507ms/step | 9,343 tok/s | epoch 1
-step  39750/250000 | loss 3.2288 | lr 8.00e-04 emb 4.00e-04 | 3506ms/step | 9,345 tok/s | epoch 1
-step  39760/250000 | loss 3.1548 | lr 8.00e-04 emb 4.00e-04 | 3506ms/step | 9,347 tok/s | epoch 1
-step  39770/250000 | loss 3.2076 | lr 8.00e-04 emb 4.00e-04 | 3505ms/step | 9,349 tok/s | epoch 1
-step  39780/250000 | loss 3.2501 | lr 8.00e-04 emb 4.00e-04 | 3504ms/step | 9,351 tok/s | epoch 1
-step  39790/250000 | loss 3.0604 | lr 8.00e-04 emb 4.00e-04 | 3503ms/step | 9,354 tok/s | epoch 1
-step  39800/250000 | loss 3.2725 | lr 8.00e-04 emb 4.00e-04 | 3502ms/step | 9,356 tok/s | epoch 1
-step  39810/250000 | loss 3.0996 | lr 8.00e-04 emb 4.00e-04 | 3501ms/step | 9,359 tok/s | epoch 1
-step  39820/250000 | loss 3.2154 | lr 8.00e-04 emb 4.00e-04 | 3500ms/step | 9,361 tok/s | epoch 1
-step  39830/250000 | loss 3.2169 | lr 8.00e-04 emb 4.00e-04 | 3499ms/step | 9,364 tok/s | epoch 1
-step  39840/250000 | loss 3.2018 | lr 8.00e-04 emb 4.00e-04 | 3498ms/step | 9,366 tok/s | epoch 1
-step  39850/250000 | loss 3.2629 | lr 8.00e-04 emb 4.00e-04 | 3497ms/step | 9,369 tok/s | epoch 1
-step  39860/250000 | loss 3.2325 | lr 8.00e-04 emb 4.00e-04 | 3496ms/step | 9,372 tok/s | epoch 1
-step  39870/250000 | loss 3.2538 | lr 8.00e-04 emb 4.00e-04 | 3496ms/step | 9,374 tok/s | epoch 1
-step  39880/250000 | loss 3.2490 | lr 8.00e-04 emb 4.00e-04 | 3495ms/step | 9,375 tok/s | epoch 1
-step  39890/250000 | loss 3.0916 | lr 8.00e-04 emb 4.00e-04 | 3494ms/step | 9,377 tok/s | epoch 1
-step  39900/250000 | loss 3.1592 | lr 8.00e-04 emb 4.00e-04 | 3494ms/step | 9,379 tok/s | epoch 1
-step  39910/250000 | loss 3.2033 | lr 8.00e-04 emb 4.00e-04 | 3493ms/step | 9,381 tok/s | epoch 1
-step  39920/250000 | loss 3.1877 | lr 8.00e-04 emb 4.00e-04 | 3493ms/step | 9,382 tok/s | epoch 1
-step  39930/250000 | loss 3.2243 | lr 8.00e-04 emb 4.00e-04 | 3492ms/step | 9,384 tok/s | epoch 1
-step  39940/250000 | loss 3.1857 | lr 8.00e-04 emb 4.00e-04 | 3491ms/step | 9,385 tok/s | epoch 1
-step  39950/250000 | loss 3.2684 | lr 8.00e-04 emb 4.00e-04 | 3491ms/step | 9,386 tok/s | epoch 1
-step  39960/250000 | loss 3.2554 | lr 8.00e-04 emb 4.00e-04 | 3491ms/step | 9,387 tok/s | epoch 1
-step  39970/250000 | loss 3.1867 | lr 8.00e-04 emb 4.00e-04 | 3491ms/step | 9,388 tok/s | epoch 1
-step  39980/250000 | loss 3.1958 | lr 8.00e-04 emb 4.00e-04 | 3490ms/step | 9,388 tok/s | epoch 1
-step  39990/250000 | loss 3.1424 | lr 8.00e-04 emb 4.00e-04 | 3490ms/step | 9,389 tok/s | epoch 1
-step  40000/250000 | loss 3.1128 | lr 8.00e-04 emb 4.00e-04 | 3490ms/step | 9,390 tok/s | epoch 1
-W0322 17:12:09.686000 95849 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8] torch._dynamo hit config.recompile_limit (8)
-W0322 17:12:09.686000 95849 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8]    function: 'rearrange' (/home/fy/Dev/zara_ml/.venv/lib/python3.12/site-packages/einops/einops.py:561)
-W0322 17:12:09.686000 95849 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8]    last reason: 7/7: tensor 'tensor' rank mismatch. expected 4, actual 3
-W0322 17:12:09.686000 95849 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8] To log all recompilation reasons, use TORCH_LOGS="recompiles".
-W0322 17:12:09.686000 95849 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8] To diagnose recompilation issues, see https://docs.pytorch.org/docs/main/user_guide/torch_compiler/compile/programming_model.recompilation.html
-W0322 17:12:11.802000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-W0322 17:12:11.805000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-W0322 17:12:11.806000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-W0322 17:12:11.808000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-W0322 17:12:11.810000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-W0322 17:12:11.812000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-W0322 17:12:11.814000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-W0322 17:12:11.819000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-W0322 17:12:11.838000 95849 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
-  >>> val_loss: 3.5245 | bpt: 5.0848 | true_bpb: 1.6357 *BEST*
-  >>> [The] The South has taken advantage of efforts to further combat the financial and political crisis in South Africa.
-It will have to face cautious planning, political turmoil and non-medical and social shocks and anxiety in order to fight the financial and political crisis in South Africa.
-The Korean War has seen an outbreak of anxiety and fear. Central Asian leaders have responded to it, and as a result, it has lost confidence and
-  >>> [Scientists have discovered] Scientists have discovered the world's greatest satellite, the asteroid Curiosity.
-The asteroid was discovered by NASA's Earth Observatory in Hawaii and by ESA's Mars Reconnaissance Orbital Observatory in Hawaii. It found that it was the first time that a planet had been seen in Earth's orbit. It is a giant object and crucial to the study of the history of our planet.
-The asteroid was spotted on December 21
-step  40010/250000 | loss 3.2186 | lr 8.00e-04 emb 4.00e-04 | 3705ms/step | 8,845 tok/s | epoch 1
-step  40020/250000 | loss 3.2086 | lr 8.00e-04 emb 4.00e-04 | 3700ms/step | 8,856 tok/s | epoch 1
-step  40030/250000 | loss 3.0996 | lr 8.00e-04 emb 4.00e-04 | 3695ms/step | 8,867 tok/s | epoch 1
-step  40040/250000 | loss 3.2232 | lr 8.00e-04 emb 4.00e-04 | 3692ms/step | 8,876 tok/s | epoch 1
-step  40050/250000 | loss 3.1587 | lr 8.00e-04 emb 4.00e-04 | 3688ms/step | 8,885 tok/s | epoch 1
-step  40060/250000 | loss 3.2029 | lr 8.00e-04 emb 4.00e-04 | 3685ms/step | 8,893 tok/s | epoch 1
-step  40070/250000 | loss 3.1416 | lr 8.00e-04 emb 4.00e-04 | 3681ms/step | 8,902 tok/s | epoch 1
-step  40080/250000 | loss 3.2118 | lr 8.00e-04 emb 4.00e-04 | 3677ms/step | 8,911 tok/s | epoch 1
-step  40090/250000 | loss 3.2299 | lr 8.00e-04 emb 4.00e-04 | 3674ms/step | 8,919 tok/s | epoch 1
-step  40100/250000 | loss 3.1665 | lr 8.00e-04 emb 4.00e-04 | 3671ms/step | 8,927 tok/s | epoch 1
+step  32010/250000 | loss 4.3253 | lr 8.00e-04 emb 4.00e-04 | 4294ms/step | 7,631 tok/s | epoch 1
+step  32020/250000 | loss 4.3835 | lr 8.00e-04 emb 4.00e-04 | 3893ms/step | 8,416 tok/s | epoch 1
+step  32030/250000 | loss 4.3459 | lr 8.00e-04 emb 4.00e-04 | 3765ms/step | 8,704 tok/s | epoch 1
+step  32040/250000 | loss 4.4343 | lr 8.00e-04 emb 4.00e-04 | 3698ms/step | 8,861 tok/s | epoch 1
+step  32050/250000 | loss 4.3790 | lr 8.00e-04 emb 4.00e-04 | 3660ms/step | 8,952 tok/s | epoch 1
+step  32060/250000 | loss 4.1950 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,011 tok/s | epoch 1
+step  32070/250000 | loss 4.1609 | lr 8.00e-04 emb 4.00e-04 | 3615ms/step | 9,063 tok/s | epoch 1
+step  32080/250000 | loss 4.1494 | lr 8.00e-04 emb 4.00e-04 | 3605ms/step | 9,089 tok/s | epoch 1
+step  32090/250000 | loss 4.2361 | lr 8.00e-04 emb 4.00e-04 | 3594ms/step | 9,118 tok/s | epoch 1
