Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74463.77325933569,
-      "required_pages_per_second_upper_estimate": 0.006454331139411316
+      "required_token_positions_per_second": 74894.68819944288,
+      "required_pages_per_second_upper_estimate": 0.006491681754275876
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9307.971657416962,
-      "required_pages_per_second_upper_estimate": 0.003713089060721622
+      "required_token_positions_per_second": 9361.83602493036,
+      "required_pages_per_second_upper_estimate": 0.0037345763622667785
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18615.943314833923,
-      "required_pages_per_second_upper_estimate": 0.004306470856474421
+      "required_token_positions_per_second": 18723.67204986072,
+      "required_pages_per_second_upper_estimate": 0.00433139200336189
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27923.914972250885,
-      "required_pages_per_second_upper_estimate": 0.001700419916350758
+      "required_token_positions_per_second": 28085.508074791083,
+      "required_pages_per_second_upper_estimate": 0.0017102600884819695
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27923.914972250885,
-      "required_pages_per_second_upper_estimate": 0.0021496602729078303
+      "required_token_positions_per_second": 28085.508074791083,
+      "required_pages_per_second_upper_estimate": 0.0021621001572597123
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27923.914972250885,
-      "required_pages_per_second_upper_estimate": 0.34672156862374914
+      "required_token_positions_per_second": 28085.508074791083,
+      "required_pages_per_second_upper_estimate": 0.34872801413997395
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.40037772304694386,
-    "producer_p95_seconds": 0.32834741799888434,
-    "gpu_update_seconds": 0.9901190441052026,
-    "producer_batches_per_second": 2.49764145814564,
-    "gpu_batches_per_second": 1.0099795635218056,
-    "producer_headroom": 2.4729623730566854,
+    "producer_batch_seconds": 0.1634502163671243,
+    "producer_p95_seconds": 0.3260848159989109,
+    "gpu_update_seconds": 0.9844222837761735,
+    "producer_batches_per_second": 6.118070824414863,
+    "gpu_batches_per_second": 1.0158242214551174,
+    "producer_headroom": 6.022765253274856,
     "sustainable": true,
-    "buffer_seconds": 60.39726169041736,
+    "buffer_seconds": 60.04975931034658,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395331.6585669071,
+    "generation_bytes_per_second": 413867.38137492625,
     "upload_bytes_per_second": 4144407.649767115,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.82897107844744,
+    "estimated_full_shards_per_hour": 88.80630570350495,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -185,3 +185,4 @@
 {"step": 103500, "loss": 0.5845393426716328, "lr": 0.0003, "grad_norm": 0.13813541500181312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15145877141, "tokens_per_second": 148842.16556128208, "data_wait_s": 1.5707002603448927e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 103600, "loss": 0.5744451247155666, "lr": 0.0003, "grad_norm": 0.13890033496057527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15160483280, "tokens_per_second": 144457.2000252497, "data_wait_s": 1.4171997463563457e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 103700, "loss": 0.605474017560482, "lr": 0.0003, "grad_norm": 0.1383430302841768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15175012661, "tokens_per_second": 147318.96638237443, "data_wait_s": 1.7316000594291836e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 103800, "loss": 0.643247103318572, "lr": 0.0003, "grad_norm": 0.13918651865510248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15189614740, "tokens_per_second": 147997.2361363034, "data_wait_s": 1.610899926163256e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2552b66f21ad0b9d29b821af97b045a774c5ec9b03e8c611f0e4fd667aeee618
-size 398580
+oid sha256:a8b2530779e7252a165f75189e2206efe40b6d3f0edfeaeb403503fc4d19e392
+size 400698