Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76355.78087234657,
-      "required_pages_per_second_upper_estimate": 0.005282470090569797
+      "required_token_positions_per_second": 76183.51539906266,
+      "required_pages_per_second_upper_estimate": 0.005270552365417049
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9544.472609043321,
-      "required_pages_per_second_upper_estimate": 0.0039960713701539235
+      "required_token_positions_per_second": 9522.939424882832,
+      "required_pages_per_second_upper_estimate": 0.003987055875609945
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19088.945218086643,
-      "required_pages_per_second_upper_estimate": 0.003975723879589235
+      "required_token_positions_per_second": 19045.878849765664,
+      "required_pages_per_second_upper_estimate": 0.003966754290804482
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28633.417827129964,
-      "required_pages_per_second_upper_estimate": 0.0017165296780389297
+      "required_token_positions_per_second": 28568.818274648496,
+      "required_pages_per_second_upper_estimate": 0.0017126570335054712
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28633.417827129964,
-      "required_pages_per_second_upper_estimate": 0.002226122290964621
+      "required_token_positions_per_second": 28568.818274648496,
+      "required_pages_per_second_upper_estimate": 0.002221099959902589
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28633.417827129964,
-      "required_pages_per_second_upper_estimate": 0.3672787397176789
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28568.818274648496,
+      "required_pages_per_second_upper_estimate": 0.366450126020042
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08798215473424875,
-    "producer_p95_seconds": 0.2692010829996434,
-    "gpu_update_seconds": 0.9655850435641571,
-    "producer_batches_per_second": 11.36594123002003,
-    "gpu_batches_per_second": 1.0356415591409855,
-    "producer_headroom": 10.97478285773654,
+    "producer_batch_seconds": 0.26964968273421164,
+    "producer_p95_seconds": 0.27072457800022676,
+    "gpu_update_seconds": 0.9677684156972773,
+    "producer_batches_per_second": 3.708515396199002,
+    "gpu_batches_per_second": 1.0333050591235713,
+    "producer_headroom": 3.5889840695684687,
     "sustainable": true,
-    "buffer_seconds": 60.8318577445419,
+    "buffer_seconds": 60.00164177323119,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397305.3105749293,
-    "upload_bytes_per_second": 3761982.5041243904,
+    "generation_bytes_per_second": 391172.3881166338,
+    "upload_bytes_per_second": 3804357.0129485405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.25247085510168,
+    "estimated_full_shards_per_hour": 83.93648846267949,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.250901432998944,
-      "bytes": 1087271,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.5110034520002955,
       "bytes": 7289248,
@@ -562,6 +556,12 @@
       "bytes": 2590562,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3539272460002394,
+      "bytes": 6288510,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -265,3 +265,4 @@
 {"step": 551500, "loss": 0.5715614557266235, "lr": 0.0003, "grad_norm": 0.1479404510899116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80722270761, "tokens_per_second": 152463.46489181934, "data_wait_s": 8.75200203154236e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 551600, "loss": 0.5878363847732544, "lr": 0.0003, "grad_norm": 0.17073314825640917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80736882669, "tokens_per_second": 152311.14371276874, "data_wait_s": 1.499599966336973e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 551700, "loss": 0.5444007515907288, "lr": 0.0003, "grad_norm": 0.14396969298588103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80751490379, "tokens_per_second": 151359.89025243273, "data_wait_s": 1.1003998224623501e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 551800, "loss": 0.5731511116027832, "lr": 0.0003, "grad_norm": 0.14905357860602683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80766091810, "tokens_per_second": 152191.24941692536, "data_wait_s": 1.7607999325264245e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f44ae21fd779a3393e471bb4209318be0006b8e00df607db13443eb7d862a07
-size 569586
+oid sha256:3f10a3946d422c10783da00258f2ebd22f48942c8cf9bb0f3cbdeee3fdabea92
+size 571704