+step  32100/250000 | loss 4.1286 | lr 8.00e-04 emb 4.00e-04 | 3585ms/step | 9,141 tok/s | epoch 1
+step  32110/250000 | loss 4.0256 | lr 8.00e-04 emb 4.00e-04 | 3576ms/step | 9,163 tok/s | epoch 1
+step  32120/250000 | loss 4.0525 | lr 8.00e-04 emb 4.00e-04 | 3569ms/step | 9,181 tok/s | epoch 1
+step  32130/250000 | loss 3.9968 | lr 8.00e-04 emb 4.00e-04 | 3563ms/step | 9,197 tok/s | epoch 1
+step  32140/250000 | loss 4.0342 | lr 8.00e-04 emb 4.00e-04 | 3559ms/step | 9,207 tok/s | epoch 1
+step  32150/250000 | loss 4.0057 | lr 8.00e-04 emb 4.00e-04 | 3555ms/step | 9,217 tok/s | epoch 1
+step  32160/250000 | loss 4.0607 | lr 8.00e-04 emb 4.00e-04 | 3550ms/step | 9,230 tok/s | epoch 1
+step  32170/250000 | loss 4.0601 | lr 8.00e-04 emb 4.00e-04 | 3548ms/step | 9,236 tok/s | epoch 1
+step  32180/250000 | loss 3.9072 | lr 8.00e-04 emb 4.00e-04 | 3546ms/step | 9,242 tok/s | epoch 1
+step  32190/250000 | loss 3.9479 | lr 8.00e-04 emb 4.00e-04 | 3543ms/step | 9,249 tok/s | epoch 1
+step  32200/250000 | loss 3.9923 | lr 8.00e-04 emb 4.00e-04 | 3541ms/step | 9,254 tok/s | epoch 1
+step  32210/250000 | loss 3.8755 | lr 8.00e-04 emb 4.00e-04 | 3539ms/step | 9,260 tok/s | epoch 1
+step  32220/250000 | loss 3.8866 | lr 8.00e-04 emb 4.00e-04 | 3537ms/step | 9,265 tok/s | epoch 1
+step  32230/250000 | loss 3.8966 | lr 8.00e-04 emb 4.00e-04 | 3535ms/step | 9,269 tok/s | epoch 1
+step  32240/250000 | loss 3.8574 | lr 8.00e-04 emb 4.00e-04 | 3534ms/step | 9,273 tok/s | epoch 1
+step  32250/250000 | loss 3.8730 | lr 8.00e-04 emb 4.00e-04 | 3532ms/step | 9,277 tok/s | epoch 1
+step  32260/250000 | loss 3.8584 | lr 8.00e-04 emb 4.00e-04 | 3531ms/step | 9,280 tok/s | epoch 1
+step  32270/250000 | loss 3.8414 | lr 8.00e-04 emb 4.00e-04 | 3530ms/step | 9,283 tok/s | epoch 1
+step  32280/250000 | loss 3.8643 | lr 8.00e-04 emb 4.00e-04 | 3529ms/step | 9,286 tok/s | epoch 1
+step  32290/250000 | loss 3.7273 | lr 8.00e-04 emb 4.00e-04 | 3528ms/step | 9,288 tok/s | epoch 1
+step  32300/250000 | loss 3.9331 | lr 8.00e-04 emb 4.00e-04 | 3527ms/step | 9,290 tok/s | epoch 1
+step  32310/250000 | loss 3.7946 | lr 8.00e-04 emb 4.00e-04 | 3527ms/step | 9,292 tok/s | epoch 1
+step  32320/250000 | loss 3.8370 | lr 8.00e-04 emb 4.00e-04 | 3525ms/step | 9,295 tok/s | epoch 1
+step  32330/250000 | loss 3.8173 | lr 8.00e-04 emb 4.00e-04 | 3524ms/step | 9,298 tok/s | epoch 1
+step  32340/250000 | loss 3.8264 | lr 8.00e-04 emb 4.00e-04 | 3523ms/step | 9,300 tok/s | epoch 1
+step  32350/250000 | loss 3.8209 | lr 8.00e-04 emb 4.00e-04 | 3523ms/step | 9,302 tok/s | epoch 1
+step  32360/250000 | loss 3.7796 | lr 8.00e-04 emb 4.00e-04 | 3522ms/step | 9,304 tok/s | epoch 1
+step  32370/250000 | loss 3.7822 | lr 8.00e-04 emb 4.00e-04 | 3521ms/step | 9,306 tok/s | epoch 1
+step  32380/250000 | loss 3.8132 | lr 8.00e-04 emb 4.00e-04 | 3521ms/step | 9,307 tok/s | epoch 1
+step  32390/250000 | loss 3.6871 | lr 8.00e-04 emb 4.00e-04 | 3520ms/step | 9,309 tok/s | epoch 1
+step  32400/250000 | loss 3.7128 | lr 8.00e-04 emb 4.00e-04 | 3520ms/step | 9,310 tok/s | epoch 1
+step  32410/250000 | loss 3.7725 | lr 8.00e-04 emb 4.00e-04 | 3519ms/step | 9,311 tok/s | epoch 1
+step  32420/250000 | loss 3.7770 | lr 8.00e-04 emb 4.00e-04 | 3520ms/step | 9,310 tok/s | epoch 1
+step  32430/250000 | loss 3.7821 | lr 8.00e-04 emb 4.00e-04 | 3519ms/step | 9,311 tok/s | epoch 1
+step  32440/250000 | loss 3.7456 | lr 8.00e-04 emb 4.00e-04 | 3519ms/step | 9,312 tok/s | epoch 1
+step  32450/250000 | loss 3.7889 | lr 8.00e-04 emb 4.00e-04 | 3519ms/step | 9,313 tok/s | epoch 1
+step  32460/250000 | loss 3.7786 | lr 8.00e-04 emb 4.00e-04 | 3518ms/step | 9,314 tok/s | epoch 1
+step  32470/250000 | loss 3.7332 | lr 8.00e-04 emb 4.00e-04 | 3518ms/step | 9,314 tok/s | epoch 1
+step  32480/250000 | loss 3.7266 | lr 8.00e-04 emb 4.00e-04 | 3518ms/step | 9,314 tok/s | epoch 1
+step  32490/250000 | loss 3.7182 | lr 8.00e-04 emb 4.00e-04 | 3518ms/step | 9,315 tok/s | epoch 1
+step  32500/250000 | loss 3.6681 | lr 8.00e-04 emb 4.00e-04 | 3517ms/step | 9,317 tok/s | epoch 1
+W0322 09:12:19.040000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/utils.py:1706] [17/1] Not enough SMs to use max_autotune_gemm mode
+W0322 09:13:31.690000 92278 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8] torch._dynamo hit config.recompile_limit (8)
+W0322 09:13:31.690000 92278 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8]    function: 'rearrange' (/home/fy/Dev/zara_ml/.venv/lib/python3.12/site-packages/einops/einops.py:561)
+W0322 09:13:31.690000 92278 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8]    last reason: 7/7: tensor 'tensor' rank mismatch. expected 4, actual 3
+W0322 09:13:31.690000 92278 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8] To log all recompilation reasons, use TORCH_LOGS="recompiles".
+W0322 09:13:31.690000 92278 .venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py:1813] [7/8] To diagnose recompilation issues, see https://docs.pytorch.org/docs/main/user_guide/torch_compiler/compile/programming_model.recompilation.html
+W0322 09:13:37.864000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+W0322 09:13:37.866000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+W0322 09:13:37.868000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+W0322 09:13:37.870000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+W0322 09:13:37.872000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+W0322 09:13:37.874000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+W0322 09:13:37.876000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+W0322 09:13:37.881000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+W0322 09:13:37.900000 92278 .venv/lib/python3.12/site-packages/torch/_inductor/cudagraph_utils.py:343] [__cudagraphs] CUDAGraph supports dynamic shapes by recording a new graph for each distinct input size. Recording too many CUDAGraphs may lead to extra overhead. We have observed 9 distinct sizes. Please consider the following options for better performance: a) padding inputs to a few fixed number of shapes; or b) set torch._inductor.config.triton.cudagraph_skip_dynamic_graphs=True. Set torch._inductor.config.triton.cudagraph_dynamic_shape_warn_limit=None to silence this warning.
+  >>> val_loss: 3.8598 | bpt: 5.5685 | true_bpb: 1.7913 *BEST*
+  >>> [The] The The Taqheim band of the United States Army has been a major force in the United States as well as the United States Army and the United States Army.
+The upper bounds of the U.S. Army have recently been made to the U.S. Army, as well as to the U.S. Army. The Army had one of the youngest positions in the Army during the war. The Army of
+  >>> [Scientists have discovered] Scientists have discovered that it has been used to treat lemonsmin.
+The drug has also been linked to increased pain. The study shows that it may be particularly beneficial for the patients with drug-induced brain damage.
+Whether the studies may help us better understand its adverse effects on the brain, or even how it is used, this information is important.
