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": 76285.89494527929,
-      "required_pages_per_second_upper_estimate": 0.005277635219977591
+      "required_token_positions_per_second": 76413.05350459921,
+      "required_pages_per_second_upper_estimate": 0.0052864323441598506
     },
     {
       "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": 9535.736868159911,
-      "required_pages_per_second_upper_estimate": 0.003992413897869054
+      "required_token_positions_per_second": 9551.631688074902,
+      "required_pages_per_second_upper_estimate": 0.003999068727045876
     },
     {
       "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": 19071.473736319822,
-      "required_pages_per_second_upper_estimate": 0.003972085030691165
+      "required_token_positions_per_second": 19103.263376149804,
+      "required_pages_per_second_upper_estimate": 0.003978705974318571
     },
     {
       "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": 28607.210604479733,
-      "required_pages_per_second_upper_estimate": 0.0017149585950571604
+      "required_token_positions_per_second": 28654.895064224704,
+      "required_pages_per_second_upper_estimate": 0.0017178172056089172
     },
     {
       "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": 28607.210604479733,
-      "required_pages_per_second_upper_estimate": 0.0022240847946769556
+      "required_token_positions_per_second": 28654.895064224704,
+      "required_pages_per_second_upper_estimate": 0.002227792051680355
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28607.210604479733,
-      "required_pages_per_second_upper_estimate": 0.36694258160464505
+      "required_token_positions_per_second": 28654.895064224704,
+      "required_pages_per_second_upper_estimate": 0.3675542266546697
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14263968398441307,
-    "producer_p95_seconds": 0.31378181299805874,
-    "gpu_update_seconds": 0.9664696213223415,
-    "producer_batches_per_second": 7.010671729399476,
-    "gpu_batches_per_second": 1.034693670590268,
-    "producer_headroom": 6.775601251527958,
+    "producer_batch_seconds": 0.11514162374223247,
+    "producer_p95_seconds": 0.30869785899994895,
+    "gpu_update_seconds": 0.9648613243228449,
+    "producer_batches_per_second": 8.68495655609912,
+    "gpu_batches_per_second": 1.0364183689317386,
+    "producer_headroom": 8.37977868440417,
     "sustainable": true,
-    "buffer_seconds": 60.887586143307516,
+    "buffer_seconds": 60.786263432339226,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312787.0631909728,
+    "generation_bytes_per_second": 313682.5314962577,
     "upload_bytes_per_second": 5298031.726554063,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.11682245060814,
+    "estimated_full_shards_per_hour": 67.308969103487,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 637200, "loss": 0.56538325548172, "lr": 0.0001, "grad_norm": 0.16798120802493155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93269558678, "tokens_per_second": 151645.69934552547, "data_wait_s": 1.0137999197468162e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 637300, "loss": 0.556537926197052, "lr": 0.0001, "grad_norm": 0.1721248058661298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93284114297, "tokens_per_second": 147904.51063344016, "data_wait_s": 8.86399720911868e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 637400, "loss": 0.5920100212097168, "lr": 0.0001, "grad_norm": 0.1860952955176649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93298632099, "tokens_per_second": 150522.39252012363, "data_wait_s": 9.453000529902056e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 637500, "loss": 0.569724440574646, "lr": 0.0001, "grad_norm": 0.17819412745127783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93313262112, "tokens_per_second": 152721.54016342905, "data_wait_s": 1.1635002010734752e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1537077ceec6afaf43c93bc475c4ed20f77684f0f69d1e13e9df551ca572820
-size 704586
+oid sha256:783c94546a3fce40ffef70a8a64b52b1fe6ade3a17af8eb55654e786b843a110
+size 706704