+About the Author
+Beedh et al., "Proper Supply
+step  32510/250000 | loss 3.7445 | lr 8.00e-04 emb 4.00e-04 | 3733ms/step | 8,777 tok/s | epoch 1
+step  32520/250000 | loss 3.7410 | lr 8.00e-04 emb 4.00e-04 | 3729ms/step | 8,788 tok/s | epoch 1
+step  32530/250000 | loss 3.6449 | lr 8.00e-04 emb 4.00e-04 | 3724ms/step | 8,798 tok/s | epoch 1
+step  32540/250000 | loss 3.7357 | lr 8.00e-04 emb 4.00e-04 | 3720ms/step | 8,808 tok/s | epoch 1
+step  32550/250000 | loss 3.6752 | lr 8.00e-04 emb 4.00e-04 | 3716ms/step | 8,818 tok/s | epoch 1
+step  32560/250000 | loss 3.7221 | lr 8.00e-04 emb 4.00e-04 | 3712ms/step | 8,827 tok/s | epoch 1
+step  32570/250000 | loss 3.6747 | lr 8.00e-04 emb 4.00e-04 | 3709ms/step | 8,835 tok/s | epoch 1
+step  32580/250000 | loss 3.7262 | lr 8.00e-04 emb 4.00e-04 | 3705ms/step | 8,843 tok/s | epoch 1
+step  32590/250000 | loss 3.7551 | lr 8.00e-04 emb 4.00e-04 | 3702ms/step | 8,852 tok/s | epoch 1
+step  32600/250000 | loss 3.6877 | lr 8.00e-04 emb 4.00e-04 | 3698ms/step | 8,860 tok/s | epoch 1
+step  32610/250000 | loss 3.6245 | lr 8.00e-04 emb 4.00e-04 | 3695ms/step | 8,868 tok/s | epoch 1
+step  32620/250000 | loss 3.6660 | lr 8.00e-04 emb 4.00e-04 | 3692ms/step | 8,875 tok/s | epoch 1
+step  32630/250000 | loss 3.7096 | lr 8.00e-04 emb 4.00e-04 | 3689ms/step | 8,883 tok/s | epoch 1
+step  32640/250000 | loss 3.6213 | lr 8.00e-04 emb 4.00e-04 | 3686ms/step | 8,890 tok/s | epoch 1
+step  32650/250000 | loss 3.6506 | lr 8.00e-04 emb 4.00e-04 | 3683ms/step | 8,897 tok/s | epoch 1
+step  32660/250000 | loss 3.7223 | lr 8.00e-04 emb 4.00e-04 | 3680ms/step | 8,904 tok/s | epoch 1
+step  32670/250000 | loss 3.6415 | lr 8.00e-04 emb 4.00e-04 | 3678ms/step | 8,910 tok/s | epoch 1
+step  32680/250000 | loss 3.6209 | lr 8.00e-04 emb 4.00e-04 | 3675ms/step | 8,917 tok/s | epoch 1
+step  32690/250000 | loss 3.7024 | lr 8.00e-04 emb 4.00e-04 | 3672ms/step | 8,923 tok/s | epoch 1
+step  32700/250000 | loss 3.6317 | lr 8.00e-04 emb 4.00e-04 | 3670ms/step | 8,930 tok/s | epoch 1
+step  32710/250000 | loss 3.6781 | lr 8.00e-04 emb 4.00e-04 | 3667ms/step | 8,936 tok/s | epoch 1
+step  32720/250000 | loss 3.6185 | lr 8.00e-04 emb 4.00e-04 | 3665ms/step | 8,941 tok/s | epoch 1
+step  32730/250000 | loss 3.5699 | lr 8.00e-04 emb 4.00e-04 | 3662ms/step | 8,947 tok/s | epoch 1
+step  32740/250000 | loss 3.6203 | lr 8.00e-04 emb 4.00e-04 | 3661ms/step | 8,952 tok/s | epoch 1
+step  32750/250000 | loss 3.6717 | lr 8.00e-04 emb 4.00e-04 | 3658ms/step | 8,957 tok/s | epoch 1
+step  32760/250000 | loss 3.6708 | lr 8.00e-04 emb 4.00e-04 | 3656ms/step | 8,962 tok/s | epoch 1
+step  32770/250000 | loss 3.6137 | lr 8.00e-04 emb 4.00e-04 | 3654ms/step | 8,967 tok/s | epoch 1
+step  32780/250000 | loss 3.6641 | lr 8.00e-04 emb 4.00e-04 | 3652ms/step | 8,972 tok/s | epoch 1
+step  32790/250000 | loss 3.6069 | lr 8.00e-04 emb 4.00e-04 | 3650ms/step | 8,976 tok/s | epoch 1
+step  32800/250000 | loss 3.6532 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,981 tok/s | epoch 1
+step  32810/250000 | loss 3.6053 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,985 tok/s | epoch 1
+step  32820/250000 | loss 3.6372 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,990 tok/s | epoch 1
+step  32830/250000 | loss 3.5646 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,995 tok/s | epoch 1
+step  32840/250000 | loss 3.6212 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  32850/250000 | loss 3.6503 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  32860/250000 | loss 3.6397 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  32870/250000 | loss 3.6498 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  32880/250000 | loss 3.6346 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,016 tok/s | epoch 1
+step  32890/250000 | loss 3.6178 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  32900/250000 | loss 3.6099 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,023 tok/s | epoch 1
+step  32910/250000 | loss 3.5542 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,027 tok/s | epoch 1
+step  32920/250000 | loss 3.6588 | lr 8.00e-04 emb 4.00e-04 | 3629ms/step | 9,030 tok/s | epoch 1
+step  32930/250000 | loss 3.5657 | lr 8.00e-04 emb 4.00e-04 | 3628ms/step | 9,033 tok/s | epoch 1
+step  32940/250000 | loss 3.5350 | lr 8.00e-04 emb 4.00e-04 | 3626ms/step | 9,036 tok/s | epoch 1
+step  32950/250000 | loss 3.5168 | lr 8.00e-04 emb 4.00e-04 | 3625ms/step | 9,039 tok/s | epoch 1
+step  32960/250000 | loss 3.5547 | lr 8.00e-04 emb 4.00e-04 | 3624ms/step | 9,042 tok/s | epoch 1
+step  32970/250000 | loss 3.5989 | lr 8.00e-04 emb 4.00e-04 | 3622ms/step | 9,046 tok/s | epoch 1
+step  32980/250000 | loss 3.5758 | lr 8.00e-04 emb 4.00e-04 | 3621ms/step | 9,049 tok/s | epoch 1
+step  32990/250000 | loss 3.5693 | lr 8.00e-04 emb 4.00e-04 | 3620ms/step | 9,052 tok/s | epoch 1
+step  33000/250000 | loss 3.5526 | lr 8.00e-04 emb 4.00e-04 | 3619ms/step | 9,055 tok/s | epoch 1
+  >>> val_loss: 3.7694 | bpt: 5.4381 | true_bpb: 1.7494 *BEST*
+  >>> [The] The treatment of dysphagia is a trauma of the neurological and behavioral disorders that provide an adaptation to a range of systemic and environmental experiences in a variety of environmental settings.
+This is to be expected, however, as most of the research is on non-Epox and its effects on human cognition, particularly in children and adults with dysphagia (e.g., dysphagia in children
+  >>> [Scientists have discovered] Scientists have discovered that the innermost polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar polar
+step  33010/250000 | loss 3.5788 | lr 8.00e-04 emb 4.00e-04 | 3691ms/step | 8,878 tok/s | epoch 1
+step  33020/250000 | loss 3.6459 | lr 8.00e-04 emb 4.00e-04 | 3689ms/step | 8,883 tok/s | epoch 1
+step  33030/250000 | loss 3.5790 | lr 8.00e-04 emb 4.00e-04 | 3687ms/step | 8,887 tok/s | epoch 1
+step  33040/250000 | loss 3.6054 | lr 8.00e-04 emb 4.00e-04 | 3685ms/step | 8,892 tok/s | epoch 1
+step  33050/250000 | loss 3.5389 | lr 8.00e-04 emb 4.00e-04 | 3683ms/step | 8,896 tok/s | epoch 1
+step  33060/250000 | loss 3.6018 | lr 8.00e-04 emb 4.00e-04 | 3682ms/step | 8,900 tok/s | epoch 1
+step  33070/250000 | loss 3.5074 | lr 8.00e-04 emb 4.00e-04 | 3680ms/step | 8,904 tok/s | epoch 1
+step  33080/250000 | loss 3.5437 | lr 8.00e-04 emb 4.00e-04 | 3679ms/step | 8,908 tok/s | epoch 1
+step  33090/250000 | loss 3.5160 | lr 8.00e-04 emb 4.00e-04 | 3677ms/step | 8,912 tok/s | epoch 1
+step  33100/250000 | loss 3.5792 | lr 8.00e-04 emb 4.00e-04 | 3675ms/step | 8,916 tok/s | epoch 1
+step  33110/250000 | loss 3.6596 | lr 8.00e-04 emb 4.00e-04 | 3674ms/step | 8,920 tok/s | epoch 1
+step  33120/250000 | loss 3.6037 | lr 8.00e-04 emb 4.00e-04 | 3672ms/step | 8,923 tok/s | epoch 1
+step  33130/250000 | loss 3.5874 | lr 8.00e-04 emb 4.00e-04 | 3671ms/step | 8,927 tok/s | epoch 1
+step  33140/250000 | loss 3.5561 | lr 8.00e-04 emb 4.00e-04 | 3669ms/step | 8,931 tok/s | epoch 1
+step  33150/250000 | loss 3.4746 | lr 8.00e-04 emb 4.00e-04 | 3668ms/step | 8,934 tok/s | epoch 1
+step  33160/250000 | loss 3.5866 | lr 8.00e-04 emb 4.00e-04 | 3666ms/step | 8,938 tok/s | epoch 1
+step  33170/250000 | loss 3.5620 | lr 8.00e-04 emb 4.00e-04 | 3665ms/step | 8,941 tok/s | epoch 1
+step  33180/250000 | loss 3.4907 | lr 8.00e-04 emb 4.00e-04 | 3663ms/step | 8,945 tok/s | epoch 1
+step  33190/250000 | loss 3.5978 | lr 8.00e-04 emb 4.00e-04 | 3662ms/step | 8,948 tok/s | epoch 1
+step  33200/250000 | loss 3.5582 | lr 8.00e-04 emb 4.00e-04 | 3661ms/step | 8,951 tok/s | epoch 1
+step  33210/250000 | loss 3.5782 | lr 8.00e-04 emb 4.00e-04 | 3659ms/step | 8,955 tok/s | epoch 1
+step  33220/250000 | loss 3.4135 | lr 8.00e-04 emb 4.00e-04 | 3658ms/step | 8,958 tok/s | epoch 1
+step  33230/250000 | loss 3.4599 | lr 8.00e-04 emb 4.00e-04 | 3657ms/step | 8,961 tok/s | epoch 1
+step  33240/250000 | loss 3.5057 | lr 8.00e-04 emb 4.00e-04 | 3655ms/step | 8,964 tok/s | epoch 1
+step  33250/250000 | loss 3.6033 | lr 8.00e-04 emb 4.00e-04 | 3654ms/step | 8,967 tok/s | epoch 1
+step  33260/250000 | loss 3.5199 | lr 8.00e-04 emb 4.00e-04 | 3653ms/step | 8,971 tok/s | epoch 1
+step  33270/250000 | loss 3.5545 | lr 8.00e-04 emb 4.00e-04 | 3652ms/step | 8,974 tok/s | epoch 1
+step  33280/250000 | loss 3.5585 | lr 8.00e-04 emb 4.00e-04 | 3650ms/step | 8,977 tok/s | epoch 1
+step  33290/250000 | loss 3.4980 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,980 tok/s | epoch 1
+step  33300/250000 | loss 3.4611 | lr 8.00e-04 emb 4.00e-04 | 3648ms/step | 8,983 tok/s | epoch 1
+step  33310/250000 | loss 3.5400 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,986 tok/s | epoch 1
+step  33320/250000 | loss 3.5382 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,989 tok/s | epoch 1
+step  33330/250000 | loss 3.5147 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,991 tok/s | epoch 1
+step  33340/250000 | loss 3.5911 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,994 tok/s | epoch 1
+step  33350/250000 | loss 3.5295 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  33360/250000 | loss 3.5845 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  33370/250000 | loss 3.5153 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  33380/250000 | loss 3.5083 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  33390/250000 | loss 3.5056 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  33400/250000 | loss 3.4801 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  33410/250000 | loss 3.4838 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  33420/250000 | loss 3.5204 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  33430/250000 | loss 3.4168 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  33440/250000 | loss 3.4907 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  33450/250000 | loss 3.5345 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  33460/250000 | loss 3.5483 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  33470/250000 | loss 3.5998 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,026 tok/s | epoch 1
+step  33480/250000 | loss 3.4438 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,028 tok/s | epoch 1
+step  33490/250000 | loss 3.4054 | lr 8.00e-04 emb 4.00e-04 | 3629ms/step | 9,030 tok/s | epoch 1
+step  33500/250000 | loss 3.4959 | lr 8.00e-04 emb 4.00e-04 | 3628ms/step | 9,032 tok/s | epoch 1
+  >>> val_loss: 3.7166 | bpt: 5.3619 | true_bpb: 1.7249 *BEST*
+  >>> [The] The the most important thing to pay attention to is to take care of your health. Catch your bite, do not try to get more.
+In addition to brushing your teeth regularly, one few things you need to do to keep your teeth healthy as a whole are:
+- Using a mouthguard or a toothbrush.
+- A mouthguard protects your teeth from gum infections.
+- Make sure your mouth is hydrated
+  >>> [Scientists have discovered] Scientists have discovered the anti-inflammatory properties of 5M, a toxin from the 4M S4M gene, which the scientists have used to acquire an immune response from the 5 M gene. The researchers believe this would be a stronger way of killing off the immune system.
+The researchers also have discovered that there are many information circuits involved in the body's immune system. These circuits are the cells that are
+step  33510/250000 | loss 3.5402 | lr 8.00e-04 emb 4.00e-04 | 3670ms/step | 8,928 tok/s | epoch 1
+step  33520/250000 | loss 3.5470 | lr 8.00e-04 emb 4.00e-04 | 3669ms/step | 8,931 tok/s | epoch 1
+step  33530/250000 | loss 3.4299 | lr 8.00e-04 emb 4.00e-04 | 3668ms/step | 8,933 tok/s | epoch 1
+step  33540/250000 | loss 3.4398 | lr 8.00e-04 emb 4.00e-04 | 3667ms/step | 8,936 tok/s | epoch 1
+step  33550/250000 | loss 3.4781 | lr 8.00e-04 emb 4.00e-04 | 3666ms/step | 8,938 tok/s | epoch 1
+step  33560/250000 | loss 3.4832 | lr 8.00e-04 emb 4.00e-04 | 3665ms/step | 8,941 tok/s | epoch 1
+step  33570/250000 | loss 3.4530 | lr 8.00e-04 emb 4.00e-04 | 3664ms/step | 8,943 tok/s | epoch 1
+step  33580/250000 | loss 3.5087 | lr 8.00e-04 emb 4.00e-04 | 3663ms/step | 8,946 tok/s | epoch 1
+step  33590/250000 | loss 3.5637 | lr 8.00e-04 emb 4.00e-04 | 3662ms/step | 8,948 tok/s | epoch 1
+step  33600/250000 | loss 3.4922 | lr 8.00e-04 emb 4.00e-04 | 3661ms/step | 8,951 tok/s | epoch 1
+step  33610/250000 | loss 3.4745 | lr 8.00e-04 emb 4.00e-04 | 3660ms/step | 8,954 tok/s | epoch 1
+step  33620/250000 | loss 3.3905 | lr 8.00e-04 emb 4.00e-04 | 3659ms/step | 8,956 tok/s | epoch 1
+step  33630/250000 | loss 3.3546 | lr 8.00e-04 emb 4.00e-04 | 3658ms/step | 8,958 tok/s | epoch 1
+step  33640/250000 | loss 3.4134 | lr 8.00e-04 emb 4.00e-04 | 3657ms/step | 8,960 tok/s | epoch 1
+step  33650/250000 | loss 3.5152 | lr 8.00e-04 emb 4.00e-04 | 3656ms/step | 8,963 tok/s | epoch 1
+step  33660/250000 | loss 3.5338 | lr 8.00e-04 emb 4.00e-04 | 3655ms/step | 8,965 tok/s | epoch 1
+step  33670/250000 | loss 3.4875 | lr 8.00e-04 emb 4.00e-04 | 3654ms/step | 8,967 tok/s | epoch 1
+step  33680/250000 | loss 3.5319 | lr 8.00e-04 emb 4.00e-04 | 3653ms/step | 8,970 tok/s | epoch 1
+step  33690/250000 | loss 3.4405 | lr 8.00e-04 emb 4.00e-04 | 3652ms/step | 8,972 tok/s | epoch 1
+step  33700/250000 | loss 3.4231 | lr 8.00e-04 emb 4.00e-04 | 3651ms/step | 8,974 tok/s | epoch 1
+step  33710/250000 | loss 3.4825 | lr 8.00e-04 emb 4.00e-04 | 3651ms/step | 8,976 tok/s | epoch 1
+step  33720/250000 | loss 3.4738 | lr 8.00e-04 emb 4.00e-04 | 3650ms/step | 8,978 tok/s | epoch 1
+step  33730/250000 | loss 3.4558 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,980 tok/s | epoch 1
+step  33740/250000 | loss 3.5521 | lr 8.00e-04 emb 4.00e-04 | 3648ms/step | 8,982 tok/s | epoch 1
+step  33750/250000 | loss 3.4787 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,984 tok/s | epoch 1
+step  33760/250000 | loss 3.4549 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,986 tok/s | epoch 1
+step  33770/250000 | loss 3.5061 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,988 tok/s | epoch 1
+step  33780/250000 | loss 3.4498 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,990 tok/s | epoch 1
+step  33790/250000 | loss 3.4908 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,992 tok/s | epoch 1
+step  33800/250000 | loss 3.5116 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,994 tok/s | epoch 1
+step  33810/250000 | loss 3.4241 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,996 tok/s | epoch 1
+step  33820/250000 | loss 3.4963 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  33830/250000 | loss 3.4805 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  33840/250000 | loss 3.5000 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  33850/250000 | loss 3.3997 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,004 tok/s | epoch 1
+step  33860/250000 | loss 3.3850 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,006 tok/s | epoch 1
+step  33870/250000 | loss 3.4977 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  33880/250000 | loss 3.4404 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  33890/250000 | loss 3.5118 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,011 tok/s | epoch 1
+step  33900/250000 | loss 3.4463 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,013 tok/s | epoch 1
+step  33910/250000 | loss 3.4067 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  33920/250000 | loss 3.4170 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  33930/250000 | loss 3.4234 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,018 tok/s | epoch 1
+step  33940/250000 | loss 3.5264 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  33950/250000 | loss 3.4246 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  33960/250000 | loss 3.4494 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,023 tok/s | epoch 1
+step  33970/250000 | loss 3.4107 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  33980/250000 | loss 3.3755 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,027 tok/s | epoch 1
+step  33990/250000 | loss 3.4925 | lr 8.00e-04 emb 4.00e-04 | 3629ms/step | 9,028 tok/s | epoch 1
+step  34000/250000 | loss 3.4538 | lr 8.00e-04 emb 4.00e-04 | 3629ms/step | 9,030 tok/s | epoch 1
+  >>> val_loss: 3.6805 | bpt: 5.3099 | true_bpb: 1.7081 *BEST*
+  >>> [The] The arguably critical fact was that for today’s generations of confinement, the president was constantly moving away from the political and economic experiment that was taking place to take the people’s lives off right. A bitter situation, however, was the situation with the victory.
+The president spent the last day of his life in mob meetings, and the administration of a cabinet minister and a cabinet minister had always been in line.
+  >>> [Scientists have discovered] Scientists have discovered that living in a linoleic environment, living in a hydrogen sulfide water, may have found a way to go a long time without oxygen. Often this kind of solvent will not just belong to the decomposition process, but also from the decomposition process itself.
+After the reaction, the hydrogen sulfide can react with tests to determine what is causing the reaction. It is also possible that the reaction is
+step  34010/250000 | loss 3.5302 | lr 8.00e-04 emb 4.00e-04 | 3660ms/step | 8,952 tok/s | epoch 1
+step  34020/250000 | loss 3.3940 | lr 8.00e-04 emb 4.00e-04 | 3659ms/step | 8,954 tok/s | epoch 1
+step  34030/250000 | loss 3.4128 | lr 8.00e-04 emb 4.00e-04 | 3659ms/step | 8,956 tok/s | epoch 1
+step  34040/250000 | loss 3.3824 | lr 8.00e-04 emb 4.00e-04 | 3658ms/step | 8,958 tok/s | epoch 1
+step  34050/250000 | loss 3.5289 | lr 8.00e-04 emb 4.00e-04 | 3657ms/step | 8,960 tok/s | epoch 1
+step  34060/250000 | loss 3.3971 | lr 8.00e-04 emb 4.00e-04 | 3656ms/step | 8,962 tok/s | epoch 1
+step  34070/250000 | loss 3.3912 | lr 8.00e-04 emb 4.00e-04 | 3656ms/step | 8,964 tok/s | epoch 1
+step  34080/250000 | loss 3.3633 | lr 8.00e-04 emb 4.00e-04 | 3655ms/step | 8,965 tok/s | epoch 1
+step  34090/250000 | loss 3.4464 | lr 8.00e-04 emb 4.00e-04 | 3654ms/step | 8,967 tok/s | epoch 1
+step  34100/250000 | loss 3.3995 | lr 8.00e-04 emb 4.00e-04 | 3653ms/step | 8,969 tok/s | epoch 1
+step  34110/250000 | loss 3.4305 | lr 8.00e-04 emb 4.00e-04 | 3653ms/step | 8,971 tok/s | epoch 1
+step  34120/250000 | loss 3.3460 | lr 8.00e-04 emb 4.00e-04 | 3652ms/step | 8,973 tok/s | epoch 1
+step  34130/250000 | loss 3.5283 | lr 8.00e-04 emb 4.00e-04 | 3651ms/step | 8,974 tok/s | epoch 1
+step  34140/250000 | loss 3.3708 | lr 8.00e-04 emb 4.00e-04 | 3651ms/step | 8,976 tok/s | epoch 1
+step  34150/250000 | loss 3.5141 | lr 8.00e-04 emb 4.00e-04 | 3650ms/step | 8,978 tok/s | epoch 1
+step  34160/250000 | loss 3.4781 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,979 tok/s | epoch 1
+step  34170/250000 | loss 3.4069 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,981 tok/s | epoch 1
+step  34180/250000 | loss 3.4207 | lr 8.00e-04 emb 4.00e-04 | 3648ms/step | 8,983 tok/s | epoch 1
+step  34190/250000 | loss 3.4781 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,985 tok/s | epoch 1
+step  34200/250000 | loss 3.3426 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,986 tok/s | epoch 1
+step  34210/250000 | loss 3.3984 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,987 tok/s | epoch 1
+step  34220/250000 | loss 3.3872 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,989 tok/s | epoch 1
+step  34230/250000 | loss 3.3977 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,991 tok/s | epoch 1
+step  34240/250000 | loss 3.4530 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,992 tok/s | epoch 1
+step  34250/250000 | loss 3.3621 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,994 tok/s | epoch 1
+step  34260/250000 | loss 3.4752 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,995 tok/s | epoch 1
+step  34270/250000 | loss 3.3996 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  34280/250000 | loss 3.4670 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  34290/250000 | loss 3.4335 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  34300/250000 | loss 3.3448 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  34310/250000 | loss 3.4642 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  34320/250000 | loss 3.4835 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  34330/250000 | loss 3.3329 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,006 tok/s | epoch 1
+step  34340/250000 | loss 3.4757 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  34350/250000 | loss 3.4868 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  34360/250000 | loss 3.4232 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,011 tok/s | epoch 1
+step  34370/250000 | loss 3.4100 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  34380/250000 | loss 3.4088 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,013 tok/s | epoch 1
+step  34390/250000 | loss 3.4764 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  34400/250000 | loss 3.4006 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,016 tok/s | epoch 1
+step  34410/250000 | loss 3.4085 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,018 tok/s | epoch 1
+step  34420/250000 | loss 3.3012 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  34430/250000 | loss 3.3864 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  34440/250000 | loss 3.3281 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  34450/250000 | loss 3.3770 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,023 tok/s | epoch 1
+step  34460/250000 | loss 3.3945 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  34470/250000 | loss 3.4091 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,026 tok/s | epoch 1
+step  34480/250000 | loss 3.4767 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,027 tok/s | epoch 1
+step  34490/250000 | loss 3.4682 | lr 8.00e-04 emb 4.00e-04 | 3629ms/step | 9,028 tok/s | epoch 1
+step  34500/250000 | loss 3.3547 | lr 8.00e-04 emb 4.00e-04 | 3629ms/step | 9,030 tok/s | epoch 1
+  >>> val_loss: 3.6542 | bpt: 5.2719 | true_bpb: 1.6959 *BEST*
+  >>> [The] The sun is quite small.
+According to the latest satellite observations that have been made from the University of Arizona, the Sun has an average temperature of 306 degrees Celsius. According to the analysis, the Sun has a temperature of 125 degrees Celsius. The temperature is at about 25 degrees Celsius. Sun and moon can be extremely hot. Sun and moon can be hot for billions
+  >>> [Scientists have discovered] Scientists have discovered that a protein called the 4M-beta-GGE-3361 molecule can, on average, destroy carbon monoxide. This molecule can also be used to help the weather. The scientists have determined that in the process of carbon monoxide oxidation, carbon dioxide levels rise with the amygdala, a way of sustaining the weather. This is the second time that carbon monoxide has been
+step  34510/250000 | loss 3.3925 | lr 8.00e-04 emb 4.00e-04 | 3654ms/step | 8,967 tok/s | epoch 1
+step  34520/250000 | loss 3.4284 | lr 8.00e-04 emb 4.00e-04 | 3654ms/step | 8,969 tok/s | epoch 1
+step  34530/250000 | loss 3.4632 | lr 8.00e-04 emb 4.00e-04 | 3653ms/step | 8,970 tok/s | epoch 1
+step  34540/250000 | loss 3.4059 | lr 8.00e-04 emb 4.00e-04 | 3653ms/step | 8,971 tok/s | epoch 1
+step  34550/250000 | loss 3.4668 | lr 8.00e-04 emb 4.00e-04 | 3652ms/step | 8,973 tok/s | epoch 1
+step  34560/250000 | loss 3.4805 | lr 8.00e-04 emb 4.00e-04 | 3651ms/step | 8,974 tok/s | epoch 1
+step  34570/250000 | loss 3.4754 | lr 8.00e-04 emb 4.00e-04 | 3651ms/step | 8,975 tok/s | epoch 1
+step  34580/250000 | loss 3.3982 | lr 8.00e-04 emb 4.00e-04 | 3650ms/step | 8,977 tok/s | epoch 1
+step  34590/250000 | loss 3.4406 | lr 8.00e-04 emb 4.00e-04 | 3650ms/step | 8,978 tok/s | epoch 1
+step  34600/250000 | loss 3.4007 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,980 tok/s | epoch 1
+step  34610/250000 | loss 3.4329 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,981 tok/s | epoch 1
+step  34620/250000 | loss 3.3486 | lr 8.00e-04 emb 4.00e-04 | 3648ms/step | 8,982 tok/s | epoch 1
+step  34630/250000 | loss 3.4499 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,984 tok/s | epoch 1
+step  34640/250000 | loss 3.3260 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,985 tok/s | epoch 1
+step  34650/250000 | loss 3.3909 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,986 tok/s | epoch 1
+step  34660/250000 | loss 3.4332 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,988 tok/s | epoch 1
+step  34670/250000 | loss 3.4123 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,989 tok/s | epoch 1
+step  34680/250000 | loss 3.3571 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,990 tok/s | epoch 1
+step  34690/250000 | loss 3.4455 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,992 tok/s | epoch 1
+step  34700/250000 | loss 3.4378 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,993 tok/s | epoch 1
+step  34710/250000 | loss 3.4784 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,994 tok/s | epoch 1
+step  34720/250000 | loss 3.4173 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,996 tok/s | epoch 1
+step  34730/250000 | loss 3.4001 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  34740/250000 | loss 3.3691 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  34750/250000 | loss 3.3846 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  34760/250000 | loss 3.4645 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,001 tok/s | epoch 1
+step  34770/250000 | loss 3.3233 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  34780/250000 | loss 3.3558 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  34790/250000 | loss 3.2782 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  34800/250000 | loss 3.4874 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,006 tok/s | epoch 1
+step  34810/250000 | loss 3.3911 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  34820/250000 | loss 3.3896 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  34830/250000 | loss 3.3643 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  34840/250000 | loss 3.4580 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,011 tok/s | epoch 1
+step  34850/250000 | loss 3.4127 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  34860/250000 | loss 3.3974 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,013 tok/s | epoch 1
+step  34870/250000 | loss 3.3480 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  34880/250000 | loss 3.4204 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  34890/250000 | loss 3.4501 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  34900/250000 | loss 3.3458 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,018 tok/s | epoch 1
+step  34910/250000 | loss 3.3457 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  34920/250000 | loss 3.2949 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  34930/250000 | loss 3.4756 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,021 tok/s | epoch 1
+step  34940/250000 | loss 3.3516 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  34950/250000 | loss 3.3684 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,023 tok/s | epoch 1
+step  34960/250000 | loss 3.4447 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  34970/250000 | loss 3.3161 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  34980/250000 | loss 3.3413 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,027 tok/s | epoch 1
+step  34990/250000 | loss 3.2853 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,028 tok/s | epoch 1
+step  35000/250000 | loss 3.4296 | lr 8.00e-04 emb 4.00e-04 | 3629ms/step | 9,029 tok/s | epoch 1
+  >>> val_loss: 3.6322 | bpt: 5.2402 | true_bpb: 1.6857 *BEST*
+  >>> [The] The main event of the event is the science of the matter, the “The Habitable Phases.” The first encounter is based on the basic notions and arguments of J. H. Lunfield and Johannes Kepler. The second encounter is based on the “Nature” of the universe. The third encounter is based on the “Nature” and is concerned with the “what is the nature of the universe”
+  >>> [Scientists have discovered] Scientists have discovered the mechanism that the grass is able to absorb water. They have now discovered the key to determining the water content of the soil.
+This comes as a great expense and a lot of people are hoping to find a way to harness the power of the pond’s ability to retain water.
+The research team has been studying the water content of the soil for the past 30 years. They have discovered that
+step  35010/250000 | loss 3.2993 | lr 8.00e-04 emb 4.00e-04 | 3650ms/step | 8,976 tok/s | epoch 1
+step  35020/250000 | loss 3.3612 | lr 8.00e-04 emb 4.00e-04 | 3650ms/step | 8,978 tok/s | epoch 1
+step  35030/250000 | loss 3.3875 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,979 tok/s | epoch 1
+step  35040/250000 | loss 3.3482 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,980 tok/s | epoch 1
+step  35050/250000 | loss 3.3825 | lr 8.00e-04 emb 4.00e-04 | 3649ms/step | 8,981 tok/s | epoch 1
+step  35060/250000 | loss 3.3984 | lr 8.00e-04 emb 4.00e-04 | 3648ms/step | 8,982 tok/s | epoch 1
+step  35070/250000 | loss 3.3212 | lr 8.00e-04 emb 4.00e-04 | 3648ms/step | 8,984 tok/s | epoch 1
+step  35080/250000 | loss 3.3745 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,985 tok/s | epoch 1
+step  35090/250000 | loss 3.3537 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,986 tok/s | epoch 1
+step  35100/250000 | loss 3.3956 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,987 tok/s | epoch 1
+step  35110/250000 | loss 3.4315 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,988 tok/s | epoch 1
+step  35120/250000 | loss 3.2905 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,989 tok/s | epoch 1
+step  35130/250000 | loss 3.3327 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,990 tok/s | epoch 1
+step  35140/250000 | loss 3.3269 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,992 tok/s | epoch 1
+step  35150/250000 | loss 3.4024 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,992 tok/s | epoch 1
+step  35160/250000 | loss 3.3953 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,994 tok/s | epoch 1
+step  35170/250000 | loss 3.4541 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,995 tok/s | epoch 1
+step  35180/250000 | loss 3.3922 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,996 tok/s | epoch 1
+step  35190/250000 | loss 3.3776 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  35200/250000 | loss 3.2523 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  35210/250000 | loss 3.3128 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  35220/250000 | loss 3.3309 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  35230/250000 | loss 3.3924 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,001 tok/s | epoch 1
+step  35240/250000 | loss 3.3719 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  35250/250000 | loss 3.3014 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  35260/250000 | loss 3.3733 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,004 tok/s | epoch 1
+step  35270/250000 | loss 3.4786 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  35280/250000 | loss 3.3329 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,006 tok/s | epoch 1
+step  35290/250000 | loss 3.2862 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  35300/250000 | loss 3.2733 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  35310/250000 | loss 3.3715 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  35320/250000 | loss 3.3791 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  35330/250000 | loss 3.3931 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,011 tok/s | epoch 1
+step  35340/250000 | loss 3.3735 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  35350/250000 | loss 3.3310 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,013 tok/s | epoch 1
+step  35360/250000 | loss 3.3479 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  35370/250000 | loss 3.4302 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  35380/250000 | loss 3.3918 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,016 tok/s | epoch 1
+step  35390/250000 | loss 3.3767 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  35400/250000 | loss 3.4158 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,018 tok/s | epoch 1
+step  35410/250000 | loss 3.3162 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  35420/250000 | loss 3.3480 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  35430/250000 | loss 3.4235 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,021 tok/s | epoch 1
+step  35440/250000 | loss 3.3241 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  35450/250000 | loss 3.3764 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,023 tok/s | epoch 1
+step  35460/250000 | loss 3.2434 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  35470/250000 | loss 3.3719 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  35480/250000 | loss 3.3390 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,026 tok/s | epoch 1
+step  35490/250000 | loss 3.2426 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,027 tok/s | epoch 1
+step  35500/250000 | loss 3.2676 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,028 tok/s | epoch 1
+  >>> val_loss: 3.6156 | bpt: 5.2162 | true_bpb: 1.6780 *BEST*
+  >>> [The] The Canadian government has a duty to enforce the
+government's First Amendment, one of the fourth
+States' most advanced laws. It requires the government to develop
+social welfare policy and to provide for the
+"spamation of the human right to life and property; to
+organize the courts and the courts; to ensure the
+constitutional and legal protection of the public
+country; to provide for
+  >>> [Scientists have discovered] Scientists have discovered that high salinity soils can enhance the growth of fish, and that the more elevated the salinity, the greater their salinity. By understanding the molecular mechanisms underlying salinity, scientists can better understand the effects of salinity on fish growth and development.
+Bringing attention to the science of salinity is critical to understanding the significance of salinity in fish, and many researchers have demonstrated that salinity
+step  35510/250000 | loss 3.2974 | lr 8.00e-04 emb 4.00e-04 | 3648ms/step | 8,983 tok/s | epoch 1
+step  35520/250000 | loss 3.3441 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,984 tok/s | epoch 1
+step  35530/250000 | loss 3.2989 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,985 tok/s | epoch 1
+step  35540/250000 | loss 3.3538 | lr 8.00e-04 emb 4.00e-04 | 3647ms/step | 8,986 tok/s | epoch 1
+step  35550/250000 | loss 3.3928 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,987 tok/s | epoch 1
+step  35560/250000 | loss 3.2912 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,988 tok/s | epoch 1
+step  35570/250000 | loss 3.3112 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,989 tok/s | epoch 1
+step  35580/250000 | loss 3.3826 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,990 tok/s | epoch 1
+step  35590/250000 | loss 3.3526 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,991 tok/s | epoch 1
+step  35600/250000 | loss 3.3309 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,992 tok/s | epoch 1
+step  35610/250000 | loss 3.2155 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,993 tok/s | epoch 1
+step  35620/250000 | loss 3.3969 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,994 tok/s | epoch 1
+step  35630/250000 | loss 3.3690 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,995 tok/s | epoch 1
+step  35640/250000 | loss 3.4006 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,996 tok/s | epoch 1
+step  35650/250000 | loss 3.4008 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  35660/250000 | loss 3.3690 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  35670/250000 | loss 3.4300 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  35680/250000 | loss 3.4131 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  35690/250000 | loss 3.3690 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  35700/250000 | loss 3.2983 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,001 tok/s | epoch 1
+step  35710/250000 | loss 3.3642 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  35720/250000 | loss 3.3700 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  35730/250000 | loss 3.3306 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,004 tok/s | epoch 1
+step  35740/250000 | loss 3.3366 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  35750/250000 | loss 3.2410 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,006 tok/s | epoch 1
+step  35760/250000 | loss 3.3227 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  35770/250000 | loss 3.4097 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  35780/250000 | loss 3.3494 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  35790/250000 | loss 3.3649 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  35800/250000 | loss 3.3454 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,011 tok/s | epoch 1
+step  35810/250000 | loss 3.3287 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,011 tok/s | epoch 1
+step  35820/250000 | loss 3.2660 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  35830/250000 | loss 3.3802 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,013 tok/s | epoch 1
+step  35840/250000 | loss 3.4147 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  35850/250000 | loss 3.2963 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  35860/250000 | loss 3.2470 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,016 tok/s | epoch 1
+step  35870/250000 | loss 3.3746 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,016 tok/s | epoch 1
+step  35880/250000 | loss 3.4335 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  35890/250000 | loss 3.4368 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,018 tok/s | epoch 1
+step  35900/250000 | loss 3.3517 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  35910/250000 | loss 3.4063 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  35920/250000 | loss 3.2891 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,021 tok/s | epoch 1
+step  35930/250000 | loss 3.3231 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,021 tok/s | epoch 1
+step  35940/250000 | loss 3.4223 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  35950/250000 | loss 3.2528 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,023 tok/s | epoch 1
+step  35960/250000 | loss 3.3763 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  35970/250000 | loss 3.2272 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  35980/250000 | loss 3.2642 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,026 tok/s | epoch 1
+step  35990/250000 | loss 3.2866 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,026 tok/s | epoch 1
+step  36000/250000 | loss 3.3822 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,027 tok/s | epoch 1
+  >>> val_loss: 3.6000 | bpt: 5.1936 | true_bpb: 1.6707 *BEST*
+  >>> [The] The world’s population of four billion has increased from 6.2 million in 1988 to 14.1 million more in 2013. The world’s population has decreased by more than 10 percent from 1988 to 2013—the number of people living in the recession.
+The recession is warming, now rising by 3.
+  >>> [Scientists have discovered] Scientists have discovered that fungi and the presence of sepsis can be associated with the development of new, sensitive fungal species. Unfortunately, these organisms are vulnerable to some viruses and bacteria because they are important hosts of viruses, viruses, and free-floating organisms.
+In addition to those viruses, they also have a potential to cause disease, and lack of a strong immune system. Whereas opportunistic infections can lead to various illnesses
+step  36010/250000 | loss 3.2403 | lr 8.00e-04 emb 4.00e-04 | 3646ms/step | 8,988 tok/s | epoch 1
+step  36020/250000 | loss 3.3996 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,989 tok/s | epoch 1
+step  36030/250000 | loss 3.3682 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,990 tok/s | epoch 1
+step  36040/250000 | loss 3.3360 | lr 8.00e-04 emb 4.00e-04 | 3645ms/step | 8,990 tok/s | epoch 1
+step  36050/250000 | loss 3.3289 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,991 tok/s | epoch 1
+step  36060/250000 | loss 3.2773 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,992 tok/s | epoch 1
+step  36070/250000 | loss 3.2617 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,993 tok/s | epoch 1
+step  36080/250000 | loss 3.3654 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,994 tok/s | epoch 1
+step  36090/250000 | loss 3.4252 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,995 tok/s | epoch 1
+step  36100/250000 | loss 3.3442 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,996 tok/s | epoch 1
+step  36110/250000 | loss 3.3692 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,996 tok/s | epoch 1
+step  36120/250000 | loss 3.4174 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  36130/250000 | loss 3.3400 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  36140/250000 | loss 3.4256 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  36150/250000 | loss 3.3738 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  36160/250000 | loss 3.4071 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,001 tok/s | epoch 1
+step  36170/250000 | loss 3.3752 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  36180/250000 | loss 3.2376 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  36190/250000 | loss 3.3378 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  36200/250000 | loss 3.4134 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,004 tok/s | epoch 1
+step  36210/250000 | loss 3.2742 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  36220/250000 | loss 3.3772 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,006 tok/s | epoch 1
+step  36230/250000 | loss 3.3091 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,006 tok/s | epoch 1
+step  36240/250000 | loss 3.3238 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  36250/250000 | loss 3.3175 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  36260/250000 | loss 3.3817 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  36270/250000 | loss 3.3461 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  36280/250000 | loss 3.3741 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  36290/250000 | loss 3.3102 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,011 tok/s | epoch 1
+step  36300/250000 | loss 3.3125 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  36310/250000 | loss 3.3266 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,013 tok/s | epoch 1
+step  36320/250000 | loss 3.3445 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  36330/250000 | loss 3.3423 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  36340/250000 | loss 3.3499 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  36350/250000 | loss 3.3966 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,016 tok/s | epoch 1
+step  36360/250000 | loss 3.3804 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  36370/250000 | loss 3.3145 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  36380/250000 | loss 3.3642 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,018 tok/s | epoch 1
+step  36390/250000 | loss 3.3796 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  36400/250000 | loss 3.2563 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  36410/250000 | loss 3.3118 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  36420/250000 | loss 3.2665 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,021 tok/s | epoch 1
+step  36430/250000 | loss 3.2769 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  36440/250000 | loss 3.3415 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  36450/250000 | loss 3.3803 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,023 tok/s | epoch 1
+step  36460/250000 | loss 3.2916 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  36470/250000 | loss 3.3481 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  36480/250000 | loss 3.3143 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  36490/250000 | loss 3.3647 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,026 tok/s | epoch 1
+step  36500/250000 | loss 3.3385 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,027 tok/s | epoch 1
+  >>> val_loss: 3.5855 | bpt: 5.1727 | true_bpb: 1.6640 *BEST*
+  >>> [The] The first thing that is the hardest part is the number of units that you have. The truth is, the number of units that are used is the amount of units that you have. You need to find the number of units that are used in your entire organization. Since you have been involved in this process, and you need to know the number of units you have, you need to know how many units
+  >>> [Scientists have discovered] Scientists have discovered a new pattern they known as the “Drinking Water.” Dr. William Smith, professor at the University of Basel, and his team had discovered that the “Drinking Water” had such a special substance called “Severe Water.” “This was a big hamster of the water,” says Brian Dickson, professor of regolithography and oceanography at the University of Leeds, who led the
+step  36510/250000 | loss 3.2834 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,992 tok/s | epoch 1
+step  36520/250000 | loss 3.3730 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,993 tok/s | epoch 1
+step  36530/250000 | loss 3.2918 | lr 8.00e-04 emb 4.00e-04 | 3644ms/step | 8,994 tok/s | epoch 1
+step  36540/250000 | loss 3.3311 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,994 tok/s | epoch 1
+step  36550/250000 | loss 3.2480 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,995 tok/s | epoch 1
+step  36560/250000 | loss 3.3487 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,996 tok/s | epoch 1
+step  36570/250000 | loss 3.2978 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  36580/250000 | loss 3.3358 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  36590/250000 | loss 3.2716 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  36600/250000 | loss 3.3092 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  36610/250000 | loss 3.3825 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  36620/250000 | loss 3.3074 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  36630/250000 | loss 3.3029 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,001 tok/s | epoch 1
+step  36640/250000 | loss 3.4511 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  36650/250000 | loss 3.2862 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  36660/250000 | loss 3.2878 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  36670/250000 | loss 3.3127 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,004 tok/s | epoch 1
+step  36680/250000 | loss 3.3518 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  36690/250000 | loss 3.3121 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  36700/250000 | loss 3.1981 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,006 tok/s | epoch 1
+step  36710/250000 | loss 3.3032 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  36720/250000 | loss 3.3371 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  36730/250000 | loss 3.3398 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  36740/250000 | loss 3.2078 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  36750/250000 | loss 3.1812 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  36760/250000 | loss 3.2894 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,011 tok/s | epoch 1
+step  36770/250000 | loss 3.1648 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,011 tok/s | epoch 1
+step  36780/250000 | loss 3.2101 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  36790/250000 | loss 3.3325 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,013 tok/s | epoch 1
+step  36800/250000 | loss 3.2289 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,013 tok/s | epoch 1
+step  36810/250000 | loss 3.2797 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  36820/250000 | loss 3.2781 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  36830/250000 | loss 3.3315 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,016 tok/s | epoch 1
+step  36840/250000 | loss 3.3463 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,016 tok/s | epoch 1
+step  36850/250000 | loss 3.2967 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  36860/250000 | loss 3.3756 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,018 tok/s | epoch 1
+step  36870/250000 | loss 3.3079 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,018 tok/s | epoch 1
+step  36880/250000 | loss 3.3288 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  36890/250000 | loss 3.2283 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  36900/250000 | loss 3.3556 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  36910/250000 | loss 3.3023 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,021 tok/s | epoch 1
+step  36920/250000 | loss 3.3723 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  36930/250000 | loss 3.3878 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  36940/250000 | loss 3.3718 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,023 tok/s | epoch 1
+step  36950/250000 | loss 3.3012 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  36960/250000 | loss 3.2372 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  36970/250000 | loss 3.2859 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  36980/250000 | loss 3.2464 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,026 tok/s | epoch 1
+step  36990/250000 | loss 3.2659 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,026 tok/s | epoch 1
+step  37000/250000 | loss 3.2169 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,027 tok/s | epoch 1
+  >>> val_loss: 3.5735 | bpt: 5.1555 | true_bpb: 1.6585 *BEST*
+  >>> [The] The aim of this study was to investigate the effect of promoting the movement of the abdomen during endurance training and in a free-standing group of women who had a high speed diet. Specifically, the women were given a moderate calorie, low-fat diet. The subjects were given the diet in which they were given a moderately continuously productive diet. After 15 weeks of training, the young women exhibited significant declines in
+  >>> [Scientists have discovered] Scientists have discovered that the environment in which plants grow creates a variety of organs, such as the digestive tract, thyroid, and digestive organs, that are essential for them to thrive. Other information that the researchers suggest is also present in other plants, such as the garden plants, leaves, and animals found in the atmosphere. But they believe the plant parts will be beneficial to plants, and that it creates a natural and
+step  37010/250000 | loss 3.3655 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,995 tok/s | epoch 1
+step  37020/250000 | loss 3.2700 | lr 8.00e-04 emb 4.00e-04 | 3643ms/step | 8,996 tok/s | epoch 1
+step  37030/250000 | loss 3.1731 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  37040/250000 | loss 3.3127 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,997 tok/s | epoch 1
+step  37050/250000 | loss 3.2735 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  37060/250000 | loss 3.3545 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  37070/250000 | loss 3.2992 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  37080/250000 | loss 3.3132 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  37090/250000 | loss 3.2593 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,001 tok/s | epoch 1
+step  37100/250000 | loss 3.3010 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,001 tok/s | epoch 1
+step  37110/250000 | loss 3.3044 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  37120/250000 | loss 3.4172 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  37130/250000 | loss 3.3649 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  37140/250000 | loss 3.3000 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,004 tok/s | epoch 1
+step  37150/250000 | loss 3.2291 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  37160/250000 | loss 3.2329 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  37170/250000 | loss 3.3746 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,006 tok/s | epoch 1
+step  37180/250000 | loss 3.3499 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  37190/250000 | loss 3.3031 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  37200/250000 | loss 3.2400 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  37210/250000 | loss 3.2502 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,008 tok/s | epoch 1
+step  37220/250000 | loss 3.2324 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  37230/250000 | loss 3.2300 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  37240/250000 | loss 3.3686 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  37250/250000 | loss 3.3447 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,011 tok/s | epoch 1
+step  37260/250000 | loss 3.3074 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  37270/250000 | loss 3.3822 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  37280/250000 | loss 3.2872 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,013 tok/s | epoch 1
+step  37290/250000 | loss 3.3020 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  37300/250000 | loss 3.2137 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  37310/250000 | loss 3.2206 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  37320/250000 | loss 3.2766 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1
+step  37330/250000 | loss 3.3264 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,016 tok/s | epoch 1
+step  37340/250000 | loss 3.2311 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  37350/250000 | loss 3.2715 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,017 tok/s | epoch 1
+step  37360/250000 | loss 3.2684 | lr 8.00e-04 emb 4.00e-04 | 3634ms/step | 9,018 tok/s | epoch 1
+step  37370/250000 | loss 3.3393 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  37380/250000 | loss 3.3798 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,019 tok/s | epoch 1
+step  37390/250000 | loss 3.2448 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  37400/250000 | loss 3.3022 | lr 8.00e-04 emb 4.00e-04 | 3633ms/step | 9,020 tok/s | epoch 1
+step  37410/250000 | loss 3.2729 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,021 tok/s | epoch 1
+step  37420/250000 | loss 3.1535 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  37430/250000 | loss 3.2199 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,022 tok/s | epoch 1
+step  37440/250000 | loss 3.2866 | lr 8.00e-04 emb 4.00e-04 | 3632ms/step | 9,023 tok/s | epoch 1
+step  37450/250000 | loss 3.2747 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,023 tok/s | epoch 1
+step  37460/250000 | loss 3.3482 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,024 tok/s | epoch 1
+step  37470/250000 | loss 3.2736 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  37480/250000 | loss 3.2861 | lr 8.00e-04 emb 4.00e-04 | 3631ms/step | 9,025 tok/s | epoch 1
+step  37490/250000 | loss 3.3569 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,026 tok/s | epoch 1
+step  37500/250000 | loss 3.3345 | lr 8.00e-04 emb 4.00e-04 | 3630ms/step | 9,026 tok/s | epoch 1
+  >>> val_loss: 3.5635 | bpt: 5.1410 | true_bpb: 1.6538 *BEST*
+  >>> [The] The basics of life.
+- The capacity to manage and manage the resources that you have provided.
+- The motivation to do similar things.
+- The ability to participate adequately in a decline and life change.
+- The ability to create a meaningful, meaningful impact.
+- The ability to perform various tasks responsibly and responsibly.
+- The ability to adapt and adapt to any changing circumstances.
+- The ability to maintain a stable
+  >>> [Scientists have discovered] Scientists have discovered that the wind is constantly turning a series of large body movements, such as jumping, into the wind during a dance. The movement of wind is often known as katy or jerk.
+When wind breaks up wind blows, you could hear blowing in the breeze. It often is that wind is the strongest or the worst of the shocks, and the movement is always very strong. This is the reason the
+step  37510/250000 | loss 3.3372 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  37520/250000 | loss 3.2945 | lr 8.00e-04 emb 4.00e-04 | 3642ms/step | 8,998 tok/s | epoch 1
+step  37530/250000 | loss 3.3617 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 8,999 tok/s | epoch 1
+step  37540/250000 | loss 3.2690 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  37550/250000 | loss 3.3603 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,000 tok/s | epoch 1
+step  37560/250000 | loss 3.2478 | lr 8.00e-04 emb 4.00e-04 | 3641ms/step | 9,001 tok/s | epoch 1
+step  37570/250000 | loss 3.2987 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,001 tok/s | epoch 1
+step  37580/250000 | loss 3.2970 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,002 tok/s | epoch 1
+step  37590/250000 | loss 3.2636 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  37600/250000 | loss 3.3056 | lr 8.00e-04 emb 4.00e-04 | 3640ms/step | 9,003 tok/s | epoch 1
+step  37610/250000 | loss 3.2568 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,004 tok/s | epoch 1
+step  37620/250000 | loss 3.2177 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,004 tok/s | epoch 1
+step  37630/250000 | loss 3.3196 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,005 tok/s | epoch 1
+step  37640/250000 | loss 3.2057 | lr 8.00e-04 emb 4.00e-04 | 3639ms/step | 9,006 tok/s | epoch 1
+step  37650/250000 | loss 3.3417 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,006 tok/s | epoch 1
+step  37660/250000 | loss 3.3532 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  37670/250000 | loss 3.2434 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,007 tok/s | epoch 1
+step  37680/250000 | loss 3.2764 | lr 8.00e-04 emb 4.00e-04 | 3638ms/step | 9,008 tok/s | epoch 1
+step  37690/250000 | loss 3.2178 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  37700/250000 | loss 3.2455 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,009 tok/s | epoch 1
+step  37710/250000 | loss 3.3845 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  37720/250000 | loss 3.3055 | lr 8.00e-04 emb 4.00e-04 | 3637ms/step | 9,010 tok/s | epoch 1
+step  37730/250000 | loss 3.1566 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,011 tok/s | epoch 1
+step  37740/250000 | loss 3.2423 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,011 tok/s | epoch 1
+step  37750/250000 | loss 3.3745 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,012 tok/s | epoch 1
+step  37760/250000 | loss 3.2900 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,013 tok/s | epoch 1
+step  37770/250000 | loss 3.3300 | lr 8.00e-04 emb 4.00e-04 | 3636ms/step | 9,013 tok/s | epoch 1
+step  37780/250000 | loss 3.1962 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  37790/250000 | loss 3.1939 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,014 tok/s | epoch 1
+step  37800/250000 | loss 3.3492 | lr 8.00e-04 emb 4.00e-04 | 3635ms/step | 9,015 tok/s | epoch 1